Lillemark et al BMCMedical Imaging 2014 1421httpwwwbiomedcentralcom1471-23421421
RESEARCH ARTICLE Open Access
Brain regionrsquos relative proximity as marker forAlzheimerrsquos disease based on structural MRILene Lillemark1 Lauge Soslashrensen1 Akshay Pai1 Erik B Dam2 Mads Nielsen12 and Alzheimerrsquos DiseaseNeuroimaging Initiative
Abstract
Background Alzheimerrsquos disease (AD) is a progressive incurable neurodegenerative disease and the most commontype of dementia It cannot be prevented cured or drastically slowed even though AD research has increased in thepast 5-10 years Instead of focusing on the brain volume or on the single brain structures like hippocampus this paperinvestigates the relationship and proximity between regions in the brain and uses this information as a novel way ofclassifying normal control (NC) mild cognitive impaired (MCI) and AD subjects
Methods A longitudinal cohort of 528 subjects (170 NC 240 MCI and 114 AD) from ADNI at baseline and month 12was studied We investigated a marker based on Procrustes aligned center of masses and the percentile surfaceconnectivity between regions These markers were classified using a linear discriminant analysis in a cross validationsetting and compared to whole brain and hippocampus volume
Results We found that both our markers was able to significantly classify the subjects The surface connectivitymarker showed the best results with an area under the curve (AUC) at 0877 (p lt 0001) 0784 (p lt 0001) 0766(p lt 0001) for NC-AD NC-MCI and MCI-AD respectively for the functional regions in the brain The surfaceconnectivity marker was able to classify MCI-converters with an AUC of 0599 (p lt 005) for the 1-year period
Conclusion Our results show that our relative proximity markers include more information than whole brain andhippocampus volume Our results demonstrate that our proximity markers have the potential to assist in earlydiagnosis of AD
Keywords Alzheimerrsquos disease Mild cognitive impairment Bio markers MRI Diagnosis and classification Proximity
BackgroundAlzheimerrsquos Disease (AD) is the sixth-leading cause ofdeath in the US and accounts for 50-56 of the cases ofdiagnosed dementia [1] AD is often diagnosed in peopleover 65 years but the onset for AD can occur much earlierThe population of aged 65+ years in the US is estimated todouble by the year 2030 [1] This means that the numberof new and existing cases of AD will increase drasticallyas well At the moment no cure or treatment is found forAD which obviously makes it a growing problem [1]The causes of AD are not fully clarified Research
indicates that accumulation of twisted protein fragmentsinside the nerve cells neurofibrillary tangles and toxic
Correspondence lenelillemarkgmailcom1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash DenmarkFull list of author information is available at the end of the article
protein fragment amyloid beta oligomers are character-istics of AD [23] The hippocampus which is associatedwith memory is particular vulnerable to damage at theearliest stages of AD [34] Hippocampal brain changessuch as loss of thickness and volume in the medial tem-poral lobe particular in the hippocampus is thought tobegin 7 years or more before AD symptoms such asmemory loss appear [5-8]The cognitive decline can be slowed when administered
during the early stage of disease [9] and therefore earlydetection of brain changes is highly desirable both toincrease the quality of AD patientrsquos life but also for futuredevelopments in drug trialsStructural magnetic resonance imaging (MRI) has
shown great applicability to map how AD spreads in theliving brain Recent literature have studied the accuracyand reproducibility of MRI-derived measurements and
copy 2014 Lillemark et al licensee BioMed Central Ltd This is an Open Access article distributed under the terms of the CreativeCommons Attribution License (httpcreativecommonsorglicensesby20) which permits unrestricted use distribution andreproduction in any medium provided the original work is properly credited The Creative Commons Public Domain Dedicationwaiver (httpcreativecommonsorgpublicdomainzero10) applies to the data made available in this article unless otherwisestated
Lillemark et al BMCMedical Imaging 2014 1421 Page 2 of 12httpwwwbiomedcentralcom1471-23421421
found correlation with clinical measurements and positiveprediction of future decline [10-12] MRI is non invasiveand is largely available in the clinical environment andhas therefore become a viable tool to monitor the progres-sion of AD Longitudinal measurement changes may bemore objective precise and reproducible when they aremeasured from MRI compared to diagnostic image mea-surements from PET scanning or CSF measurement ofamyloid and tau proteinDifferent studies have focused on the hippocampus
region and used atrophy scoring and shape analysis fordetection of AD [13-16] Also whole brain atrophy scor-ing have been extensively used for detection of AD [1718]The same techniques have been used to classify MCI-converters (MCI-c) from MCI-non-converters (MCI-nc)indicating that it is possible to make a prognosis of ADbased on atrophy rates shape analysis region hippocam-pal shapes and machine learning techniques [1019-24]The general focus on studies from MRI have been on
the atrophy rates for hippocampus or the whole brainor the shape of hippocampus but other brain structureslike amygdala putamen thalamus and the ventricles havealso shown relation to AD [25-27] We want to includeall of these structures in order to investigated the rela-tionship and proximity between different regions in brainin hope to characterize how the brain develops and usethis as a marker for AD We believe that the relationshipbetween the positions of different regions or the surfaceconnectivity between the different regions in the brain cancapture how the atrophy spreads We have used a Pro-crustes marker that classified AD subjects based on theposition of the center of mass of each region in a Pro-crustes aligned environment and a surface connectivitymarker that extracted the percentile surface connectivitybetween the individual regions The Procrustes markercan capture how the regions move away or toward eachother indicating how the volume loss is different acrossthe brain The surface connectivity marker can describethe individual volume loss of the regions and how theymove apart due to for example the increase in ventriclesand cerebrospinal fluid (CSF) These new markers couldgive a more detailed view of the AD progression and maybe used in addition to the traditional morphometric mark-ers Our markers were used in three different groupingsof the brain regions a group of all Freesurfer segmentedregions a subset of the functional regions and a subset ofthe small potato shaped regions (for example hippocam-pus and amygdala) to classify using a linear discriminantanalysis NC MCI and AD This was done in compari-son to the whole brain volume and hippocampus volumePotentially this could lead to a fine-to-coarse scale fromwhere one can study the progression of AD from theglobal brain scale down to the local scale of the shapeandor texture of the individual sub-regions
MethodsADNI brain MRI and preprocessingData was obtained from the Alzheimerrsquos Disease Neu-roimaging Initiative (ADNI) database (adniloniuscedu) [28] The ADNIwas launched in 2003 by theNationalInstitute for Aging (NIA) the National Institute of Bio-chemical Imaging and Bioengineering (NIBIB) the Foodand Drug Administration (FDA) private pharmaceuticalcompanies and non-profit organizations as a $ 60 mil-lion 5-year public-private partnership The primary goalof ADNI has been to test whether serial MRI positronemission tomography PET other biological markers andclinical and neurophysiological assessments can be com-bined to measure the progression of MCI and early ADDetermination of sensitive and specific markers of veryearly AD progression is intended to aid researchers andclinicians to develop new treatments and monitor theireffectiveness as well as lessen the time and cost of clin-ical trials The Principal Investigator of this initiative isMichael W Weiner MD VA Medical Center and Uni-versity of California - San Francisco ADNI is the resultof efforts of many co-investigators from a broad range ofacademic institutions and private corporations and sub-jects have been recruited from over 50 sites across the USand Canada The initial goal of ADNI was to recruit 800subjects but ADNI has been followed by ADNI-GO andADNI-2 To date these three protocols have recruited over1500 adults ages 55 to 90 to participate in the researchconsisting of cognitively normal older individuals peoplewith early or late MCI and people with early AD The fol-low up duration of each group is specified in the protocolsfor ADNI-1 ADNI-2 and ADNI-GO Subjects originallyrecruited for ADNI-1 and ADNI-GO had the option tobe followed in ADNI-2 For up-to-date information seewwwadni-infoorgLongitudinal brain T1 weighted MRI and associated
data for the study population including age gender anddiagnosis were downloaded from the ADNI database Alldata in this paper were from ADNI-1 ADNI-1 was afive year study launched in 2004 to develop longitudi-nal outcome measures of Alzheimerrsquos progression usingserial MRI PET biochemical changes in CSF blood andurine and cognitive and neuropsychological assessmentacquired at multiple sites similar to typical clinical tri-als All subjects underwent clinical and cognitive assess-ment at the time of scan acquisition All AD subjectsmet NINCDSADRDA criteria for probable AD [29] Thestudy was conducted according to the Good Clinical Prac-tice guidelines the Declaration of Helsinki and US 21CFR Part-50 Protection of Human Subject and Part 56-Institutional Review Boards This study was approved bythe Institutional Review Boards of all of the participatinginstitutions and informed written consent was obtainedfrom all participants at each site
Lillemark et al BMCMedical Imaging 2014 1421 Page 3 of 12httpwwwbiomedcentralcom1471-23421421
MRI acquisitionHigh-Resolution structural brain MRI were acquired at59 ADNI sites using 15 Tesla T1-weighted MRI scansusing volumetric 3D MPRAGE or equivalent protocolswith varying resolution typically 125 times 125 mm in-plane spatial resolution and 12 mm thick sagital slicesThe MPRAGE sequence was acquired twice for all sub-jects at each visit to improve the chance that at least onescan would be suitable for analysis The image quality wasgraded qualitatively by ADNI investigators of the ADNIMRQ quality control center at the Mayo Clinic for arti-facts and general image quality Each scan was graded onseveral separate criteria blurringghosting flow artifactsintensity a homogeneity signal-to-noise ratio susceptibil-ity artifacts and gray-whitecerebrospinal fluid contrastWe have only used the MRI scan which was graded asthe best scan for each subject No other exclusion crite-ria based on image quality were applied We have used theraw ADNI data
ParticipantsThe criteria for inclusion were those defined in theADNI protocol normal control (NC) subjects had amini mental state examamination score (MMSE) between24 - 30 a clinical dementia rating (CDR) score of zerothey were non-depressed non MCI and non-dementedMCI had MMSE scores between 24-30 a memory com-plaint had objective memory loss measured by educationadjusted scores onWechsler Memory Scale Logical Mem-ory II [30] a CDR of 05 absence of significant lev-els of impairment in other cognitive domains AD sub-jects had MMSE scores between 20-26 CDR of 05or 10 and met NINCDSADRDA criteria for proba-ble AD We selected a subset of 528 participants inthe ADNI study We have chosen a training set of101 subjects based on statistics and visual inspectionin order to get representative data which also includedthe difficult images eg with image noise and hugedeformation to allow validation of our methods on ahard data set which makes significant results moreplausibleThe remaining 427 were taken as ADNI-1 data set [31]
minus the overlap with the 101 subjects selected for train-ing Our subset population included 174 NC (age at base-line (bl) 760 years (y) plusmn51 y 89 males (M)85 females(F) 240 MCI (age at bl 749 yplusmn70y 159M81F) and 114AD subjects (age at bl 74 y plusmn73 y 58M56F) Therewas 4 NC 21 MCI and 7 AD subjects in our study thatwas under 65 y Even though there is evidence that thepathology is different in early-onset AD and late-onset ADwe have included the subjects under 65 because they donot have verified early onset AD [32] The demographicdetails of our training and testing subjects are shown inTable 1
Freesurfer segmentationThe segmentation of the regions was performed by staticFreeSurfer [33] implemented on a Linux cluster with 24cores with 18 GB RAM per CPU Freesurfer is a set ofsoftware tools designed to study the cortical and sub-cortical anatomy of the brain Freesurfer do an affineregistration of the volumes with the Talairach atlas [34]a non-uniform intensity normalization (N3) [35] and aB1 bias field correction [36] A skull stripping step wasperformed using a deformable template model Voxelswere then defined as white matter or not white matterbased on intensities Hereafter cutting planes were usedto separate the hemispheres cerebellum and brain stemA cortical and subcortical labeling was performed basedon a transformation that maps the individual subjects intoa probabilistic atlas The atlas was build based on a train-ing set where the subjects have been labeled by hand andcurrently consists of 39 subjects distributed in age and ADpathology (28 NC and 11 with questionable or probableAD) [37] The classification of each point was achieved byfinding the segmentation that maximized the probabilityof input given the prior probability from the training setin a iteratively manner
Grouping of the segmented regionsThe FreeSurfer segmentation provided 40 regions fromwhich a visualization is shown in Figure 1 AD do notspread evenly across the brain and we are interested incapturing early signs of AD and the conversion fromMCIto AD [325] Therefore have we divided our regions intothree groups all functional (func) and potato describedin Table 2 These groups are spread across the brain sowe are not biasing toward anatomical placed groupingsThe all group included the FreeSurfer segmented regionsexcluding left-vessel right vessel and 5th ventricle becausethese regions were not segmented by FreeSurfer in all sub-jects The functional group has excluded all non-functionregions like CSF and hypointensities The choroid plexuswas included in the functional regions due to suggestionsthat the functionality is altered in the choroid plexus dueto AD [38] To get a even smaller subset the potato groupconsisted of small potato shaped regions from a visualperspective where shape is clearly defined
Surface connectivity marker procrustes marker andvolumemarkerWe assume that proximity may reflect aspects of func-tional brain connectivity and have therefore looked atboth the individual regions positional relationship andhow they relate to each other We have calculated thepercentage of how much of a regions own surface wasconnected to the surface of all other regions resultingin a surface connectivity marker This was done non-symmetric in a voxel-count based manner on the three
Lillemark et al BMCMedical Imaging 2014 1421 Page 4 of 12httpwwwbiomedcentralcom1471-23421421
Table 1 The demographic details of our study population
Group Number Age at bl (years) Gender (MF) MMSE at bl
NC training set 24 753 plusmn 44 [651 minus 859] 14 M10 F 293 plusmn 11 [26 minus 30]
MCI training set 29 736 plusmn 73 [552 minus 855] 19 M10 F 272 plusmn 16 [24 minus 30]
AD training set 48 748 plusmn 67 [625 minus 879] 24 M24 F 235 plusmn 19 [21 minus 26]
NC 174 760 plusmn 51 [600 minus 897] 89 M85 F 292 plusmn 10 [25 minus 30]
MCI 240 749 plusmn 70 [552 minus 884] 159 M81 F 271 plusmn 17 [24 minus 30]
AD 114 747 plusmn 73 [565 minus 892] 58 M56 F 233 plusmn 19 [20 minus 26]
MMSE = mini mental state examination score Values are indicated as mean plusmn standard deviation[range
] There is 4 NC 21 MCI and 7 AD subjects in our study that is
under 65
dimensional data so we had a unique image of each regionwhere zero means that there was no connections betweenthe regions and an increasing percentage number referredto how much surface connectivity existed This way wecould observe if shrinkage of regions relates to more fluidin between regions or general shrinkage where the relativesizes did not changeThe individual regions and their internally relationship
was investigated as a change in position of the individualregion We calculated the center of mass c isin R for eachregion P as the mean position of all the points inside theregions in all of the coordinate directions
c middot ed = 12V
Nminus1sumi=0
intAi
(x middot ed)2(ni middot ed) d =123 (1)
where ed denote the standard basis in R by e1 e2 e3 andV denote the volume These points were aligned with aProcrustes alignment where they were adjusted to be in
Figure 1 A slide of the segmented brain where the segmentedregions have different colors
the same space by translation rotation and scaling of thepoints [39] We used the mean shape as the starting shapeThis resulted in a feature vector in a machine learning set-ting that was able to describe the variations in the pointsrelated to the disease statusFor comparison we have used the volume measurement
for the whole brain and for hippocampus for which goodclassification results earlier have been reported [181940]The whole brain volume fraction included all regions inthe skull-stripped brain except for vessels and CSF dividedwith the intracranial volume The hippocampus volumefraction was also measured as the lateral hippocampusvolume divided with the intracranial volume A summaryof our markers is shown in Table 3
Dimensionality reduction and classificationWe wanted to reduce the number of parameters in thecase of Procrustes and surface connectivity due to thecurse of dimensionality where we had more parame-ters than observations We wanted to maintain the rela-tionship between the predictive and target parametersand have therefore chosen to do dimensionality reduc-tion using partial least square regression (PLS) [41] Theidea behind PLS is to find the relevant variables X that
Table 2 The three different groups of the regions allfunctional and potato and the regions belonging to eachgroup
All CSF 3rd-Ventricle 4th-Ventricle Brain-Stem Optic-ChiasmWM-hypointensities non-WM-hypointensities left and rightcerebral white matter cerebral cortex lateral ventricle inflateral ventricle cerebellum white matter cerebellum cortexthalamus caudate putamen pallidum hippocampusamygdala accumbens area ventralDC choroid-plexus
Func Left and right cerebral white matter cerebral cortex inflateral ventricle cerebellum white matter cerebellum cortexthalamus caudate putamen pallidum hippocampusamygdala accumbens area choroid-plexus
Potato Left and right lateral ventricle cerebralwhite matter thalamus caudate putamen pallidumhippocampus amygdala
Lillemark et al BMCMedical Imaging 2014 1421 Page 5 of 12httpwwwbiomedcentralcom1471-23421421
Table 3 An overview of the names and description of themarkers we used in this paper
Marker Description
Procrustes The center of mass of each regions alignedto the same space with a Procrustes alignment
Surface connectivity The percentage of how much each regionhave connected to other regions relatedto the surface of the region
Hippocampus volume The volume of the hippocampus dividedwith the intracranial volume
Whole brain volume The volume of the whole brain dividedwith the intracranial volume
accounts for as much information of the data Y as pos-sible PLS searches for the set of components (latentvariables) that performs a simulation decomposition of Xand Y with the constraint that these components shouldexplain as much as possible of the covariance betweenX and Y It is followed by a linear regression step wherethe decomposition of X is used to predict Y The PLSmodel will try to find the multidimensionality direction inthe X space that explains the maximummultidimensionalvariance direction in the Y space The number of PLScomponents were set to 10 based on our training experi-ments Due to its simple functionality we have used lineardiscriminate analysis (LDA) for the classification [42]LDA tries to reduce the dimensionality while preservingas much of the class discriminatory information as pos-sible LDA seeks to obtain a scalar y by projecting thesamples x onto a line y = wTx where x is the samplesand w contains the class information Of all possible waysto discriminate these we would like to select the one thatmaximizes the separability between the scalars yAll experiments were done in a leave-one-of-each-
class out fashion The data were adjusted for age andgender when there existed a linear correlation betweenthose
ResultThe fractional volume scores for the whole brain volumeand hippocampus volume for NC MCI and AD respec-tively is shown in Table 4 NC had a larger volume inboth whole brain and hippocampus than MCI and ADand MCI had a larger volume score than AD AD had thelargest volume lost between bl and m12For each feature set the area under the curve (AUC) was
computed and summarized in Table 5 for NC versus ADNC versus MCI and MCI versus AD and the correspond-ing ROC curves are shown in Figure 2 The classificationwas tested with a ranksum test and the p-values are alsoshown in Table 5 All markers were able to significantlydiscriminate between the three groups NC-AD NC-MCIand MCI-AD The AUC score were highest for the NC-AD group where our surface connectivity marker werecomparable to the hippocampus volume for the AD-NCand NC-MCI cases and better in the discrimination forthe MCI-AD case than the hippocampus volume TheAUC for the Procrustes marker were in general a littlelower than for the surface connectivity scoreNext we adjusted our markers for whole brain volume
and for hippocampus volume to investigate if our mark-ers contained additional information than the volumesThese results are shown in Table 5 The signal lowersbut was still significant Again the surface connectivitymarkers performed better then the Procrustes markersand the NC-AD classification result were the best Thesurface connectivity markers were generally better to dis-criminate NC-MCI than MCI-AD and for the Procrustesmarkers it was vice versa It was the smaller group-ings functional and potato-shaped that gave the bestperformanceWe have also investigated how our markers performed
on the period to month 12 using the score differencesbetween bl and month 12 for each marker and the AUCand the corresponding ranksum p-values are shown inTable 6 and roc curves in Figure 3 Hippocampus andwhole brain showed relatively low AUC result due to the
Table 4 Fractional volume scores for the hippocampus and the whole brain at bl andmonth 12 and the volume loss
Group Time point Whole brain Hippocampusvolume fraction (cm3) volume fraction (cm3)
The last two markers were divided in three groups all functional and potato-shaped 5(a) is the non-adjusted case 5(b) and 5(c) is adjusted by whole brain fractionand hippocampus fraction respectively All markers were able to significantly distinguish the classes Our markers were still significant after adjustment for the twovolume scores but AUC scores were in general lower than the non-adjusted scores The surface connectivity score for the functional groups performed the best
use of static Freesurfer volumes from bl and month 12Our surface connectivity scores performed the best for allthree groups NC-AD NC-MCI andMCI-AD The resultsbetween NC-AD and NC-MCI are very similarWe have adjusted the month 12 classification results for
both the baseline whole brain and the baseline hippocam-pus volume shown in Table 6 The results showed a sig-nificant classification for our markers When adjusted forwhole brain volume the surface connectivity performedthe best The classification result for MCI-AD case wasbetter than the NC-AD resultFinally we have classifiedMCI-c against MCI-nc where
the non-adjusted result is shown in Table 7 The sur-face connectivity markers was the only marker that wasable to distinguish the two groups and only in the func-tional and potato-shaped grouping of regions When weadjusted for whole brain volume the surface connectiv-ity marker was still significant with an AUC at 0631
(p = 0012) and for the potato group it was borderlinesignificant with an AUC at 0595 (p = 0067) In thecase where we adjusted for hippocampus volume onlythe surface connectivity marker for the functional groupswas borderline significant with an AUC of 0599 (p =0055) No other significance were shown in the adjustedcases
Discussion and conclusionWe have investigated a novel way of looking at the rela-tionship between different regions in the brain We eval-uated a surface connectivity marker and center of massbased marker and their ability to classify between NCMCI and AD subjects Both markers have been able tosignificantly discriminate between the three classes AD-NC NC-MCI andMCI-AD both at baseline and betweenbaseline and month 12 Our surface connectivity markerwas also able to classify MCI-c
Lillemark et al BMCMedical Imaging 2014 1421 Page 7 of 12httpwwwbiomedcentralcom1471-23421421
Figure 2 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
The large variabilityrsquos in the brain regions is related toAlzheimerrsquos Disease [171925-27] and this have moti-vated our two markers describing the proximity betweenthe regions in the brain Both our markers were ableto significantly differentiate between AD and NC alsowhen adjusted for whole brain and hippocampus volumeThe surface connectivity marker was comparable to hip-pocampus volume which is one of known most effect fullmarkers from MRI Also after adjustment for volumes wehad a significant classification results this indicates thatour markers hold additional information about the devel-opment of the brain in relation to progression of ADWe believe that our markers capture an individual shrink-age due to pathological alterations In subjects with ADthe cerebral cortex is shrinking the sulcirsquos is widenedthe cortical ribbon may be thinned and ventricles aredilated [24344] Our surface connectivity markers maycapture some of these pathological alterations in measur-ing the proximity between regionsWe have evaluated our markers over a 1 - year period
where we have investigated the change in the Procrustesaligned positions and the change in surface connectivityIn this case we were also able to significantly discrim-inate between the classes although the signal was lessstrong The weakened signal can be due to noise in thesegmentation of the data Our markers were not taken
from registered brains but normalized within the samebrain so they captured comparable information acrosstime and study population The segmentation of the indi-vidual regions at two time steps can still be quite differ-ent and when we were using the difference between thescore values it can introduce noise in our markers Thisis also visible in the values for hippocampus and wholebrain volume in the longitudinal part of our study whichshowed lower results for classification than other reportedresults [1745]Our surface connectivity marker performed the best
indicating that it captured how the cell death caused byAD minimizes the surface connectivity between regionsThis was most visible in the functional regions The func-tional group were limited to functional regions of thebrain and the good performance of this grouping is in linewith the knowledge that AD affect the network aroundand including the medial temporal lobe and disruption inthis region contributes to memory impairment [46] Thelower performance of our Procrustes marker could be dueto the captured information is closer to volume and thatno particular regions moves related to the others but allregions moved due to general volume lossCuingnet et al [18] have made a comparison study
for classification of NC versus AD NC versus MCI-converters (MCI-c) and MCI-c versus MCI-non-
Lillemark et al BMCMedical Imaging 2014 1421 Page 8 of 12httpwwwbiomedcentralcom1471-23421421
Table 6 Classification result for NC-AD NC-MCI andMCI-AD for the difference between the bl andmonth 12makers 6(a)is the not adjusted case 6(b) is adjusted for bl whole brain volume and 6(c) is adjusted for baseline hippocampus volume
Our markers was still able to significantly discriminate between the three groups Our surface connectivity markers for the two subgroups functional and potatoperformed the best
converters (MCI-nc) based on 81 NC 67 MCI-nc 39MCI-c and 69 AD subjects from the ADNI databaseThey investigated voxel based segmented tissue regionsfor the whole brain in six different variants and for graymatter (GM) and GM white matter (WM) and cere-brospinal fluid (CSF) combined cortical thickness inthree different variants and finally hippocampus volumeand shape in three different variants a total of ten differ-ent methods They conclude that all methods were able toclassify NC vs AD with a sensitivity and specificity at therange from 59 - 81 and 77 - 98 respectively whichis comparable to our classification Other prediction stud-ies have shown better classification rates at 67 - 92 forcross-sectional studies [1417194547] and 69 - 815for longitudinal studies [19-21] The difference in theclassification accuracy between our method and the otherpapers can be explained by the tuning of methods and theuse of different data sets
Only our surface connectivity marker was able to clas-sify MCI-c fromMCI-nc and not with a highly significantresult This is in line with Cuignet et al comparison studyfor AD classification where they found that only fourmethods managed to predict MCI-c vs MCI-nc betterthan a random classifier and none of those got signifi-cantly better results [18] The main reason for the lowresult in the conversion case could be due to the fact thatMCI is a very in heterogeneous group that possibly couldconvert rapidly to AD or be stable for many years beforeconversionOther studies have investigated the change locally in
the hippocampus Wang et al [13] have used large-deformation diffeomorphic high-dimensional brain map-ping to quantify and compare changes in the hippocampalshape as well as volume They found that shape changeswere largely confined to the head of hippocampus andsubiculum for normal controls (NC) Other studies have
Lillemark et al BMCMedical Imaging 2014 1421 Page 9 of 12httpwwwbiomedcentralcom1471-23421421
Figure 3 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
confirmed these shape changes for the hippocampus[14-16] based on shape models and local hippocampalatrophy patterns We have focused on investigating therelationship between the different regions of the brain andhow they differ between healthy subjects and AD patientsThis way of investigating the regions could make it pos-sible to incorporate different kind of knowledge into thesame model where one could go from the individual scaleof each region to the interaction between the regionsand finally to combined picture of the brain as one wholeregion
Table 7 The AUC and corresponding p-values for theclassification of MCI-c andMCI-nc
Markers AUC pminusvalue
HPICV 0466 0516
WBICV 0512 0823
Surface all 0542 0416
Surface func 0624 0017
Surface potato 0603 0048
Procrustes all 0465 0486
Procrustes func 0498 0964
Procrustes potato 0534 0501
Only the surface connectivity markers was able to significantly discriminate thetwo groups functional and potato-shaped
An alternative use of MRI images for early predictionof AD is by using texture analysis where different texturesfeatures is used to construct a computational frameworkwhich have been able to discriminate AD MCI and NCwith a separability of up to 95 [234048] This indicatesthat one can combine the three different kinds of mark-ers volume texture and shapeproximity markers to get amore sophisticated picture of the disease progressionOther image modalities such as single-photon emis-
sion computed tomography (SPECT) functional MRI andMR spectroscopy (MRS) positron emission tomography(PET) and molecular imaging have been used for investi-gation of brain changes related to AD SPECT combinedwith MRI images can give additional information aboutdisease progression when combined [49] Functional MRIand MR spectroscopy (MRS) have shown changes inmetabolic levels even prior to symptom onset in ADbut are difficult to implement in clinical settings due totechnical support [5051] PET metabolic imaging withradioactive glucose has also been used to examined thefunctional change and tracking of the AD disease progres-sion [5253] Due to the invasiveness radiation dose limi-tation requiring lumbar punctures and high cost PET isunsuitable for repeated measurements of a single patientor screening programs for large populations Molecularimaging with amyloid tracers have showed great potential
Lillemark et al BMCMedical Imaging 2014 1421 Page 10 of 12httpwwwbiomedcentralcom1471-23421421
as to be accurate markers for early diagnosis of AD but donot show progression in established disease [5455] whichis our object of interestTo conclude structural MRI is an suitable image modal-
ity for detection of AD and AD progression Our mark-ers have shown promising results in capturing how theproximity of different regions in the brain can aid inAD diagnosis and prognosis The proximity analysis cap-tures additional information about the whole brain com-pared to atrophy scores This additional information cancontribute to the refinement of the AD markers andmay be able to give a more detailed picture of ADprogression
Competing interestsThe authors declare that they have no competing interests
Authorsrsquo contributionsLL have contributed in study design data analysis and interpretation preparedand submitted the manuscript LS and AP performed study design and datacollection EBD and MN participated in design and reviewed manuscript Allauthors have read and approved the final manuscript
AcknowledgementsWe gratefully acknowledge the funding from the Danish Research Foundation(Den Danske Forskningsfond) and The Danish National Advanced TechnologyFoundation supporting this work and FreeSurfer for providing the softwareused for the segmentations in this paper Data collection and sharing for thisproject was funded by the Alzheimerrsquos Disease Neuroimaging Initiative (ADNI)(National Institutes of Health Grant U01 AG024904) and DOD ADNI(Department of Defense award number W81XWH-12-2-0012) ADNI is fundedby the National Institute on Aging the National Institute of Bio medicalImaging and Bioengineering and through generous contributions from thefollowing Alzheimerrsquos Association Alzheimerrsquos Drug Discovery FoundationBioClinica Inc Biogen Idec Inc Bristol-Myers Squibb Company Eisai Inc ElanPharmaceuticals Inc Eli Lilly and Company F Hoffmann-La Roche Ltd and itsaffiliated company Genentech Inc GE Healthcare Innogenetics NV IXICOLtd Janssen Alzheimer Immunotherapy Research amp Development LLCJohnson amp Johnson Pharmaceutical Research amp Development LLC MedpaceInc Merck amp Co Inc Meso Scale Diagnostics LLC NeuroRx Research NovartisPharmaceuticals Corporation Pfizer Inc Piramal Imaging Servier Synarc Incand Takeda Pharmaceutical Company The Canadian Institutes of HealthResearch is providing funds to Rev December 5 2013 support ADNI clinicalsites in Canada Private sector contributions are facilitated by the Foundationfor the National Institutes of Health (wwwfnihorg) The grantee organizationis the Northern California Institute for Research and Education and the study iscoordinated by the Alzheimerrsquos Disease Cooperative Study at the University ofCalifornia San Diego ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern CaliforniaData used in preparation of this article were obtained from the AlzheimerrsquosDisease Neuroimaging Initiative (ADNI) database (adniloniuscedu) As suchthe investigators within the ADNI contributed to the design andimplementation of ADNI andor provided data but did not participate inanalysis or writing of this report A complete listing of ADNI investigators canbe found at httpadniloniusceduwp-contentuploadshow_to_applyADNI_Acknowledgement_Listpdf
Author details1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash Denmark 2Biomediq Fruebjergvej3 2100 Copenhagen Oslash Denmark
Received 2 January 2014 Accepted 9 May 2014Published 2 June 2014
References1 Alzheimerrsquos association 2011
[httpwwwalzorgdownloadsFacts_Figures_2011pdf]2 Braskie MN Klunder AD Hayashi KM Protas H Kepe V Miller KJ Huang SC
Barrio JR Ercoli LM Siddarth P Satyamurthy N Liu J Toga AWBookheimer SY Small GW Thompson PM Plaque and tangle imagingand cognition in normal aging and Alzheimerrsquos disease NeurobiolAging 2010 311669ndash1678
3 Braak H Braak E Neuropathological stageing of alzheimer-relatedchanges Acta neuropathologica 1991 82(4)239ndash259
4 West MJ Coleman PD Flood DG Troncoso JC Differences in thepattern of hippocampal neuronal loss in normal ageing andAlzheimerrsquos disease Lancet 1994 344769ndash772
5 Apostolova LG Mosconi L Thompson PM Green AE Hwang KS RamirezA Mistur R Tsui WH de Leon MJ Subregional hippocampal atrophypredicts alzheimerrsquos dementia in the cognitively normal NeurobiolAging 2010 31(7)1077ndash1088
6 Tondelli M Wilcock GK Nichelli P De Jager CA Jenkinson M Zamboni GStructural mri changes detectable up to ten years before clinicalalzheimerrsquos disease Neurobiol Aging 2012 33(4)825ndash25
7 Bernard C Helmer C Dilharreguy B Amieva H Auriacombe S DartiguesJ-F Allard M Catheline G Time course of brain volume changes in thepreclinical phase of alzheimerrsquos disease Alzheimerrsquos Dementia 201410(2)143ndash151
8 Dickerson B Stoub T Shah R Sperling R Killiany R Albert M Hyman BBlacker D deToledo-Morrell L Alzheimer-signature mri biomarkerpredicts ad dementia in cognitively normal adults Neurology 201176(16)1395ndash1402
9 Hansson O Zetterberg H Buchhave P Londos E Blennow K Minthon LAssociation between csf biomarkers and incipient alzheimerrsquosdisease in patients with mild cognitive impairment a follow-upstudy Lancet Neurol 2006 5(3)228ndash234
10 Leung KK Shen K-K Barnes J Ridgway GR Clarkson MJ Fripp JSalvado O Meriaudeau F Fox NC Bourgeat P Ourselin S Increasingpower to predict mild cognitive impairment conversion toalzheimerrsquos disease using hippocampal atrophy rate andstatistical shape models In Proceedings of the 13th InternationalConference onMedical Image Computing and Computer-assistedIntervention Part II MICCAIrsquo10 Berlin Heidelberg Springer2010125ndash132
11 Holland D Dale AM Nonlinear registration of longitudinal imagesandmeasurement of change in regions of interestMed Image Anal2011 15(4)489ndash497
12 Smith SM Zhang Y Jenkinson M Chen J Matthews P Federico ADe Stefano N Accurate robust and automated longitudinal andcross-sectional brain change analysis Neuroimage 200217(1)479ndash489
13 Wang L Swank JS Glick IE Gado MH Miller MI Morris JC Csernansky JGChanges in hippocampal volume and shape across time distinguishdementia of the Alzheimer type from healthy aging Neuroimage2003 20667ndash682
14 Li S Shi F Pu F Li X Jiang T Xie S Wang Y Hippocampal shape analysisof Alzheimer disease based onmachine learning methods AJNR AmJ Neuroradiol 2007 281339ndash1345
15 Costafreda SG Dinov ID Tu Z Shi Y Liu CY Kloszewska I Mecocci PSoininen H Tsolakif M Vellasg B Wahlundh L-O Spengerh C Togab AWLovestonea S Simmonsa A Automated hippocampal shape analysispredicts the onset of dementia in mild cognitive impairmentNeuroImage 2011
16 Scher AI Xu Y Korf ES White LR Scheltens P Toga AW Thompson PMHartley SW Witter MP Valentino DJ Launer LJ Hippocampal shapeanalysis in Alzheimerrsquos disease a population-based studyNeuroimage 2007 368ndash18
17 Klein S Loog M van der Lijn F den Heijer T Hammers A de Bruijne Mvan der Lugt A Duin RPW Breteler MMB Niessen WJ Early diagnosis ofdementia based on intersubject whole-brain dissimilarities InProceedings of the 2010 IEEE International Conference on BiomedicalImaging fromNano toMacro ISBIrsquo10 Piscataway NJ USA IEEE Press2010249ndash252
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 2 of 12httpwwwbiomedcentralcom1471-23421421
found correlation with clinical measurements and positiveprediction of future decline [10-12] MRI is non invasiveand is largely available in the clinical environment andhas therefore become a viable tool to monitor the progres-sion of AD Longitudinal measurement changes may bemore objective precise and reproducible when they aremeasured from MRI compared to diagnostic image mea-surements from PET scanning or CSF measurement ofamyloid and tau proteinDifferent studies have focused on the hippocampus
region and used atrophy scoring and shape analysis fordetection of AD [13-16] Also whole brain atrophy scor-ing have been extensively used for detection of AD [1718]The same techniques have been used to classify MCI-converters (MCI-c) from MCI-non-converters (MCI-nc)indicating that it is possible to make a prognosis of ADbased on atrophy rates shape analysis region hippocam-pal shapes and machine learning techniques [1019-24]The general focus on studies from MRI have been on
the atrophy rates for hippocampus or the whole brainor the shape of hippocampus but other brain structureslike amygdala putamen thalamus and the ventricles havealso shown relation to AD [25-27] We want to includeall of these structures in order to investigated the rela-tionship and proximity between different regions in brainin hope to characterize how the brain develops and usethis as a marker for AD We believe that the relationshipbetween the positions of different regions or the surfaceconnectivity between the different regions in the brain cancapture how the atrophy spreads We have used a Pro-crustes marker that classified AD subjects based on theposition of the center of mass of each region in a Pro-crustes aligned environment and a surface connectivitymarker that extracted the percentile surface connectivitybetween the individual regions The Procrustes markercan capture how the regions move away or toward eachother indicating how the volume loss is different acrossthe brain The surface connectivity marker can describethe individual volume loss of the regions and how theymove apart due to for example the increase in ventriclesand cerebrospinal fluid (CSF) These new markers couldgive a more detailed view of the AD progression and maybe used in addition to the traditional morphometric mark-ers Our markers were used in three different groupingsof the brain regions a group of all Freesurfer segmentedregions a subset of the functional regions and a subset ofthe small potato shaped regions (for example hippocam-pus and amygdala) to classify using a linear discriminantanalysis NC MCI and AD This was done in compari-son to the whole brain volume and hippocampus volumePotentially this could lead to a fine-to-coarse scale fromwhere one can study the progression of AD from theglobal brain scale down to the local scale of the shapeandor texture of the individual sub-regions
MethodsADNI brain MRI and preprocessingData was obtained from the Alzheimerrsquos Disease Neu-roimaging Initiative (ADNI) database (adniloniuscedu) [28] The ADNIwas launched in 2003 by theNationalInstitute for Aging (NIA) the National Institute of Bio-chemical Imaging and Bioengineering (NIBIB) the Foodand Drug Administration (FDA) private pharmaceuticalcompanies and non-profit organizations as a $ 60 mil-lion 5-year public-private partnership The primary goalof ADNI has been to test whether serial MRI positronemission tomography PET other biological markers andclinical and neurophysiological assessments can be com-bined to measure the progression of MCI and early ADDetermination of sensitive and specific markers of veryearly AD progression is intended to aid researchers andclinicians to develop new treatments and monitor theireffectiveness as well as lessen the time and cost of clin-ical trials The Principal Investigator of this initiative isMichael W Weiner MD VA Medical Center and Uni-versity of California - San Francisco ADNI is the resultof efforts of many co-investigators from a broad range ofacademic institutions and private corporations and sub-jects have been recruited from over 50 sites across the USand Canada The initial goal of ADNI was to recruit 800subjects but ADNI has been followed by ADNI-GO andADNI-2 To date these three protocols have recruited over1500 adults ages 55 to 90 to participate in the researchconsisting of cognitively normal older individuals peoplewith early or late MCI and people with early AD The fol-low up duration of each group is specified in the protocolsfor ADNI-1 ADNI-2 and ADNI-GO Subjects originallyrecruited for ADNI-1 and ADNI-GO had the option tobe followed in ADNI-2 For up-to-date information seewwwadni-infoorgLongitudinal brain T1 weighted MRI and associated
data for the study population including age gender anddiagnosis were downloaded from the ADNI database Alldata in this paper were from ADNI-1 ADNI-1 was afive year study launched in 2004 to develop longitudi-nal outcome measures of Alzheimerrsquos progression usingserial MRI PET biochemical changes in CSF blood andurine and cognitive and neuropsychological assessmentacquired at multiple sites similar to typical clinical tri-als All subjects underwent clinical and cognitive assess-ment at the time of scan acquisition All AD subjectsmet NINCDSADRDA criteria for probable AD [29] Thestudy was conducted according to the Good Clinical Prac-tice guidelines the Declaration of Helsinki and US 21CFR Part-50 Protection of Human Subject and Part 56-Institutional Review Boards This study was approved bythe Institutional Review Boards of all of the participatinginstitutions and informed written consent was obtainedfrom all participants at each site
Lillemark et al BMCMedical Imaging 2014 1421 Page 3 of 12httpwwwbiomedcentralcom1471-23421421
MRI acquisitionHigh-Resolution structural brain MRI were acquired at59 ADNI sites using 15 Tesla T1-weighted MRI scansusing volumetric 3D MPRAGE or equivalent protocolswith varying resolution typically 125 times 125 mm in-plane spatial resolution and 12 mm thick sagital slicesThe MPRAGE sequence was acquired twice for all sub-jects at each visit to improve the chance that at least onescan would be suitable for analysis The image quality wasgraded qualitatively by ADNI investigators of the ADNIMRQ quality control center at the Mayo Clinic for arti-facts and general image quality Each scan was graded onseveral separate criteria blurringghosting flow artifactsintensity a homogeneity signal-to-noise ratio susceptibil-ity artifacts and gray-whitecerebrospinal fluid contrastWe have only used the MRI scan which was graded asthe best scan for each subject No other exclusion crite-ria based on image quality were applied We have used theraw ADNI data
ParticipantsThe criteria for inclusion were those defined in theADNI protocol normal control (NC) subjects had amini mental state examamination score (MMSE) between24 - 30 a clinical dementia rating (CDR) score of zerothey were non-depressed non MCI and non-dementedMCI had MMSE scores between 24-30 a memory com-plaint had objective memory loss measured by educationadjusted scores onWechsler Memory Scale Logical Mem-ory II [30] a CDR of 05 absence of significant lev-els of impairment in other cognitive domains AD sub-jects had MMSE scores between 20-26 CDR of 05or 10 and met NINCDSADRDA criteria for proba-ble AD We selected a subset of 528 participants inthe ADNI study We have chosen a training set of101 subjects based on statistics and visual inspectionin order to get representative data which also includedthe difficult images eg with image noise and hugedeformation to allow validation of our methods on ahard data set which makes significant results moreplausibleThe remaining 427 were taken as ADNI-1 data set [31]
minus the overlap with the 101 subjects selected for train-ing Our subset population included 174 NC (age at base-line (bl) 760 years (y) plusmn51 y 89 males (M)85 females(F) 240 MCI (age at bl 749 yplusmn70y 159M81F) and 114AD subjects (age at bl 74 y plusmn73 y 58M56F) Therewas 4 NC 21 MCI and 7 AD subjects in our study thatwas under 65 y Even though there is evidence that thepathology is different in early-onset AD and late-onset ADwe have included the subjects under 65 because they donot have verified early onset AD [32] The demographicdetails of our training and testing subjects are shown inTable 1
Freesurfer segmentationThe segmentation of the regions was performed by staticFreeSurfer [33] implemented on a Linux cluster with 24cores with 18 GB RAM per CPU Freesurfer is a set ofsoftware tools designed to study the cortical and sub-cortical anatomy of the brain Freesurfer do an affineregistration of the volumes with the Talairach atlas [34]a non-uniform intensity normalization (N3) [35] and aB1 bias field correction [36] A skull stripping step wasperformed using a deformable template model Voxelswere then defined as white matter or not white matterbased on intensities Hereafter cutting planes were usedto separate the hemispheres cerebellum and brain stemA cortical and subcortical labeling was performed basedon a transformation that maps the individual subjects intoa probabilistic atlas The atlas was build based on a train-ing set where the subjects have been labeled by hand andcurrently consists of 39 subjects distributed in age and ADpathology (28 NC and 11 with questionable or probableAD) [37] The classification of each point was achieved byfinding the segmentation that maximized the probabilityof input given the prior probability from the training setin a iteratively manner
Grouping of the segmented regionsThe FreeSurfer segmentation provided 40 regions fromwhich a visualization is shown in Figure 1 AD do notspread evenly across the brain and we are interested incapturing early signs of AD and the conversion fromMCIto AD [325] Therefore have we divided our regions intothree groups all functional (func) and potato describedin Table 2 These groups are spread across the brain sowe are not biasing toward anatomical placed groupingsThe all group included the FreeSurfer segmented regionsexcluding left-vessel right vessel and 5th ventricle becausethese regions were not segmented by FreeSurfer in all sub-jects The functional group has excluded all non-functionregions like CSF and hypointensities The choroid plexuswas included in the functional regions due to suggestionsthat the functionality is altered in the choroid plexus dueto AD [38] To get a even smaller subset the potato groupconsisted of small potato shaped regions from a visualperspective where shape is clearly defined
Surface connectivity marker procrustes marker andvolumemarkerWe assume that proximity may reflect aspects of func-tional brain connectivity and have therefore looked atboth the individual regions positional relationship andhow they relate to each other We have calculated thepercentage of how much of a regions own surface wasconnected to the surface of all other regions resultingin a surface connectivity marker This was done non-symmetric in a voxel-count based manner on the three
Lillemark et al BMCMedical Imaging 2014 1421 Page 4 of 12httpwwwbiomedcentralcom1471-23421421
Table 1 The demographic details of our study population
Group Number Age at bl (years) Gender (MF) MMSE at bl
NC training set 24 753 plusmn 44 [651 minus 859] 14 M10 F 293 plusmn 11 [26 minus 30]
MCI training set 29 736 plusmn 73 [552 minus 855] 19 M10 F 272 plusmn 16 [24 minus 30]
AD training set 48 748 plusmn 67 [625 minus 879] 24 M24 F 235 plusmn 19 [21 minus 26]
NC 174 760 plusmn 51 [600 minus 897] 89 M85 F 292 plusmn 10 [25 minus 30]
MCI 240 749 plusmn 70 [552 minus 884] 159 M81 F 271 plusmn 17 [24 minus 30]
AD 114 747 plusmn 73 [565 minus 892] 58 M56 F 233 plusmn 19 [20 minus 26]
MMSE = mini mental state examination score Values are indicated as mean plusmn standard deviation[range
] There is 4 NC 21 MCI and 7 AD subjects in our study that is
under 65
dimensional data so we had a unique image of each regionwhere zero means that there was no connections betweenthe regions and an increasing percentage number referredto how much surface connectivity existed This way wecould observe if shrinkage of regions relates to more fluidin between regions or general shrinkage where the relativesizes did not changeThe individual regions and their internally relationship
was investigated as a change in position of the individualregion We calculated the center of mass c isin R for eachregion P as the mean position of all the points inside theregions in all of the coordinate directions
c middot ed = 12V
Nminus1sumi=0
intAi
(x middot ed)2(ni middot ed) d =123 (1)
where ed denote the standard basis in R by e1 e2 e3 andV denote the volume These points were aligned with aProcrustes alignment where they were adjusted to be in
Figure 1 A slide of the segmented brain where the segmentedregions have different colors
the same space by translation rotation and scaling of thepoints [39] We used the mean shape as the starting shapeThis resulted in a feature vector in a machine learning set-ting that was able to describe the variations in the pointsrelated to the disease statusFor comparison we have used the volume measurement
for the whole brain and for hippocampus for which goodclassification results earlier have been reported [181940]The whole brain volume fraction included all regions inthe skull-stripped brain except for vessels and CSF dividedwith the intracranial volume The hippocampus volumefraction was also measured as the lateral hippocampusvolume divided with the intracranial volume A summaryof our markers is shown in Table 3
Dimensionality reduction and classificationWe wanted to reduce the number of parameters in thecase of Procrustes and surface connectivity due to thecurse of dimensionality where we had more parame-ters than observations We wanted to maintain the rela-tionship between the predictive and target parametersand have therefore chosen to do dimensionality reduc-tion using partial least square regression (PLS) [41] Theidea behind PLS is to find the relevant variables X that
Table 2 The three different groups of the regions allfunctional and potato and the regions belonging to eachgroup
All CSF 3rd-Ventricle 4th-Ventricle Brain-Stem Optic-ChiasmWM-hypointensities non-WM-hypointensities left and rightcerebral white matter cerebral cortex lateral ventricle inflateral ventricle cerebellum white matter cerebellum cortexthalamus caudate putamen pallidum hippocampusamygdala accumbens area ventralDC choroid-plexus
Func Left and right cerebral white matter cerebral cortex inflateral ventricle cerebellum white matter cerebellum cortexthalamus caudate putamen pallidum hippocampusamygdala accumbens area choroid-plexus
Potato Left and right lateral ventricle cerebralwhite matter thalamus caudate putamen pallidumhippocampus amygdala
Lillemark et al BMCMedical Imaging 2014 1421 Page 5 of 12httpwwwbiomedcentralcom1471-23421421
Table 3 An overview of the names and description of themarkers we used in this paper
Marker Description
Procrustes The center of mass of each regions alignedto the same space with a Procrustes alignment
Surface connectivity The percentage of how much each regionhave connected to other regions relatedto the surface of the region
Hippocampus volume The volume of the hippocampus dividedwith the intracranial volume
Whole brain volume The volume of the whole brain dividedwith the intracranial volume
accounts for as much information of the data Y as pos-sible PLS searches for the set of components (latentvariables) that performs a simulation decomposition of Xand Y with the constraint that these components shouldexplain as much as possible of the covariance betweenX and Y It is followed by a linear regression step wherethe decomposition of X is used to predict Y The PLSmodel will try to find the multidimensionality direction inthe X space that explains the maximummultidimensionalvariance direction in the Y space The number of PLScomponents were set to 10 based on our training experi-ments Due to its simple functionality we have used lineardiscriminate analysis (LDA) for the classification [42]LDA tries to reduce the dimensionality while preservingas much of the class discriminatory information as pos-sible LDA seeks to obtain a scalar y by projecting thesamples x onto a line y = wTx where x is the samplesand w contains the class information Of all possible waysto discriminate these we would like to select the one thatmaximizes the separability between the scalars yAll experiments were done in a leave-one-of-each-
class out fashion The data were adjusted for age andgender when there existed a linear correlation betweenthose
ResultThe fractional volume scores for the whole brain volumeand hippocampus volume for NC MCI and AD respec-tively is shown in Table 4 NC had a larger volume inboth whole brain and hippocampus than MCI and ADand MCI had a larger volume score than AD AD had thelargest volume lost between bl and m12For each feature set the area under the curve (AUC) was
computed and summarized in Table 5 for NC versus ADNC versus MCI and MCI versus AD and the correspond-ing ROC curves are shown in Figure 2 The classificationwas tested with a ranksum test and the p-values are alsoshown in Table 5 All markers were able to significantlydiscriminate between the three groups NC-AD NC-MCIand MCI-AD The AUC score were highest for the NC-AD group where our surface connectivity marker werecomparable to the hippocampus volume for the AD-NCand NC-MCI cases and better in the discrimination forthe MCI-AD case than the hippocampus volume TheAUC for the Procrustes marker were in general a littlelower than for the surface connectivity scoreNext we adjusted our markers for whole brain volume
and for hippocampus volume to investigate if our mark-ers contained additional information than the volumesThese results are shown in Table 5 The signal lowersbut was still significant Again the surface connectivitymarkers performed better then the Procrustes markersand the NC-AD classification result were the best Thesurface connectivity markers were generally better to dis-criminate NC-MCI than MCI-AD and for the Procrustesmarkers it was vice versa It was the smaller group-ings functional and potato-shaped that gave the bestperformanceWe have also investigated how our markers performed
on the period to month 12 using the score differencesbetween bl and month 12 for each marker and the AUCand the corresponding ranksum p-values are shown inTable 6 and roc curves in Figure 3 Hippocampus andwhole brain showed relatively low AUC result due to the
Table 4 Fractional volume scores for the hippocampus and the whole brain at bl andmonth 12 and the volume loss
Group Time point Whole brain Hippocampusvolume fraction (cm3) volume fraction (cm3)
The last two markers were divided in three groups all functional and potato-shaped 5(a) is the non-adjusted case 5(b) and 5(c) is adjusted by whole brain fractionand hippocampus fraction respectively All markers were able to significantly distinguish the classes Our markers were still significant after adjustment for the twovolume scores but AUC scores were in general lower than the non-adjusted scores The surface connectivity score for the functional groups performed the best
use of static Freesurfer volumes from bl and month 12Our surface connectivity scores performed the best for allthree groups NC-AD NC-MCI andMCI-AD The resultsbetween NC-AD and NC-MCI are very similarWe have adjusted the month 12 classification results for
both the baseline whole brain and the baseline hippocam-pus volume shown in Table 6 The results showed a sig-nificant classification for our markers When adjusted forwhole brain volume the surface connectivity performedthe best The classification result for MCI-AD case wasbetter than the NC-AD resultFinally we have classifiedMCI-c against MCI-nc where
the non-adjusted result is shown in Table 7 The sur-face connectivity markers was the only marker that wasable to distinguish the two groups and only in the func-tional and potato-shaped grouping of regions When weadjusted for whole brain volume the surface connectiv-ity marker was still significant with an AUC at 0631
(p = 0012) and for the potato group it was borderlinesignificant with an AUC at 0595 (p = 0067) In thecase where we adjusted for hippocampus volume onlythe surface connectivity marker for the functional groupswas borderline significant with an AUC of 0599 (p =0055) No other significance were shown in the adjustedcases
Discussion and conclusionWe have investigated a novel way of looking at the rela-tionship between different regions in the brain We eval-uated a surface connectivity marker and center of massbased marker and their ability to classify between NCMCI and AD subjects Both markers have been able tosignificantly discriminate between the three classes AD-NC NC-MCI andMCI-AD both at baseline and betweenbaseline and month 12 Our surface connectivity markerwas also able to classify MCI-c
Lillemark et al BMCMedical Imaging 2014 1421 Page 7 of 12httpwwwbiomedcentralcom1471-23421421
Figure 2 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
The large variabilityrsquos in the brain regions is related toAlzheimerrsquos Disease [171925-27] and this have moti-vated our two markers describing the proximity betweenthe regions in the brain Both our markers were ableto significantly differentiate between AD and NC alsowhen adjusted for whole brain and hippocampus volumeThe surface connectivity marker was comparable to hip-pocampus volume which is one of known most effect fullmarkers from MRI Also after adjustment for volumes wehad a significant classification results this indicates thatour markers hold additional information about the devel-opment of the brain in relation to progression of ADWe believe that our markers capture an individual shrink-age due to pathological alterations In subjects with ADthe cerebral cortex is shrinking the sulcirsquos is widenedthe cortical ribbon may be thinned and ventricles aredilated [24344] Our surface connectivity markers maycapture some of these pathological alterations in measur-ing the proximity between regionsWe have evaluated our markers over a 1 - year period
where we have investigated the change in the Procrustesaligned positions and the change in surface connectivityIn this case we were also able to significantly discrim-inate between the classes although the signal was lessstrong The weakened signal can be due to noise in thesegmentation of the data Our markers were not taken
from registered brains but normalized within the samebrain so they captured comparable information acrosstime and study population The segmentation of the indi-vidual regions at two time steps can still be quite differ-ent and when we were using the difference between thescore values it can introduce noise in our markers Thisis also visible in the values for hippocampus and wholebrain volume in the longitudinal part of our study whichshowed lower results for classification than other reportedresults [1745]Our surface connectivity marker performed the best
indicating that it captured how the cell death caused byAD minimizes the surface connectivity between regionsThis was most visible in the functional regions The func-tional group were limited to functional regions of thebrain and the good performance of this grouping is in linewith the knowledge that AD affect the network aroundand including the medial temporal lobe and disruption inthis region contributes to memory impairment [46] Thelower performance of our Procrustes marker could be dueto the captured information is closer to volume and thatno particular regions moves related to the others but allregions moved due to general volume lossCuingnet et al [18] have made a comparison study
for classification of NC versus AD NC versus MCI-converters (MCI-c) and MCI-c versus MCI-non-
Lillemark et al BMCMedical Imaging 2014 1421 Page 8 of 12httpwwwbiomedcentralcom1471-23421421
Table 6 Classification result for NC-AD NC-MCI andMCI-AD for the difference between the bl andmonth 12makers 6(a)is the not adjusted case 6(b) is adjusted for bl whole brain volume and 6(c) is adjusted for baseline hippocampus volume
Our markers was still able to significantly discriminate between the three groups Our surface connectivity markers for the two subgroups functional and potatoperformed the best
converters (MCI-nc) based on 81 NC 67 MCI-nc 39MCI-c and 69 AD subjects from the ADNI databaseThey investigated voxel based segmented tissue regionsfor the whole brain in six different variants and for graymatter (GM) and GM white matter (WM) and cere-brospinal fluid (CSF) combined cortical thickness inthree different variants and finally hippocampus volumeand shape in three different variants a total of ten differ-ent methods They conclude that all methods were able toclassify NC vs AD with a sensitivity and specificity at therange from 59 - 81 and 77 - 98 respectively whichis comparable to our classification Other prediction stud-ies have shown better classification rates at 67 - 92 forcross-sectional studies [1417194547] and 69 - 815for longitudinal studies [19-21] The difference in theclassification accuracy between our method and the otherpapers can be explained by the tuning of methods and theuse of different data sets
Only our surface connectivity marker was able to clas-sify MCI-c fromMCI-nc and not with a highly significantresult This is in line with Cuignet et al comparison studyfor AD classification where they found that only fourmethods managed to predict MCI-c vs MCI-nc betterthan a random classifier and none of those got signifi-cantly better results [18] The main reason for the lowresult in the conversion case could be due to the fact thatMCI is a very in heterogeneous group that possibly couldconvert rapidly to AD or be stable for many years beforeconversionOther studies have investigated the change locally in
the hippocampus Wang et al [13] have used large-deformation diffeomorphic high-dimensional brain map-ping to quantify and compare changes in the hippocampalshape as well as volume They found that shape changeswere largely confined to the head of hippocampus andsubiculum for normal controls (NC) Other studies have
Lillemark et al BMCMedical Imaging 2014 1421 Page 9 of 12httpwwwbiomedcentralcom1471-23421421
Figure 3 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
confirmed these shape changes for the hippocampus[14-16] based on shape models and local hippocampalatrophy patterns We have focused on investigating therelationship between the different regions of the brain andhow they differ between healthy subjects and AD patientsThis way of investigating the regions could make it pos-sible to incorporate different kind of knowledge into thesame model where one could go from the individual scaleof each region to the interaction between the regionsand finally to combined picture of the brain as one wholeregion
Table 7 The AUC and corresponding p-values for theclassification of MCI-c andMCI-nc
Markers AUC pminusvalue
HPICV 0466 0516
WBICV 0512 0823
Surface all 0542 0416
Surface func 0624 0017
Surface potato 0603 0048
Procrustes all 0465 0486
Procrustes func 0498 0964
Procrustes potato 0534 0501
Only the surface connectivity markers was able to significantly discriminate thetwo groups functional and potato-shaped
An alternative use of MRI images for early predictionof AD is by using texture analysis where different texturesfeatures is used to construct a computational frameworkwhich have been able to discriminate AD MCI and NCwith a separability of up to 95 [234048] This indicatesthat one can combine the three different kinds of mark-ers volume texture and shapeproximity markers to get amore sophisticated picture of the disease progressionOther image modalities such as single-photon emis-
sion computed tomography (SPECT) functional MRI andMR spectroscopy (MRS) positron emission tomography(PET) and molecular imaging have been used for investi-gation of brain changes related to AD SPECT combinedwith MRI images can give additional information aboutdisease progression when combined [49] Functional MRIand MR spectroscopy (MRS) have shown changes inmetabolic levels even prior to symptom onset in ADbut are difficult to implement in clinical settings due totechnical support [5051] PET metabolic imaging withradioactive glucose has also been used to examined thefunctional change and tracking of the AD disease progres-sion [5253] Due to the invasiveness radiation dose limi-tation requiring lumbar punctures and high cost PET isunsuitable for repeated measurements of a single patientor screening programs for large populations Molecularimaging with amyloid tracers have showed great potential
Lillemark et al BMCMedical Imaging 2014 1421 Page 10 of 12httpwwwbiomedcentralcom1471-23421421
as to be accurate markers for early diagnosis of AD but donot show progression in established disease [5455] whichis our object of interestTo conclude structural MRI is an suitable image modal-
ity for detection of AD and AD progression Our mark-ers have shown promising results in capturing how theproximity of different regions in the brain can aid inAD diagnosis and prognosis The proximity analysis cap-tures additional information about the whole brain com-pared to atrophy scores This additional information cancontribute to the refinement of the AD markers andmay be able to give a more detailed picture of ADprogression
Competing interestsThe authors declare that they have no competing interests
Authorsrsquo contributionsLL have contributed in study design data analysis and interpretation preparedand submitted the manuscript LS and AP performed study design and datacollection EBD and MN participated in design and reviewed manuscript Allauthors have read and approved the final manuscript
AcknowledgementsWe gratefully acknowledge the funding from the Danish Research Foundation(Den Danske Forskningsfond) and The Danish National Advanced TechnologyFoundation supporting this work and FreeSurfer for providing the softwareused for the segmentations in this paper Data collection and sharing for thisproject was funded by the Alzheimerrsquos Disease Neuroimaging Initiative (ADNI)(National Institutes of Health Grant U01 AG024904) and DOD ADNI(Department of Defense award number W81XWH-12-2-0012) ADNI is fundedby the National Institute on Aging the National Institute of Bio medicalImaging and Bioengineering and through generous contributions from thefollowing Alzheimerrsquos Association Alzheimerrsquos Drug Discovery FoundationBioClinica Inc Biogen Idec Inc Bristol-Myers Squibb Company Eisai Inc ElanPharmaceuticals Inc Eli Lilly and Company F Hoffmann-La Roche Ltd and itsaffiliated company Genentech Inc GE Healthcare Innogenetics NV IXICOLtd Janssen Alzheimer Immunotherapy Research amp Development LLCJohnson amp Johnson Pharmaceutical Research amp Development LLC MedpaceInc Merck amp Co Inc Meso Scale Diagnostics LLC NeuroRx Research NovartisPharmaceuticals Corporation Pfizer Inc Piramal Imaging Servier Synarc Incand Takeda Pharmaceutical Company The Canadian Institutes of HealthResearch is providing funds to Rev December 5 2013 support ADNI clinicalsites in Canada Private sector contributions are facilitated by the Foundationfor the National Institutes of Health (wwwfnihorg) The grantee organizationis the Northern California Institute for Research and Education and the study iscoordinated by the Alzheimerrsquos Disease Cooperative Study at the University ofCalifornia San Diego ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern CaliforniaData used in preparation of this article were obtained from the AlzheimerrsquosDisease Neuroimaging Initiative (ADNI) database (adniloniuscedu) As suchthe investigators within the ADNI contributed to the design andimplementation of ADNI andor provided data but did not participate inanalysis or writing of this report A complete listing of ADNI investigators canbe found at httpadniloniusceduwp-contentuploadshow_to_applyADNI_Acknowledgement_Listpdf
Author details1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash Denmark 2Biomediq Fruebjergvej3 2100 Copenhagen Oslash Denmark
Received 2 January 2014 Accepted 9 May 2014Published 2 June 2014
References1 Alzheimerrsquos association 2011
[httpwwwalzorgdownloadsFacts_Figures_2011pdf]2 Braskie MN Klunder AD Hayashi KM Protas H Kepe V Miller KJ Huang SC
Barrio JR Ercoli LM Siddarth P Satyamurthy N Liu J Toga AWBookheimer SY Small GW Thompson PM Plaque and tangle imagingand cognition in normal aging and Alzheimerrsquos disease NeurobiolAging 2010 311669ndash1678
3 Braak H Braak E Neuropathological stageing of alzheimer-relatedchanges Acta neuropathologica 1991 82(4)239ndash259
4 West MJ Coleman PD Flood DG Troncoso JC Differences in thepattern of hippocampal neuronal loss in normal ageing andAlzheimerrsquos disease Lancet 1994 344769ndash772
5 Apostolova LG Mosconi L Thompson PM Green AE Hwang KS RamirezA Mistur R Tsui WH de Leon MJ Subregional hippocampal atrophypredicts alzheimerrsquos dementia in the cognitively normal NeurobiolAging 2010 31(7)1077ndash1088
6 Tondelli M Wilcock GK Nichelli P De Jager CA Jenkinson M Zamboni GStructural mri changes detectable up to ten years before clinicalalzheimerrsquos disease Neurobiol Aging 2012 33(4)825ndash25
7 Bernard C Helmer C Dilharreguy B Amieva H Auriacombe S DartiguesJ-F Allard M Catheline G Time course of brain volume changes in thepreclinical phase of alzheimerrsquos disease Alzheimerrsquos Dementia 201410(2)143ndash151
8 Dickerson B Stoub T Shah R Sperling R Killiany R Albert M Hyman BBlacker D deToledo-Morrell L Alzheimer-signature mri biomarkerpredicts ad dementia in cognitively normal adults Neurology 201176(16)1395ndash1402
9 Hansson O Zetterberg H Buchhave P Londos E Blennow K Minthon LAssociation between csf biomarkers and incipient alzheimerrsquosdisease in patients with mild cognitive impairment a follow-upstudy Lancet Neurol 2006 5(3)228ndash234
10 Leung KK Shen K-K Barnes J Ridgway GR Clarkson MJ Fripp JSalvado O Meriaudeau F Fox NC Bourgeat P Ourselin S Increasingpower to predict mild cognitive impairment conversion toalzheimerrsquos disease using hippocampal atrophy rate andstatistical shape models In Proceedings of the 13th InternationalConference onMedical Image Computing and Computer-assistedIntervention Part II MICCAIrsquo10 Berlin Heidelberg Springer2010125ndash132
11 Holland D Dale AM Nonlinear registration of longitudinal imagesandmeasurement of change in regions of interestMed Image Anal2011 15(4)489ndash497
12 Smith SM Zhang Y Jenkinson M Chen J Matthews P Federico ADe Stefano N Accurate robust and automated longitudinal andcross-sectional brain change analysis Neuroimage 200217(1)479ndash489
13 Wang L Swank JS Glick IE Gado MH Miller MI Morris JC Csernansky JGChanges in hippocampal volume and shape across time distinguishdementia of the Alzheimer type from healthy aging Neuroimage2003 20667ndash682
14 Li S Shi F Pu F Li X Jiang T Xie S Wang Y Hippocampal shape analysisof Alzheimer disease based onmachine learning methods AJNR AmJ Neuroradiol 2007 281339ndash1345
15 Costafreda SG Dinov ID Tu Z Shi Y Liu CY Kloszewska I Mecocci PSoininen H Tsolakif M Vellasg B Wahlundh L-O Spengerh C Togab AWLovestonea S Simmonsa A Automated hippocampal shape analysispredicts the onset of dementia in mild cognitive impairmentNeuroImage 2011
16 Scher AI Xu Y Korf ES White LR Scheltens P Toga AW Thompson PMHartley SW Witter MP Valentino DJ Launer LJ Hippocampal shapeanalysis in Alzheimerrsquos disease a population-based studyNeuroimage 2007 368ndash18
17 Klein S Loog M van der Lijn F den Heijer T Hammers A de Bruijne Mvan der Lugt A Duin RPW Breteler MMB Niessen WJ Early diagnosis ofdementia based on intersubject whole-brain dissimilarities InProceedings of the 2010 IEEE International Conference on BiomedicalImaging fromNano toMacro ISBIrsquo10 Piscataway NJ USA IEEE Press2010249ndash252
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 3 of 12httpwwwbiomedcentralcom1471-23421421
MRI acquisitionHigh-Resolution structural brain MRI were acquired at59 ADNI sites using 15 Tesla T1-weighted MRI scansusing volumetric 3D MPRAGE or equivalent protocolswith varying resolution typically 125 times 125 mm in-plane spatial resolution and 12 mm thick sagital slicesThe MPRAGE sequence was acquired twice for all sub-jects at each visit to improve the chance that at least onescan would be suitable for analysis The image quality wasgraded qualitatively by ADNI investigators of the ADNIMRQ quality control center at the Mayo Clinic for arti-facts and general image quality Each scan was graded onseveral separate criteria blurringghosting flow artifactsintensity a homogeneity signal-to-noise ratio susceptibil-ity artifacts and gray-whitecerebrospinal fluid contrastWe have only used the MRI scan which was graded asthe best scan for each subject No other exclusion crite-ria based on image quality were applied We have used theraw ADNI data
ParticipantsThe criteria for inclusion were those defined in theADNI protocol normal control (NC) subjects had amini mental state examamination score (MMSE) between24 - 30 a clinical dementia rating (CDR) score of zerothey were non-depressed non MCI and non-dementedMCI had MMSE scores between 24-30 a memory com-plaint had objective memory loss measured by educationadjusted scores onWechsler Memory Scale Logical Mem-ory II [30] a CDR of 05 absence of significant lev-els of impairment in other cognitive domains AD sub-jects had MMSE scores between 20-26 CDR of 05or 10 and met NINCDSADRDA criteria for proba-ble AD We selected a subset of 528 participants inthe ADNI study We have chosen a training set of101 subjects based on statistics and visual inspectionin order to get representative data which also includedthe difficult images eg with image noise and hugedeformation to allow validation of our methods on ahard data set which makes significant results moreplausibleThe remaining 427 were taken as ADNI-1 data set [31]
minus the overlap with the 101 subjects selected for train-ing Our subset population included 174 NC (age at base-line (bl) 760 years (y) plusmn51 y 89 males (M)85 females(F) 240 MCI (age at bl 749 yplusmn70y 159M81F) and 114AD subjects (age at bl 74 y plusmn73 y 58M56F) Therewas 4 NC 21 MCI and 7 AD subjects in our study thatwas under 65 y Even though there is evidence that thepathology is different in early-onset AD and late-onset ADwe have included the subjects under 65 because they donot have verified early onset AD [32] The demographicdetails of our training and testing subjects are shown inTable 1
Freesurfer segmentationThe segmentation of the regions was performed by staticFreeSurfer [33] implemented on a Linux cluster with 24cores with 18 GB RAM per CPU Freesurfer is a set ofsoftware tools designed to study the cortical and sub-cortical anatomy of the brain Freesurfer do an affineregistration of the volumes with the Talairach atlas [34]a non-uniform intensity normalization (N3) [35] and aB1 bias field correction [36] A skull stripping step wasperformed using a deformable template model Voxelswere then defined as white matter or not white matterbased on intensities Hereafter cutting planes were usedto separate the hemispheres cerebellum and brain stemA cortical and subcortical labeling was performed basedon a transformation that maps the individual subjects intoa probabilistic atlas The atlas was build based on a train-ing set where the subjects have been labeled by hand andcurrently consists of 39 subjects distributed in age and ADpathology (28 NC and 11 with questionable or probableAD) [37] The classification of each point was achieved byfinding the segmentation that maximized the probabilityof input given the prior probability from the training setin a iteratively manner
Grouping of the segmented regionsThe FreeSurfer segmentation provided 40 regions fromwhich a visualization is shown in Figure 1 AD do notspread evenly across the brain and we are interested incapturing early signs of AD and the conversion fromMCIto AD [325] Therefore have we divided our regions intothree groups all functional (func) and potato describedin Table 2 These groups are spread across the brain sowe are not biasing toward anatomical placed groupingsThe all group included the FreeSurfer segmented regionsexcluding left-vessel right vessel and 5th ventricle becausethese regions were not segmented by FreeSurfer in all sub-jects The functional group has excluded all non-functionregions like CSF and hypointensities The choroid plexuswas included in the functional regions due to suggestionsthat the functionality is altered in the choroid plexus dueto AD [38] To get a even smaller subset the potato groupconsisted of small potato shaped regions from a visualperspective where shape is clearly defined
Surface connectivity marker procrustes marker andvolumemarkerWe assume that proximity may reflect aspects of func-tional brain connectivity and have therefore looked atboth the individual regions positional relationship andhow they relate to each other We have calculated thepercentage of how much of a regions own surface wasconnected to the surface of all other regions resultingin a surface connectivity marker This was done non-symmetric in a voxel-count based manner on the three
Lillemark et al BMCMedical Imaging 2014 1421 Page 4 of 12httpwwwbiomedcentralcom1471-23421421
Table 1 The demographic details of our study population
Group Number Age at bl (years) Gender (MF) MMSE at bl
NC training set 24 753 plusmn 44 [651 minus 859] 14 M10 F 293 plusmn 11 [26 minus 30]
MCI training set 29 736 plusmn 73 [552 minus 855] 19 M10 F 272 plusmn 16 [24 minus 30]
AD training set 48 748 plusmn 67 [625 minus 879] 24 M24 F 235 plusmn 19 [21 minus 26]
NC 174 760 plusmn 51 [600 minus 897] 89 M85 F 292 plusmn 10 [25 minus 30]
MCI 240 749 plusmn 70 [552 minus 884] 159 M81 F 271 plusmn 17 [24 minus 30]
AD 114 747 plusmn 73 [565 minus 892] 58 M56 F 233 plusmn 19 [20 minus 26]
MMSE = mini mental state examination score Values are indicated as mean plusmn standard deviation[range
] There is 4 NC 21 MCI and 7 AD subjects in our study that is
under 65
dimensional data so we had a unique image of each regionwhere zero means that there was no connections betweenthe regions and an increasing percentage number referredto how much surface connectivity existed This way wecould observe if shrinkage of regions relates to more fluidin between regions or general shrinkage where the relativesizes did not changeThe individual regions and their internally relationship
was investigated as a change in position of the individualregion We calculated the center of mass c isin R for eachregion P as the mean position of all the points inside theregions in all of the coordinate directions
c middot ed = 12V
Nminus1sumi=0
intAi
(x middot ed)2(ni middot ed) d =123 (1)
where ed denote the standard basis in R by e1 e2 e3 andV denote the volume These points were aligned with aProcrustes alignment where they were adjusted to be in
Figure 1 A slide of the segmented brain where the segmentedregions have different colors
the same space by translation rotation and scaling of thepoints [39] We used the mean shape as the starting shapeThis resulted in a feature vector in a machine learning set-ting that was able to describe the variations in the pointsrelated to the disease statusFor comparison we have used the volume measurement
for the whole brain and for hippocampus for which goodclassification results earlier have been reported [181940]The whole brain volume fraction included all regions inthe skull-stripped brain except for vessels and CSF dividedwith the intracranial volume The hippocampus volumefraction was also measured as the lateral hippocampusvolume divided with the intracranial volume A summaryof our markers is shown in Table 3
Dimensionality reduction and classificationWe wanted to reduce the number of parameters in thecase of Procrustes and surface connectivity due to thecurse of dimensionality where we had more parame-ters than observations We wanted to maintain the rela-tionship between the predictive and target parametersand have therefore chosen to do dimensionality reduc-tion using partial least square regression (PLS) [41] Theidea behind PLS is to find the relevant variables X that
Table 2 The three different groups of the regions allfunctional and potato and the regions belonging to eachgroup
All CSF 3rd-Ventricle 4th-Ventricle Brain-Stem Optic-ChiasmWM-hypointensities non-WM-hypointensities left and rightcerebral white matter cerebral cortex lateral ventricle inflateral ventricle cerebellum white matter cerebellum cortexthalamus caudate putamen pallidum hippocampusamygdala accumbens area ventralDC choroid-plexus
Func Left and right cerebral white matter cerebral cortex inflateral ventricle cerebellum white matter cerebellum cortexthalamus caudate putamen pallidum hippocampusamygdala accumbens area choroid-plexus
Potato Left and right lateral ventricle cerebralwhite matter thalamus caudate putamen pallidumhippocampus amygdala
Lillemark et al BMCMedical Imaging 2014 1421 Page 5 of 12httpwwwbiomedcentralcom1471-23421421
Table 3 An overview of the names and description of themarkers we used in this paper
Marker Description
Procrustes The center of mass of each regions alignedto the same space with a Procrustes alignment
Surface connectivity The percentage of how much each regionhave connected to other regions relatedto the surface of the region
Hippocampus volume The volume of the hippocampus dividedwith the intracranial volume
Whole brain volume The volume of the whole brain dividedwith the intracranial volume
accounts for as much information of the data Y as pos-sible PLS searches for the set of components (latentvariables) that performs a simulation decomposition of Xand Y with the constraint that these components shouldexplain as much as possible of the covariance betweenX and Y It is followed by a linear regression step wherethe decomposition of X is used to predict Y The PLSmodel will try to find the multidimensionality direction inthe X space that explains the maximummultidimensionalvariance direction in the Y space The number of PLScomponents were set to 10 based on our training experi-ments Due to its simple functionality we have used lineardiscriminate analysis (LDA) for the classification [42]LDA tries to reduce the dimensionality while preservingas much of the class discriminatory information as pos-sible LDA seeks to obtain a scalar y by projecting thesamples x onto a line y = wTx where x is the samplesand w contains the class information Of all possible waysto discriminate these we would like to select the one thatmaximizes the separability between the scalars yAll experiments were done in a leave-one-of-each-
class out fashion The data were adjusted for age andgender when there existed a linear correlation betweenthose
ResultThe fractional volume scores for the whole brain volumeand hippocampus volume for NC MCI and AD respec-tively is shown in Table 4 NC had a larger volume inboth whole brain and hippocampus than MCI and ADand MCI had a larger volume score than AD AD had thelargest volume lost between bl and m12For each feature set the area under the curve (AUC) was
computed and summarized in Table 5 for NC versus ADNC versus MCI and MCI versus AD and the correspond-ing ROC curves are shown in Figure 2 The classificationwas tested with a ranksum test and the p-values are alsoshown in Table 5 All markers were able to significantlydiscriminate between the three groups NC-AD NC-MCIand MCI-AD The AUC score were highest for the NC-AD group where our surface connectivity marker werecomparable to the hippocampus volume for the AD-NCand NC-MCI cases and better in the discrimination forthe MCI-AD case than the hippocampus volume TheAUC for the Procrustes marker were in general a littlelower than for the surface connectivity scoreNext we adjusted our markers for whole brain volume
and for hippocampus volume to investigate if our mark-ers contained additional information than the volumesThese results are shown in Table 5 The signal lowersbut was still significant Again the surface connectivitymarkers performed better then the Procrustes markersand the NC-AD classification result were the best Thesurface connectivity markers were generally better to dis-criminate NC-MCI than MCI-AD and for the Procrustesmarkers it was vice versa It was the smaller group-ings functional and potato-shaped that gave the bestperformanceWe have also investigated how our markers performed
on the period to month 12 using the score differencesbetween bl and month 12 for each marker and the AUCand the corresponding ranksum p-values are shown inTable 6 and roc curves in Figure 3 Hippocampus andwhole brain showed relatively low AUC result due to the
Table 4 Fractional volume scores for the hippocampus and the whole brain at bl andmonth 12 and the volume loss
Group Time point Whole brain Hippocampusvolume fraction (cm3) volume fraction (cm3)
The last two markers were divided in three groups all functional and potato-shaped 5(a) is the non-adjusted case 5(b) and 5(c) is adjusted by whole brain fractionand hippocampus fraction respectively All markers were able to significantly distinguish the classes Our markers were still significant after adjustment for the twovolume scores but AUC scores were in general lower than the non-adjusted scores The surface connectivity score for the functional groups performed the best
use of static Freesurfer volumes from bl and month 12Our surface connectivity scores performed the best for allthree groups NC-AD NC-MCI andMCI-AD The resultsbetween NC-AD and NC-MCI are very similarWe have adjusted the month 12 classification results for
both the baseline whole brain and the baseline hippocam-pus volume shown in Table 6 The results showed a sig-nificant classification for our markers When adjusted forwhole brain volume the surface connectivity performedthe best The classification result for MCI-AD case wasbetter than the NC-AD resultFinally we have classifiedMCI-c against MCI-nc where
the non-adjusted result is shown in Table 7 The sur-face connectivity markers was the only marker that wasable to distinguish the two groups and only in the func-tional and potato-shaped grouping of regions When weadjusted for whole brain volume the surface connectiv-ity marker was still significant with an AUC at 0631
(p = 0012) and for the potato group it was borderlinesignificant with an AUC at 0595 (p = 0067) In thecase where we adjusted for hippocampus volume onlythe surface connectivity marker for the functional groupswas borderline significant with an AUC of 0599 (p =0055) No other significance were shown in the adjustedcases
Discussion and conclusionWe have investigated a novel way of looking at the rela-tionship between different regions in the brain We eval-uated a surface connectivity marker and center of massbased marker and their ability to classify between NCMCI and AD subjects Both markers have been able tosignificantly discriminate between the three classes AD-NC NC-MCI andMCI-AD both at baseline and betweenbaseline and month 12 Our surface connectivity markerwas also able to classify MCI-c
Lillemark et al BMCMedical Imaging 2014 1421 Page 7 of 12httpwwwbiomedcentralcom1471-23421421
Figure 2 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
The large variabilityrsquos in the brain regions is related toAlzheimerrsquos Disease [171925-27] and this have moti-vated our two markers describing the proximity betweenthe regions in the brain Both our markers were ableto significantly differentiate between AD and NC alsowhen adjusted for whole brain and hippocampus volumeThe surface connectivity marker was comparable to hip-pocampus volume which is one of known most effect fullmarkers from MRI Also after adjustment for volumes wehad a significant classification results this indicates thatour markers hold additional information about the devel-opment of the brain in relation to progression of ADWe believe that our markers capture an individual shrink-age due to pathological alterations In subjects with ADthe cerebral cortex is shrinking the sulcirsquos is widenedthe cortical ribbon may be thinned and ventricles aredilated [24344] Our surface connectivity markers maycapture some of these pathological alterations in measur-ing the proximity between regionsWe have evaluated our markers over a 1 - year period
where we have investigated the change in the Procrustesaligned positions and the change in surface connectivityIn this case we were also able to significantly discrim-inate between the classes although the signal was lessstrong The weakened signal can be due to noise in thesegmentation of the data Our markers were not taken
from registered brains but normalized within the samebrain so they captured comparable information acrosstime and study population The segmentation of the indi-vidual regions at two time steps can still be quite differ-ent and when we were using the difference between thescore values it can introduce noise in our markers Thisis also visible in the values for hippocampus and wholebrain volume in the longitudinal part of our study whichshowed lower results for classification than other reportedresults [1745]Our surface connectivity marker performed the best
indicating that it captured how the cell death caused byAD minimizes the surface connectivity between regionsThis was most visible in the functional regions The func-tional group were limited to functional regions of thebrain and the good performance of this grouping is in linewith the knowledge that AD affect the network aroundand including the medial temporal lobe and disruption inthis region contributes to memory impairment [46] Thelower performance of our Procrustes marker could be dueto the captured information is closer to volume and thatno particular regions moves related to the others but allregions moved due to general volume lossCuingnet et al [18] have made a comparison study
for classification of NC versus AD NC versus MCI-converters (MCI-c) and MCI-c versus MCI-non-
Lillemark et al BMCMedical Imaging 2014 1421 Page 8 of 12httpwwwbiomedcentralcom1471-23421421
Table 6 Classification result for NC-AD NC-MCI andMCI-AD for the difference between the bl andmonth 12makers 6(a)is the not adjusted case 6(b) is adjusted for bl whole brain volume and 6(c) is adjusted for baseline hippocampus volume
Our markers was still able to significantly discriminate between the three groups Our surface connectivity markers for the two subgroups functional and potatoperformed the best
converters (MCI-nc) based on 81 NC 67 MCI-nc 39MCI-c and 69 AD subjects from the ADNI databaseThey investigated voxel based segmented tissue regionsfor the whole brain in six different variants and for graymatter (GM) and GM white matter (WM) and cere-brospinal fluid (CSF) combined cortical thickness inthree different variants and finally hippocampus volumeand shape in three different variants a total of ten differ-ent methods They conclude that all methods were able toclassify NC vs AD with a sensitivity and specificity at therange from 59 - 81 and 77 - 98 respectively whichis comparable to our classification Other prediction stud-ies have shown better classification rates at 67 - 92 forcross-sectional studies [1417194547] and 69 - 815for longitudinal studies [19-21] The difference in theclassification accuracy between our method and the otherpapers can be explained by the tuning of methods and theuse of different data sets
Only our surface connectivity marker was able to clas-sify MCI-c fromMCI-nc and not with a highly significantresult This is in line with Cuignet et al comparison studyfor AD classification where they found that only fourmethods managed to predict MCI-c vs MCI-nc betterthan a random classifier and none of those got signifi-cantly better results [18] The main reason for the lowresult in the conversion case could be due to the fact thatMCI is a very in heterogeneous group that possibly couldconvert rapidly to AD or be stable for many years beforeconversionOther studies have investigated the change locally in
the hippocampus Wang et al [13] have used large-deformation diffeomorphic high-dimensional brain map-ping to quantify and compare changes in the hippocampalshape as well as volume They found that shape changeswere largely confined to the head of hippocampus andsubiculum for normal controls (NC) Other studies have
Lillemark et al BMCMedical Imaging 2014 1421 Page 9 of 12httpwwwbiomedcentralcom1471-23421421
Figure 3 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
confirmed these shape changes for the hippocampus[14-16] based on shape models and local hippocampalatrophy patterns We have focused on investigating therelationship between the different regions of the brain andhow they differ between healthy subjects and AD patientsThis way of investigating the regions could make it pos-sible to incorporate different kind of knowledge into thesame model where one could go from the individual scaleof each region to the interaction between the regionsand finally to combined picture of the brain as one wholeregion
Table 7 The AUC and corresponding p-values for theclassification of MCI-c andMCI-nc
Markers AUC pminusvalue
HPICV 0466 0516
WBICV 0512 0823
Surface all 0542 0416
Surface func 0624 0017
Surface potato 0603 0048
Procrustes all 0465 0486
Procrustes func 0498 0964
Procrustes potato 0534 0501
Only the surface connectivity markers was able to significantly discriminate thetwo groups functional and potato-shaped
An alternative use of MRI images for early predictionof AD is by using texture analysis where different texturesfeatures is used to construct a computational frameworkwhich have been able to discriminate AD MCI and NCwith a separability of up to 95 [234048] This indicatesthat one can combine the three different kinds of mark-ers volume texture and shapeproximity markers to get amore sophisticated picture of the disease progressionOther image modalities such as single-photon emis-
sion computed tomography (SPECT) functional MRI andMR spectroscopy (MRS) positron emission tomography(PET) and molecular imaging have been used for investi-gation of brain changes related to AD SPECT combinedwith MRI images can give additional information aboutdisease progression when combined [49] Functional MRIand MR spectroscopy (MRS) have shown changes inmetabolic levels even prior to symptom onset in ADbut are difficult to implement in clinical settings due totechnical support [5051] PET metabolic imaging withradioactive glucose has also been used to examined thefunctional change and tracking of the AD disease progres-sion [5253] Due to the invasiveness radiation dose limi-tation requiring lumbar punctures and high cost PET isunsuitable for repeated measurements of a single patientor screening programs for large populations Molecularimaging with amyloid tracers have showed great potential
Lillemark et al BMCMedical Imaging 2014 1421 Page 10 of 12httpwwwbiomedcentralcom1471-23421421
as to be accurate markers for early diagnosis of AD but donot show progression in established disease [5455] whichis our object of interestTo conclude structural MRI is an suitable image modal-
ity for detection of AD and AD progression Our mark-ers have shown promising results in capturing how theproximity of different regions in the brain can aid inAD diagnosis and prognosis The proximity analysis cap-tures additional information about the whole brain com-pared to atrophy scores This additional information cancontribute to the refinement of the AD markers andmay be able to give a more detailed picture of ADprogression
Competing interestsThe authors declare that they have no competing interests
Authorsrsquo contributionsLL have contributed in study design data analysis and interpretation preparedand submitted the manuscript LS and AP performed study design and datacollection EBD and MN participated in design and reviewed manuscript Allauthors have read and approved the final manuscript
AcknowledgementsWe gratefully acknowledge the funding from the Danish Research Foundation(Den Danske Forskningsfond) and The Danish National Advanced TechnologyFoundation supporting this work and FreeSurfer for providing the softwareused for the segmentations in this paper Data collection and sharing for thisproject was funded by the Alzheimerrsquos Disease Neuroimaging Initiative (ADNI)(National Institutes of Health Grant U01 AG024904) and DOD ADNI(Department of Defense award number W81XWH-12-2-0012) ADNI is fundedby the National Institute on Aging the National Institute of Bio medicalImaging and Bioengineering and through generous contributions from thefollowing Alzheimerrsquos Association Alzheimerrsquos Drug Discovery FoundationBioClinica Inc Biogen Idec Inc Bristol-Myers Squibb Company Eisai Inc ElanPharmaceuticals Inc Eli Lilly and Company F Hoffmann-La Roche Ltd and itsaffiliated company Genentech Inc GE Healthcare Innogenetics NV IXICOLtd Janssen Alzheimer Immunotherapy Research amp Development LLCJohnson amp Johnson Pharmaceutical Research amp Development LLC MedpaceInc Merck amp Co Inc Meso Scale Diagnostics LLC NeuroRx Research NovartisPharmaceuticals Corporation Pfizer Inc Piramal Imaging Servier Synarc Incand Takeda Pharmaceutical Company The Canadian Institutes of HealthResearch is providing funds to Rev December 5 2013 support ADNI clinicalsites in Canada Private sector contributions are facilitated by the Foundationfor the National Institutes of Health (wwwfnihorg) The grantee organizationis the Northern California Institute for Research and Education and the study iscoordinated by the Alzheimerrsquos Disease Cooperative Study at the University ofCalifornia San Diego ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern CaliforniaData used in preparation of this article were obtained from the AlzheimerrsquosDisease Neuroimaging Initiative (ADNI) database (adniloniuscedu) As suchthe investigators within the ADNI contributed to the design andimplementation of ADNI andor provided data but did not participate inanalysis or writing of this report A complete listing of ADNI investigators canbe found at httpadniloniusceduwp-contentuploadshow_to_applyADNI_Acknowledgement_Listpdf
Author details1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash Denmark 2Biomediq Fruebjergvej3 2100 Copenhagen Oslash Denmark
Received 2 January 2014 Accepted 9 May 2014Published 2 June 2014
References1 Alzheimerrsquos association 2011
[httpwwwalzorgdownloadsFacts_Figures_2011pdf]2 Braskie MN Klunder AD Hayashi KM Protas H Kepe V Miller KJ Huang SC
Barrio JR Ercoli LM Siddarth P Satyamurthy N Liu J Toga AWBookheimer SY Small GW Thompson PM Plaque and tangle imagingand cognition in normal aging and Alzheimerrsquos disease NeurobiolAging 2010 311669ndash1678
3 Braak H Braak E Neuropathological stageing of alzheimer-relatedchanges Acta neuropathologica 1991 82(4)239ndash259
4 West MJ Coleman PD Flood DG Troncoso JC Differences in thepattern of hippocampal neuronal loss in normal ageing andAlzheimerrsquos disease Lancet 1994 344769ndash772
5 Apostolova LG Mosconi L Thompson PM Green AE Hwang KS RamirezA Mistur R Tsui WH de Leon MJ Subregional hippocampal atrophypredicts alzheimerrsquos dementia in the cognitively normal NeurobiolAging 2010 31(7)1077ndash1088
6 Tondelli M Wilcock GK Nichelli P De Jager CA Jenkinson M Zamboni GStructural mri changes detectable up to ten years before clinicalalzheimerrsquos disease Neurobiol Aging 2012 33(4)825ndash25
7 Bernard C Helmer C Dilharreguy B Amieva H Auriacombe S DartiguesJ-F Allard M Catheline G Time course of brain volume changes in thepreclinical phase of alzheimerrsquos disease Alzheimerrsquos Dementia 201410(2)143ndash151
8 Dickerson B Stoub T Shah R Sperling R Killiany R Albert M Hyman BBlacker D deToledo-Morrell L Alzheimer-signature mri biomarkerpredicts ad dementia in cognitively normal adults Neurology 201176(16)1395ndash1402
9 Hansson O Zetterberg H Buchhave P Londos E Blennow K Minthon LAssociation between csf biomarkers and incipient alzheimerrsquosdisease in patients with mild cognitive impairment a follow-upstudy Lancet Neurol 2006 5(3)228ndash234
10 Leung KK Shen K-K Barnes J Ridgway GR Clarkson MJ Fripp JSalvado O Meriaudeau F Fox NC Bourgeat P Ourselin S Increasingpower to predict mild cognitive impairment conversion toalzheimerrsquos disease using hippocampal atrophy rate andstatistical shape models In Proceedings of the 13th InternationalConference onMedical Image Computing and Computer-assistedIntervention Part II MICCAIrsquo10 Berlin Heidelberg Springer2010125ndash132
11 Holland D Dale AM Nonlinear registration of longitudinal imagesandmeasurement of change in regions of interestMed Image Anal2011 15(4)489ndash497
12 Smith SM Zhang Y Jenkinson M Chen J Matthews P Federico ADe Stefano N Accurate robust and automated longitudinal andcross-sectional brain change analysis Neuroimage 200217(1)479ndash489
13 Wang L Swank JS Glick IE Gado MH Miller MI Morris JC Csernansky JGChanges in hippocampal volume and shape across time distinguishdementia of the Alzheimer type from healthy aging Neuroimage2003 20667ndash682
14 Li S Shi F Pu F Li X Jiang T Xie S Wang Y Hippocampal shape analysisof Alzheimer disease based onmachine learning methods AJNR AmJ Neuroradiol 2007 281339ndash1345
15 Costafreda SG Dinov ID Tu Z Shi Y Liu CY Kloszewska I Mecocci PSoininen H Tsolakif M Vellasg B Wahlundh L-O Spengerh C Togab AWLovestonea S Simmonsa A Automated hippocampal shape analysispredicts the onset of dementia in mild cognitive impairmentNeuroImage 2011
16 Scher AI Xu Y Korf ES White LR Scheltens P Toga AW Thompson PMHartley SW Witter MP Valentino DJ Launer LJ Hippocampal shapeanalysis in Alzheimerrsquos disease a population-based studyNeuroimage 2007 368ndash18
17 Klein S Loog M van der Lijn F den Heijer T Hammers A de Bruijne Mvan der Lugt A Duin RPW Breteler MMB Niessen WJ Early diagnosis ofdementia based on intersubject whole-brain dissimilarities InProceedings of the 2010 IEEE International Conference on BiomedicalImaging fromNano toMacro ISBIrsquo10 Piscataway NJ USA IEEE Press2010249ndash252
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 4 of 12httpwwwbiomedcentralcom1471-23421421
Table 1 The demographic details of our study population
Group Number Age at bl (years) Gender (MF) MMSE at bl
NC training set 24 753 plusmn 44 [651 minus 859] 14 M10 F 293 plusmn 11 [26 minus 30]
MCI training set 29 736 plusmn 73 [552 minus 855] 19 M10 F 272 plusmn 16 [24 minus 30]
AD training set 48 748 plusmn 67 [625 minus 879] 24 M24 F 235 plusmn 19 [21 minus 26]
NC 174 760 plusmn 51 [600 minus 897] 89 M85 F 292 plusmn 10 [25 minus 30]
MCI 240 749 plusmn 70 [552 minus 884] 159 M81 F 271 plusmn 17 [24 minus 30]
AD 114 747 plusmn 73 [565 minus 892] 58 M56 F 233 plusmn 19 [20 minus 26]
MMSE = mini mental state examination score Values are indicated as mean plusmn standard deviation[range
] There is 4 NC 21 MCI and 7 AD subjects in our study that is
under 65
dimensional data so we had a unique image of each regionwhere zero means that there was no connections betweenthe regions and an increasing percentage number referredto how much surface connectivity existed This way wecould observe if shrinkage of regions relates to more fluidin between regions or general shrinkage where the relativesizes did not changeThe individual regions and their internally relationship
was investigated as a change in position of the individualregion We calculated the center of mass c isin R for eachregion P as the mean position of all the points inside theregions in all of the coordinate directions
c middot ed = 12V
Nminus1sumi=0
intAi
(x middot ed)2(ni middot ed) d =123 (1)
where ed denote the standard basis in R by e1 e2 e3 andV denote the volume These points were aligned with aProcrustes alignment where they were adjusted to be in
Figure 1 A slide of the segmented brain where the segmentedregions have different colors
the same space by translation rotation and scaling of thepoints [39] We used the mean shape as the starting shapeThis resulted in a feature vector in a machine learning set-ting that was able to describe the variations in the pointsrelated to the disease statusFor comparison we have used the volume measurement
for the whole brain and for hippocampus for which goodclassification results earlier have been reported [181940]The whole brain volume fraction included all regions inthe skull-stripped brain except for vessels and CSF dividedwith the intracranial volume The hippocampus volumefraction was also measured as the lateral hippocampusvolume divided with the intracranial volume A summaryof our markers is shown in Table 3
Dimensionality reduction and classificationWe wanted to reduce the number of parameters in thecase of Procrustes and surface connectivity due to thecurse of dimensionality where we had more parame-ters than observations We wanted to maintain the rela-tionship between the predictive and target parametersand have therefore chosen to do dimensionality reduc-tion using partial least square regression (PLS) [41] Theidea behind PLS is to find the relevant variables X that
Table 2 The three different groups of the regions allfunctional and potato and the regions belonging to eachgroup
All CSF 3rd-Ventricle 4th-Ventricle Brain-Stem Optic-ChiasmWM-hypointensities non-WM-hypointensities left and rightcerebral white matter cerebral cortex lateral ventricle inflateral ventricle cerebellum white matter cerebellum cortexthalamus caudate putamen pallidum hippocampusamygdala accumbens area ventralDC choroid-plexus
Func Left and right cerebral white matter cerebral cortex inflateral ventricle cerebellum white matter cerebellum cortexthalamus caudate putamen pallidum hippocampusamygdala accumbens area choroid-plexus
Potato Left and right lateral ventricle cerebralwhite matter thalamus caudate putamen pallidumhippocampus amygdala
Lillemark et al BMCMedical Imaging 2014 1421 Page 5 of 12httpwwwbiomedcentralcom1471-23421421
Table 3 An overview of the names and description of themarkers we used in this paper
Marker Description
Procrustes The center of mass of each regions alignedto the same space with a Procrustes alignment
Surface connectivity The percentage of how much each regionhave connected to other regions relatedto the surface of the region
Hippocampus volume The volume of the hippocampus dividedwith the intracranial volume
Whole brain volume The volume of the whole brain dividedwith the intracranial volume
accounts for as much information of the data Y as pos-sible PLS searches for the set of components (latentvariables) that performs a simulation decomposition of Xand Y with the constraint that these components shouldexplain as much as possible of the covariance betweenX and Y It is followed by a linear regression step wherethe decomposition of X is used to predict Y The PLSmodel will try to find the multidimensionality direction inthe X space that explains the maximummultidimensionalvariance direction in the Y space The number of PLScomponents were set to 10 based on our training experi-ments Due to its simple functionality we have used lineardiscriminate analysis (LDA) for the classification [42]LDA tries to reduce the dimensionality while preservingas much of the class discriminatory information as pos-sible LDA seeks to obtain a scalar y by projecting thesamples x onto a line y = wTx where x is the samplesand w contains the class information Of all possible waysto discriminate these we would like to select the one thatmaximizes the separability between the scalars yAll experiments were done in a leave-one-of-each-
class out fashion The data were adjusted for age andgender when there existed a linear correlation betweenthose
ResultThe fractional volume scores for the whole brain volumeand hippocampus volume for NC MCI and AD respec-tively is shown in Table 4 NC had a larger volume inboth whole brain and hippocampus than MCI and ADand MCI had a larger volume score than AD AD had thelargest volume lost between bl and m12For each feature set the area under the curve (AUC) was
computed and summarized in Table 5 for NC versus ADNC versus MCI and MCI versus AD and the correspond-ing ROC curves are shown in Figure 2 The classificationwas tested with a ranksum test and the p-values are alsoshown in Table 5 All markers were able to significantlydiscriminate between the three groups NC-AD NC-MCIand MCI-AD The AUC score were highest for the NC-AD group where our surface connectivity marker werecomparable to the hippocampus volume for the AD-NCand NC-MCI cases and better in the discrimination forthe MCI-AD case than the hippocampus volume TheAUC for the Procrustes marker were in general a littlelower than for the surface connectivity scoreNext we adjusted our markers for whole brain volume
and for hippocampus volume to investigate if our mark-ers contained additional information than the volumesThese results are shown in Table 5 The signal lowersbut was still significant Again the surface connectivitymarkers performed better then the Procrustes markersand the NC-AD classification result were the best Thesurface connectivity markers were generally better to dis-criminate NC-MCI than MCI-AD and for the Procrustesmarkers it was vice versa It was the smaller group-ings functional and potato-shaped that gave the bestperformanceWe have also investigated how our markers performed
on the period to month 12 using the score differencesbetween bl and month 12 for each marker and the AUCand the corresponding ranksum p-values are shown inTable 6 and roc curves in Figure 3 Hippocampus andwhole brain showed relatively low AUC result due to the
Table 4 Fractional volume scores for the hippocampus and the whole brain at bl andmonth 12 and the volume loss
Group Time point Whole brain Hippocampusvolume fraction (cm3) volume fraction (cm3)
The last two markers were divided in three groups all functional and potato-shaped 5(a) is the non-adjusted case 5(b) and 5(c) is adjusted by whole brain fractionand hippocampus fraction respectively All markers were able to significantly distinguish the classes Our markers were still significant after adjustment for the twovolume scores but AUC scores were in general lower than the non-adjusted scores The surface connectivity score for the functional groups performed the best
use of static Freesurfer volumes from bl and month 12Our surface connectivity scores performed the best for allthree groups NC-AD NC-MCI andMCI-AD The resultsbetween NC-AD and NC-MCI are very similarWe have adjusted the month 12 classification results for
both the baseline whole brain and the baseline hippocam-pus volume shown in Table 6 The results showed a sig-nificant classification for our markers When adjusted forwhole brain volume the surface connectivity performedthe best The classification result for MCI-AD case wasbetter than the NC-AD resultFinally we have classifiedMCI-c against MCI-nc where
the non-adjusted result is shown in Table 7 The sur-face connectivity markers was the only marker that wasable to distinguish the two groups and only in the func-tional and potato-shaped grouping of regions When weadjusted for whole brain volume the surface connectiv-ity marker was still significant with an AUC at 0631
(p = 0012) and for the potato group it was borderlinesignificant with an AUC at 0595 (p = 0067) In thecase where we adjusted for hippocampus volume onlythe surface connectivity marker for the functional groupswas borderline significant with an AUC of 0599 (p =0055) No other significance were shown in the adjustedcases
Discussion and conclusionWe have investigated a novel way of looking at the rela-tionship between different regions in the brain We eval-uated a surface connectivity marker and center of massbased marker and their ability to classify between NCMCI and AD subjects Both markers have been able tosignificantly discriminate between the three classes AD-NC NC-MCI andMCI-AD both at baseline and betweenbaseline and month 12 Our surface connectivity markerwas also able to classify MCI-c
Lillemark et al BMCMedical Imaging 2014 1421 Page 7 of 12httpwwwbiomedcentralcom1471-23421421
Figure 2 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
The large variabilityrsquos in the brain regions is related toAlzheimerrsquos Disease [171925-27] and this have moti-vated our two markers describing the proximity betweenthe regions in the brain Both our markers were ableto significantly differentiate between AD and NC alsowhen adjusted for whole brain and hippocampus volumeThe surface connectivity marker was comparable to hip-pocampus volume which is one of known most effect fullmarkers from MRI Also after adjustment for volumes wehad a significant classification results this indicates thatour markers hold additional information about the devel-opment of the brain in relation to progression of ADWe believe that our markers capture an individual shrink-age due to pathological alterations In subjects with ADthe cerebral cortex is shrinking the sulcirsquos is widenedthe cortical ribbon may be thinned and ventricles aredilated [24344] Our surface connectivity markers maycapture some of these pathological alterations in measur-ing the proximity between regionsWe have evaluated our markers over a 1 - year period
where we have investigated the change in the Procrustesaligned positions and the change in surface connectivityIn this case we were also able to significantly discrim-inate between the classes although the signal was lessstrong The weakened signal can be due to noise in thesegmentation of the data Our markers were not taken
from registered brains but normalized within the samebrain so they captured comparable information acrosstime and study population The segmentation of the indi-vidual regions at two time steps can still be quite differ-ent and when we were using the difference between thescore values it can introduce noise in our markers Thisis also visible in the values for hippocampus and wholebrain volume in the longitudinal part of our study whichshowed lower results for classification than other reportedresults [1745]Our surface connectivity marker performed the best
indicating that it captured how the cell death caused byAD minimizes the surface connectivity between regionsThis was most visible in the functional regions The func-tional group were limited to functional regions of thebrain and the good performance of this grouping is in linewith the knowledge that AD affect the network aroundand including the medial temporal lobe and disruption inthis region contributes to memory impairment [46] Thelower performance of our Procrustes marker could be dueto the captured information is closer to volume and thatno particular regions moves related to the others but allregions moved due to general volume lossCuingnet et al [18] have made a comparison study
for classification of NC versus AD NC versus MCI-converters (MCI-c) and MCI-c versus MCI-non-
Lillemark et al BMCMedical Imaging 2014 1421 Page 8 of 12httpwwwbiomedcentralcom1471-23421421
Table 6 Classification result for NC-AD NC-MCI andMCI-AD for the difference between the bl andmonth 12makers 6(a)is the not adjusted case 6(b) is adjusted for bl whole brain volume and 6(c) is adjusted for baseline hippocampus volume
Our markers was still able to significantly discriminate between the three groups Our surface connectivity markers for the two subgroups functional and potatoperformed the best
converters (MCI-nc) based on 81 NC 67 MCI-nc 39MCI-c and 69 AD subjects from the ADNI databaseThey investigated voxel based segmented tissue regionsfor the whole brain in six different variants and for graymatter (GM) and GM white matter (WM) and cere-brospinal fluid (CSF) combined cortical thickness inthree different variants and finally hippocampus volumeand shape in three different variants a total of ten differ-ent methods They conclude that all methods were able toclassify NC vs AD with a sensitivity and specificity at therange from 59 - 81 and 77 - 98 respectively whichis comparable to our classification Other prediction stud-ies have shown better classification rates at 67 - 92 forcross-sectional studies [1417194547] and 69 - 815for longitudinal studies [19-21] The difference in theclassification accuracy between our method and the otherpapers can be explained by the tuning of methods and theuse of different data sets
Only our surface connectivity marker was able to clas-sify MCI-c fromMCI-nc and not with a highly significantresult This is in line with Cuignet et al comparison studyfor AD classification where they found that only fourmethods managed to predict MCI-c vs MCI-nc betterthan a random classifier and none of those got signifi-cantly better results [18] The main reason for the lowresult in the conversion case could be due to the fact thatMCI is a very in heterogeneous group that possibly couldconvert rapidly to AD or be stable for many years beforeconversionOther studies have investigated the change locally in
the hippocampus Wang et al [13] have used large-deformation diffeomorphic high-dimensional brain map-ping to quantify and compare changes in the hippocampalshape as well as volume They found that shape changeswere largely confined to the head of hippocampus andsubiculum for normal controls (NC) Other studies have
Lillemark et al BMCMedical Imaging 2014 1421 Page 9 of 12httpwwwbiomedcentralcom1471-23421421
Figure 3 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
confirmed these shape changes for the hippocampus[14-16] based on shape models and local hippocampalatrophy patterns We have focused on investigating therelationship between the different regions of the brain andhow they differ between healthy subjects and AD patientsThis way of investigating the regions could make it pos-sible to incorporate different kind of knowledge into thesame model where one could go from the individual scaleof each region to the interaction between the regionsand finally to combined picture of the brain as one wholeregion
Table 7 The AUC and corresponding p-values for theclassification of MCI-c andMCI-nc
Markers AUC pminusvalue
HPICV 0466 0516
WBICV 0512 0823
Surface all 0542 0416
Surface func 0624 0017
Surface potato 0603 0048
Procrustes all 0465 0486
Procrustes func 0498 0964
Procrustes potato 0534 0501
Only the surface connectivity markers was able to significantly discriminate thetwo groups functional and potato-shaped
An alternative use of MRI images for early predictionof AD is by using texture analysis where different texturesfeatures is used to construct a computational frameworkwhich have been able to discriminate AD MCI and NCwith a separability of up to 95 [234048] This indicatesthat one can combine the three different kinds of mark-ers volume texture and shapeproximity markers to get amore sophisticated picture of the disease progressionOther image modalities such as single-photon emis-
sion computed tomography (SPECT) functional MRI andMR spectroscopy (MRS) positron emission tomography(PET) and molecular imaging have been used for investi-gation of brain changes related to AD SPECT combinedwith MRI images can give additional information aboutdisease progression when combined [49] Functional MRIand MR spectroscopy (MRS) have shown changes inmetabolic levels even prior to symptom onset in ADbut are difficult to implement in clinical settings due totechnical support [5051] PET metabolic imaging withradioactive glucose has also been used to examined thefunctional change and tracking of the AD disease progres-sion [5253] Due to the invasiveness radiation dose limi-tation requiring lumbar punctures and high cost PET isunsuitable for repeated measurements of a single patientor screening programs for large populations Molecularimaging with amyloid tracers have showed great potential
Lillemark et al BMCMedical Imaging 2014 1421 Page 10 of 12httpwwwbiomedcentralcom1471-23421421
as to be accurate markers for early diagnosis of AD but donot show progression in established disease [5455] whichis our object of interestTo conclude structural MRI is an suitable image modal-
ity for detection of AD and AD progression Our mark-ers have shown promising results in capturing how theproximity of different regions in the brain can aid inAD diagnosis and prognosis The proximity analysis cap-tures additional information about the whole brain com-pared to atrophy scores This additional information cancontribute to the refinement of the AD markers andmay be able to give a more detailed picture of ADprogression
Competing interestsThe authors declare that they have no competing interests
Authorsrsquo contributionsLL have contributed in study design data analysis and interpretation preparedand submitted the manuscript LS and AP performed study design and datacollection EBD and MN participated in design and reviewed manuscript Allauthors have read and approved the final manuscript
AcknowledgementsWe gratefully acknowledge the funding from the Danish Research Foundation(Den Danske Forskningsfond) and The Danish National Advanced TechnologyFoundation supporting this work and FreeSurfer for providing the softwareused for the segmentations in this paper Data collection and sharing for thisproject was funded by the Alzheimerrsquos Disease Neuroimaging Initiative (ADNI)(National Institutes of Health Grant U01 AG024904) and DOD ADNI(Department of Defense award number W81XWH-12-2-0012) ADNI is fundedby the National Institute on Aging the National Institute of Bio medicalImaging and Bioengineering and through generous contributions from thefollowing Alzheimerrsquos Association Alzheimerrsquos Drug Discovery FoundationBioClinica Inc Biogen Idec Inc Bristol-Myers Squibb Company Eisai Inc ElanPharmaceuticals Inc Eli Lilly and Company F Hoffmann-La Roche Ltd and itsaffiliated company Genentech Inc GE Healthcare Innogenetics NV IXICOLtd Janssen Alzheimer Immunotherapy Research amp Development LLCJohnson amp Johnson Pharmaceutical Research amp Development LLC MedpaceInc Merck amp Co Inc Meso Scale Diagnostics LLC NeuroRx Research NovartisPharmaceuticals Corporation Pfizer Inc Piramal Imaging Servier Synarc Incand Takeda Pharmaceutical Company The Canadian Institutes of HealthResearch is providing funds to Rev December 5 2013 support ADNI clinicalsites in Canada Private sector contributions are facilitated by the Foundationfor the National Institutes of Health (wwwfnihorg) The grantee organizationis the Northern California Institute for Research and Education and the study iscoordinated by the Alzheimerrsquos Disease Cooperative Study at the University ofCalifornia San Diego ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern CaliforniaData used in preparation of this article were obtained from the AlzheimerrsquosDisease Neuroimaging Initiative (ADNI) database (adniloniuscedu) As suchthe investigators within the ADNI contributed to the design andimplementation of ADNI andor provided data but did not participate inanalysis or writing of this report A complete listing of ADNI investigators canbe found at httpadniloniusceduwp-contentuploadshow_to_applyADNI_Acknowledgement_Listpdf
Author details1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash Denmark 2Biomediq Fruebjergvej3 2100 Copenhagen Oslash Denmark
Received 2 January 2014 Accepted 9 May 2014Published 2 June 2014
References1 Alzheimerrsquos association 2011
[httpwwwalzorgdownloadsFacts_Figures_2011pdf]2 Braskie MN Klunder AD Hayashi KM Protas H Kepe V Miller KJ Huang SC
Barrio JR Ercoli LM Siddarth P Satyamurthy N Liu J Toga AWBookheimer SY Small GW Thompson PM Plaque and tangle imagingand cognition in normal aging and Alzheimerrsquos disease NeurobiolAging 2010 311669ndash1678
3 Braak H Braak E Neuropathological stageing of alzheimer-relatedchanges Acta neuropathologica 1991 82(4)239ndash259
4 West MJ Coleman PD Flood DG Troncoso JC Differences in thepattern of hippocampal neuronal loss in normal ageing andAlzheimerrsquos disease Lancet 1994 344769ndash772
5 Apostolova LG Mosconi L Thompson PM Green AE Hwang KS RamirezA Mistur R Tsui WH de Leon MJ Subregional hippocampal atrophypredicts alzheimerrsquos dementia in the cognitively normal NeurobiolAging 2010 31(7)1077ndash1088
6 Tondelli M Wilcock GK Nichelli P De Jager CA Jenkinson M Zamboni GStructural mri changes detectable up to ten years before clinicalalzheimerrsquos disease Neurobiol Aging 2012 33(4)825ndash25
7 Bernard C Helmer C Dilharreguy B Amieva H Auriacombe S DartiguesJ-F Allard M Catheline G Time course of brain volume changes in thepreclinical phase of alzheimerrsquos disease Alzheimerrsquos Dementia 201410(2)143ndash151
8 Dickerson B Stoub T Shah R Sperling R Killiany R Albert M Hyman BBlacker D deToledo-Morrell L Alzheimer-signature mri biomarkerpredicts ad dementia in cognitively normal adults Neurology 201176(16)1395ndash1402
9 Hansson O Zetterberg H Buchhave P Londos E Blennow K Minthon LAssociation between csf biomarkers and incipient alzheimerrsquosdisease in patients with mild cognitive impairment a follow-upstudy Lancet Neurol 2006 5(3)228ndash234
10 Leung KK Shen K-K Barnes J Ridgway GR Clarkson MJ Fripp JSalvado O Meriaudeau F Fox NC Bourgeat P Ourselin S Increasingpower to predict mild cognitive impairment conversion toalzheimerrsquos disease using hippocampal atrophy rate andstatistical shape models In Proceedings of the 13th InternationalConference onMedical Image Computing and Computer-assistedIntervention Part II MICCAIrsquo10 Berlin Heidelberg Springer2010125ndash132
11 Holland D Dale AM Nonlinear registration of longitudinal imagesandmeasurement of change in regions of interestMed Image Anal2011 15(4)489ndash497
12 Smith SM Zhang Y Jenkinson M Chen J Matthews P Federico ADe Stefano N Accurate robust and automated longitudinal andcross-sectional brain change analysis Neuroimage 200217(1)479ndash489
13 Wang L Swank JS Glick IE Gado MH Miller MI Morris JC Csernansky JGChanges in hippocampal volume and shape across time distinguishdementia of the Alzheimer type from healthy aging Neuroimage2003 20667ndash682
14 Li S Shi F Pu F Li X Jiang T Xie S Wang Y Hippocampal shape analysisof Alzheimer disease based onmachine learning methods AJNR AmJ Neuroradiol 2007 281339ndash1345
15 Costafreda SG Dinov ID Tu Z Shi Y Liu CY Kloszewska I Mecocci PSoininen H Tsolakif M Vellasg B Wahlundh L-O Spengerh C Togab AWLovestonea S Simmonsa A Automated hippocampal shape analysispredicts the onset of dementia in mild cognitive impairmentNeuroImage 2011
16 Scher AI Xu Y Korf ES White LR Scheltens P Toga AW Thompson PMHartley SW Witter MP Valentino DJ Launer LJ Hippocampal shapeanalysis in Alzheimerrsquos disease a population-based studyNeuroimage 2007 368ndash18
17 Klein S Loog M van der Lijn F den Heijer T Hammers A de Bruijne Mvan der Lugt A Duin RPW Breteler MMB Niessen WJ Early diagnosis ofdementia based on intersubject whole-brain dissimilarities InProceedings of the 2010 IEEE International Conference on BiomedicalImaging fromNano toMacro ISBIrsquo10 Piscataway NJ USA IEEE Press2010249ndash252
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 5 of 12httpwwwbiomedcentralcom1471-23421421
Table 3 An overview of the names and description of themarkers we used in this paper
Marker Description
Procrustes The center of mass of each regions alignedto the same space with a Procrustes alignment
Surface connectivity The percentage of how much each regionhave connected to other regions relatedto the surface of the region
Hippocampus volume The volume of the hippocampus dividedwith the intracranial volume
Whole brain volume The volume of the whole brain dividedwith the intracranial volume
accounts for as much information of the data Y as pos-sible PLS searches for the set of components (latentvariables) that performs a simulation decomposition of Xand Y with the constraint that these components shouldexplain as much as possible of the covariance betweenX and Y It is followed by a linear regression step wherethe decomposition of X is used to predict Y The PLSmodel will try to find the multidimensionality direction inthe X space that explains the maximummultidimensionalvariance direction in the Y space The number of PLScomponents were set to 10 based on our training experi-ments Due to its simple functionality we have used lineardiscriminate analysis (LDA) for the classification [42]LDA tries to reduce the dimensionality while preservingas much of the class discriminatory information as pos-sible LDA seeks to obtain a scalar y by projecting thesamples x onto a line y = wTx where x is the samplesand w contains the class information Of all possible waysto discriminate these we would like to select the one thatmaximizes the separability between the scalars yAll experiments were done in a leave-one-of-each-
class out fashion The data were adjusted for age andgender when there existed a linear correlation betweenthose
ResultThe fractional volume scores for the whole brain volumeand hippocampus volume for NC MCI and AD respec-tively is shown in Table 4 NC had a larger volume inboth whole brain and hippocampus than MCI and ADand MCI had a larger volume score than AD AD had thelargest volume lost between bl and m12For each feature set the area under the curve (AUC) was
computed and summarized in Table 5 for NC versus ADNC versus MCI and MCI versus AD and the correspond-ing ROC curves are shown in Figure 2 The classificationwas tested with a ranksum test and the p-values are alsoshown in Table 5 All markers were able to significantlydiscriminate between the three groups NC-AD NC-MCIand MCI-AD The AUC score were highest for the NC-AD group where our surface connectivity marker werecomparable to the hippocampus volume for the AD-NCand NC-MCI cases and better in the discrimination forthe MCI-AD case than the hippocampus volume TheAUC for the Procrustes marker were in general a littlelower than for the surface connectivity scoreNext we adjusted our markers for whole brain volume
and for hippocampus volume to investigate if our mark-ers contained additional information than the volumesThese results are shown in Table 5 The signal lowersbut was still significant Again the surface connectivitymarkers performed better then the Procrustes markersand the NC-AD classification result were the best Thesurface connectivity markers were generally better to dis-criminate NC-MCI than MCI-AD and for the Procrustesmarkers it was vice versa It was the smaller group-ings functional and potato-shaped that gave the bestperformanceWe have also investigated how our markers performed
on the period to month 12 using the score differencesbetween bl and month 12 for each marker and the AUCand the corresponding ranksum p-values are shown inTable 6 and roc curves in Figure 3 Hippocampus andwhole brain showed relatively low AUC result due to the
Table 4 Fractional volume scores for the hippocampus and the whole brain at bl andmonth 12 and the volume loss
Group Time point Whole brain Hippocampusvolume fraction (cm3) volume fraction (cm3)
The last two markers were divided in three groups all functional and potato-shaped 5(a) is the non-adjusted case 5(b) and 5(c) is adjusted by whole brain fractionand hippocampus fraction respectively All markers were able to significantly distinguish the classes Our markers were still significant after adjustment for the twovolume scores but AUC scores were in general lower than the non-adjusted scores The surface connectivity score for the functional groups performed the best
use of static Freesurfer volumes from bl and month 12Our surface connectivity scores performed the best for allthree groups NC-AD NC-MCI andMCI-AD The resultsbetween NC-AD and NC-MCI are very similarWe have adjusted the month 12 classification results for
both the baseline whole brain and the baseline hippocam-pus volume shown in Table 6 The results showed a sig-nificant classification for our markers When adjusted forwhole brain volume the surface connectivity performedthe best The classification result for MCI-AD case wasbetter than the NC-AD resultFinally we have classifiedMCI-c against MCI-nc where
the non-adjusted result is shown in Table 7 The sur-face connectivity markers was the only marker that wasable to distinguish the two groups and only in the func-tional and potato-shaped grouping of regions When weadjusted for whole brain volume the surface connectiv-ity marker was still significant with an AUC at 0631
(p = 0012) and for the potato group it was borderlinesignificant with an AUC at 0595 (p = 0067) In thecase where we adjusted for hippocampus volume onlythe surface connectivity marker for the functional groupswas borderline significant with an AUC of 0599 (p =0055) No other significance were shown in the adjustedcases
Discussion and conclusionWe have investigated a novel way of looking at the rela-tionship between different regions in the brain We eval-uated a surface connectivity marker and center of massbased marker and their ability to classify between NCMCI and AD subjects Both markers have been able tosignificantly discriminate between the three classes AD-NC NC-MCI andMCI-AD both at baseline and betweenbaseline and month 12 Our surface connectivity markerwas also able to classify MCI-c
Lillemark et al BMCMedical Imaging 2014 1421 Page 7 of 12httpwwwbiomedcentralcom1471-23421421
Figure 2 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
The large variabilityrsquos in the brain regions is related toAlzheimerrsquos Disease [171925-27] and this have moti-vated our two markers describing the proximity betweenthe regions in the brain Both our markers were ableto significantly differentiate between AD and NC alsowhen adjusted for whole brain and hippocampus volumeThe surface connectivity marker was comparable to hip-pocampus volume which is one of known most effect fullmarkers from MRI Also after adjustment for volumes wehad a significant classification results this indicates thatour markers hold additional information about the devel-opment of the brain in relation to progression of ADWe believe that our markers capture an individual shrink-age due to pathological alterations In subjects with ADthe cerebral cortex is shrinking the sulcirsquos is widenedthe cortical ribbon may be thinned and ventricles aredilated [24344] Our surface connectivity markers maycapture some of these pathological alterations in measur-ing the proximity between regionsWe have evaluated our markers over a 1 - year period
where we have investigated the change in the Procrustesaligned positions and the change in surface connectivityIn this case we were also able to significantly discrim-inate between the classes although the signal was lessstrong The weakened signal can be due to noise in thesegmentation of the data Our markers were not taken
from registered brains but normalized within the samebrain so they captured comparable information acrosstime and study population The segmentation of the indi-vidual regions at two time steps can still be quite differ-ent and when we were using the difference between thescore values it can introduce noise in our markers Thisis also visible in the values for hippocampus and wholebrain volume in the longitudinal part of our study whichshowed lower results for classification than other reportedresults [1745]Our surface connectivity marker performed the best
indicating that it captured how the cell death caused byAD minimizes the surface connectivity between regionsThis was most visible in the functional regions The func-tional group were limited to functional regions of thebrain and the good performance of this grouping is in linewith the knowledge that AD affect the network aroundand including the medial temporal lobe and disruption inthis region contributes to memory impairment [46] Thelower performance of our Procrustes marker could be dueto the captured information is closer to volume and thatno particular regions moves related to the others but allregions moved due to general volume lossCuingnet et al [18] have made a comparison study
for classification of NC versus AD NC versus MCI-converters (MCI-c) and MCI-c versus MCI-non-
Lillemark et al BMCMedical Imaging 2014 1421 Page 8 of 12httpwwwbiomedcentralcom1471-23421421
Table 6 Classification result for NC-AD NC-MCI andMCI-AD for the difference between the bl andmonth 12makers 6(a)is the not adjusted case 6(b) is adjusted for bl whole brain volume and 6(c) is adjusted for baseline hippocampus volume
Our markers was still able to significantly discriminate between the three groups Our surface connectivity markers for the two subgroups functional and potatoperformed the best
converters (MCI-nc) based on 81 NC 67 MCI-nc 39MCI-c and 69 AD subjects from the ADNI databaseThey investigated voxel based segmented tissue regionsfor the whole brain in six different variants and for graymatter (GM) and GM white matter (WM) and cere-brospinal fluid (CSF) combined cortical thickness inthree different variants and finally hippocampus volumeand shape in three different variants a total of ten differ-ent methods They conclude that all methods were able toclassify NC vs AD with a sensitivity and specificity at therange from 59 - 81 and 77 - 98 respectively whichis comparable to our classification Other prediction stud-ies have shown better classification rates at 67 - 92 forcross-sectional studies [1417194547] and 69 - 815for longitudinal studies [19-21] The difference in theclassification accuracy between our method and the otherpapers can be explained by the tuning of methods and theuse of different data sets
Only our surface connectivity marker was able to clas-sify MCI-c fromMCI-nc and not with a highly significantresult This is in line with Cuignet et al comparison studyfor AD classification where they found that only fourmethods managed to predict MCI-c vs MCI-nc betterthan a random classifier and none of those got signifi-cantly better results [18] The main reason for the lowresult in the conversion case could be due to the fact thatMCI is a very in heterogeneous group that possibly couldconvert rapidly to AD or be stable for many years beforeconversionOther studies have investigated the change locally in
the hippocampus Wang et al [13] have used large-deformation diffeomorphic high-dimensional brain map-ping to quantify and compare changes in the hippocampalshape as well as volume They found that shape changeswere largely confined to the head of hippocampus andsubiculum for normal controls (NC) Other studies have
Lillemark et al BMCMedical Imaging 2014 1421 Page 9 of 12httpwwwbiomedcentralcom1471-23421421
Figure 3 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
confirmed these shape changes for the hippocampus[14-16] based on shape models and local hippocampalatrophy patterns We have focused on investigating therelationship between the different regions of the brain andhow they differ between healthy subjects and AD patientsThis way of investigating the regions could make it pos-sible to incorporate different kind of knowledge into thesame model where one could go from the individual scaleof each region to the interaction between the regionsand finally to combined picture of the brain as one wholeregion
Table 7 The AUC and corresponding p-values for theclassification of MCI-c andMCI-nc
Markers AUC pminusvalue
HPICV 0466 0516
WBICV 0512 0823
Surface all 0542 0416
Surface func 0624 0017
Surface potato 0603 0048
Procrustes all 0465 0486
Procrustes func 0498 0964
Procrustes potato 0534 0501
Only the surface connectivity markers was able to significantly discriminate thetwo groups functional and potato-shaped
An alternative use of MRI images for early predictionof AD is by using texture analysis where different texturesfeatures is used to construct a computational frameworkwhich have been able to discriminate AD MCI and NCwith a separability of up to 95 [234048] This indicatesthat one can combine the three different kinds of mark-ers volume texture and shapeproximity markers to get amore sophisticated picture of the disease progressionOther image modalities such as single-photon emis-
sion computed tomography (SPECT) functional MRI andMR spectroscopy (MRS) positron emission tomography(PET) and molecular imaging have been used for investi-gation of brain changes related to AD SPECT combinedwith MRI images can give additional information aboutdisease progression when combined [49] Functional MRIand MR spectroscopy (MRS) have shown changes inmetabolic levels even prior to symptom onset in ADbut are difficult to implement in clinical settings due totechnical support [5051] PET metabolic imaging withradioactive glucose has also been used to examined thefunctional change and tracking of the AD disease progres-sion [5253] Due to the invasiveness radiation dose limi-tation requiring lumbar punctures and high cost PET isunsuitable for repeated measurements of a single patientor screening programs for large populations Molecularimaging with amyloid tracers have showed great potential
Lillemark et al BMCMedical Imaging 2014 1421 Page 10 of 12httpwwwbiomedcentralcom1471-23421421
as to be accurate markers for early diagnosis of AD but donot show progression in established disease [5455] whichis our object of interestTo conclude structural MRI is an suitable image modal-
ity for detection of AD and AD progression Our mark-ers have shown promising results in capturing how theproximity of different regions in the brain can aid inAD diagnosis and prognosis The proximity analysis cap-tures additional information about the whole brain com-pared to atrophy scores This additional information cancontribute to the refinement of the AD markers andmay be able to give a more detailed picture of ADprogression
Competing interestsThe authors declare that they have no competing interests
Authorsrsquo contributionsLL have contributed in study design data analysis and interpretation preparedand submitted the manuscript LS and AP performed study design and datacollection EBD and MN participated in design and reviewed manuscript Allauthors have read and approved the final manuscript
AcknowledgementsWe gratefully acknowledge the funding from the Danish Research Foundation(Den Danske Forskningsfond) and The Danish National Advanced TechnologyFoundation supporting this work and FreeSurfer for providing the softwareused for the segmentations in this paper Data collection and sharing for thisproject was funded by the Alzheimerrsquos Disease Neuroimaging Initiative (ADNI)(National Institutes of Health Grant U01 AG024904) and DOD ADNI(Department of Defense award number W81XWH-12-2-0012) ADNI is fundedby the National Institute on Aging the National Institute of Bio medicalImaging and Bioengineering and through generous contributions from thefollowing Alzheimerrsquos Association Alzheimerrsquos Drug Discovery FoundationBioClinica Inc Biogen Idec Inc Bristol-Myers Squibb Company Eisai Inc ElanPharmaceuticals Inc Eli Lilly and Company F Hoffmann-La Roche Ltd and itsaffiliated company Genentech Inc GE Healthcare Innogenetics NV IXICOLtd Janssen Alzheimer Immunotherapy Research amp Development LLCJohnson amp Johnson Pharmaceutical Research amp Development LLC MedpaceInc Merck amp Co Inc Meso Scale Diagnostics LLC NeuroRx Research NovartisPharmaceuticals Corporation Pfizer Inc Piramal Imaging Servier Synarc Incand Takeda Pharmaceutical Company The Canadian Institutes of HealthResearch is providing funds to Rev December 5 2013 support ADNI clinicalsites in Canada Private sector contributions are facilitated by the Foundationfor the National Institutes of Health (wwwfnihorg) The grantee organizationis the Northern California Institute for Research and Education and the study iscoordinated by the Alzheimerrsquos Disease Cooperative Study at the University ofCalifornia San Diego ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern CaliforniaData used in preparation of this article were obtained from the AlzheimerrsquosDisease Neuroimaging Initiative (ADNI) database (adniloniuscedu) As suchthe investigators within the ADNI contributed to the design andimplementation of ADNI andor provided data but did not participate inanalysis or writing of this report A complete listing of ADNI investigators canbe found at httpadniloniusceduwp-contentuploadshow_to_applyADNI_Acknowledgement_Listpdf
Author details1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash Denmark 2Biomediq Fruebjergvej3 2100 Copenhagen Oslash Denmark
Received 2 January 2014 Accepted 9 May 2014Published 2 June 2014
References1 Alzheimerrsquos association 2011
[httpwwwalzorgdownloadsFacts_Figures_2011pdf]2 Braskie MN Klunder AD Hayashi KM Protas H Kepe V Miller KJ Huang SC
Barrio JR Ercoli LM Siddarth P Satyamurthy N Liu J Toga AWBookheimer SY Small GW Thompson PM Plaque and tangle imagingand cognition in normal aging and Alzheimerrsquos disease NeurobiolAging 2010 311669ndash1678
3 Braak H Braak E Neuropathological stageing of alzheimer-relatedchanges Acta neuropathologica 1991 82(4)239ndash259
4 West MJ Coleman PD Flood DG Troncoso JC Differences in thepattern of hippocampal neuronal loss in normal ageing andAlzheimerrsquos disease Lancet 1994 344769ndash772
5 Apostolova LG Mosconi L Thompson PM Green AE Hwang KS RamirezA Mistur R Tsui WH de Leon MJ Subregional hippocampal atrophypredicts alzheimerrsquos dementia in the cognitively normal NeurobiolAging 2010 31(7)1077ndash1088
6 Tondelli M Wilcock GK Nichelli P De Jager CA Jenkinson M Zamboni GStructural mri changes detectable up to ten years before clinicalalzheimerrsquos disease Neurobiol Aging 2012 33(4)825ndash25
7 Bernard C Helmer C Dilharreguy B Amieva H Auriacombe S DartiguesJ-F Allard M Catheline G Time course of brain volume changes in thepreclinical phase of alzheimerrsquos disease Alzheimerrsquos Dementia 201410(2)143ndash151
8 Dickerson B Stoub T Shah R Sperling R Killiany R Albert M Hyman BBlacker D deToledo-Morrell L Alzheimer-signature mri biomarkerpredicts ad dementia in cognitively normal adults Neurology 201176(16)1395ndash1402
9 Hansson O Zetterberg H Buchhave P Londos E Blennow K Minthon LAssociation between csf biomarkers and incipient alzheimerrsquosdisease in patients with mild cognitive impairment a follow-upstudy Lancet Neurol 2006 5(3)228ndash234
10 Leung KK Shen K-K Barnes J Ridgway GR Clarkson MJ Fripp JSalvado O Meriaudeau F Fox NC Bourgeat P Ourselin S Increasingpower to predict mild cognitive impairment conversion toalzheimerrsquos disease using hippocampal atrophy rate andstatistical shape models In Proceedings of the 13th InternationalConference onMedical Image Computing and Computer-assistedIntervention Part II MICCAIrsquo10 Berlin Heidelberg Springer2010125ndash132
11 Holland D Dale AM Nonlinear registration of longitudinal imagesandmeasurement of change in regions of interestMed Image Anal2011 15(4)489ndash497
12 Smith SM Zhang Y Jenkinson M Chen J Matthews P Federico ADe Stefano N Accurate robust and automated longitudinal andcross-sectional brain change analysis Neuroimage 200217(1)479ndash489
13 Wang L Swank JS Glick IE Gado MH Miller MI Morris JC Csernansky JGChanges in hippocampal volume and shape across time distinguishdementia of the Alzheimer type from healthy aging Neuroimage2003 20667ndash682
14 Li S Shi F Pu F Li X Jiang T Xie S Wang Y Hippocampal shape analysisof Alzheimer disease based onmachine learning methods AJNR AmJ Neuroradiol 2007 281339ndash1345
15 Costafreda SG Dinov ID Tu Z Shi Y Liu CY Kloszewska I Mecocci PSoininen H Tsolakif M Vellasg B Wahlundh L-O Spengerh C Togab AWLovestonea S Simmonsa A Automated hippocampal shape analysispredicts the onset of dementia in mild cognitive impairmentNeuroImage 2011
16 Scher AI Xu Y Korf ES White LR Scheltens P Toga AW Thompson PMHartley SW Witter MP Valentino DJ Launer LJ Hippocampal shapeanalysis in Alzheimerrsquos disease a population-based studyNeuroimage 2007 368ndash18
17 Klein S Loog M van der Lijn F den Heijer T Hammers A de Bruijne Mvan der Lugt A Duin RPW Breteler MMB Niessen WJ Early diagnosis ofdementia based on intersubject whole-brain dissimilarities InProceedings of the 2010 IEEE International Conference on BiomedicalImaging fromNano toMacro ISBIrsquo10 Piscataway NJ USA IEEE Press2010249ndash252
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 6 of 12httpwwwbiomedcentralcom1471-23421421
Table 5 The AUC values and corresponding ranksum p-values for classification of AD-NC NC-MCI andMCI-AD
The last two markers were divided in three groups all functional and potato-shaped 5(a) is the non-adjusted case 5(b) and 5(c) is adjusted by whole brain fractionand hippocampus fraction respectively All markers were able to significantly distinguish the classes Our markers were still significant after adjustment for the twovolume scores but AUC scores were in general lower than the non-adjusted scores The surface connectivity score for the functional groups performed the best
use of static Freesurfer volumes from bl and month 12Our surface connectivity scores performed the best for allthree groups NC-AD NC-MCI andMCI-AD The resultsbetween NC-AD and NC-MCI are very similarWe have adjusted the month 12 classification results for
both the baseline whole brain and the baseline hippocam-pus volume shown in Table 6 The results showed a sig-nificant classification for our markers When adjusted forwhole brain volume the surface connectivity performedthe best The classification result for MCI-AD case wasbetter than the NC-AD resultFinally we have classifiedMCI-c against MCI-nc where
the non-adjusted result is shown in Table 7 The sur-face connectivity markers was the only marker that wasable to distinguish the two groups and only in the func-tional and potato-shaped grouping of regions When weadjusted for whole brain volume the surface connectiv-ity marker was still significant with an AUC at 0631
(p = 0012) and for the potato group it was borderlinesignificant with an AUC at 0595 (p = 0067) In thecase where we adjusted for hippocampus volume onlythe surface connectivity marker for the functional groupswas borderline significant with an AUC of 0599 (p =0055) No other significance were shown in the adjustedcases
Discussion and conclusionWe have investigated a novel way of looking at the rela-tionship between different regions in the brain We eval-uated a surface connectivity marker and center of massbased marker and their ability to classify between NCMCI and AD subjects Both markers have been able tosignificantly discriminate between the three classes AD-NC NC-MCI andMCI-AD both at baseline and betweenbaseline and month 12 Our surface connectivity markerwas also able to classify MCI-c
Lillemark et al BMCMedical Imaging 2014 1421 Page 7 of 12httpwwwbiomedcentralcom1471-23421421
Figure 2 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
The large variabilityrsquos in the brain regions is related toAlzheimerrsquos Disease [171925-27] and this have moti-vated our two markers describing the proximity betweenthe regions in the brain Both our markers were ableto significantly differentiate between AD and NC alsowhen adjusted for whole brain and hippocampus volumeThe surface connectivity marker was comparable to hip-pocampus volume which is one of known most effect fullmarkers from MRI Also after adjustment for volumes wehad a significant classification results this indicates thatour markers hold additional information about the devel-opment of the brain in relation to progression of ADWe believe that our markers capture an individual shrink-age due to pathological alterations In subjects with ADthe cerebral cortex is shrinking the sulcirsquos is widenedthe cortical ribbon may be thinned and ventricles aredilated [24344] Our surface connectivity markers maycapture some of these pathological alterations in measur-ing the proximity between regionsWe have evaluated our markers over a 1 - year period
where we have investigated the change in the Procrustesaligned positions and the change in surface connectivityIn this case we were also able to significantly discrim-inate between the classes although the signal was lessstrong The weakened signal can be due to noise in thesegmentation of the data Our markers were not taken
from registered brains but normalized within the samebrain so they captured comparable information acrosstime and study population The segmentation of the indi-vidual regions at two time steps can still be quite differ-ent and when we were using the difference between thescore values it can introduce noise in our markers Thisis also visible in the values for hippocampus and wholebrain volume in the longitudinal part of our study whichshowed lower results for classification than other reportedresults [1745]Our surface connectivity marker performed the best
indicating that it captured how the cell death caused byAD minimizes the surface connectivity between regionsThis was most visible in the functional regions The func-tional group were limited to functional regions of thebrain and the good performance of this grouping is in linewith the knowledge that AD affect the network aroundand including the medial temporal lobe and disruption inthis region contributes to memory impairment [46] Thelower performance of our Procrustes marker could be dueto the captured information is closer to volume and thatno particular regions moves related to the others but allregions moved due to general volume lossCuingnet et al [18] have made a comparison study
for classification of NC versus AD NC versus MCI-converters (MCI-c) and MCI-c versus MCI-non-
Lillemark et al BMCMedical Imaging 2014 1421 Page 8 of 12httpwwwbiomedcentralcom1471-23421421
Table 6 Classification result for NC-AD NC-MCI andMCI-AD for the difference between the bl andmonth 12makers 6(a)is the not adjusted case 6(b) is adjusted for bl whole brain volume and 6(c) is adjusted for baseline hippocampus volume
Our markers was still able to significantly discriminate between the three groups Our surface connectivity markers for the two subgroups functional and potatoperformed the best
converters (MCI-nc) based on 81 NC 67 MCI-nc 39MCI-c and 69 AD subjects from the ADNI databaseThey investigated voxel based segmented tissue regionsfor the whole brain in six different variants and for graymatter (GM) and GM white matter (WM) and cere-brospinal fluid (CSF) combined cortical thickness inthree different variants and finally hippocampus volumeand shape in three different variants a total of ten differ-ent methods They conclude that all methods were able toclassify NC vs AD with a sensitivity and specificity at therange from 59 - 81 and 77 - 98 respectively whichis comparable to our classification Other prediction stud-ies have shown better classification rates at 67 - 92 forcross-sectional studies [1417194547] and 69 - 815for longitudinal studies [19-21] The difference in theclassification accuracy between our method and the otherpapers can be explained by the tuning of methods and theuse of different data sets
Only our surface connectivity marker was able to clas-sify MCI-c fromMCI-nc and not with a highly significantresult This is in line with Cuignet et al comparison studyfor AD classification where they found that only fourmethods managed to predict MCI-c vs MCI-nc betterthan a random classifier and none of those got signifi-cantly better results [18] The main reason for the lowresult in the conversion case could be due to the fact thatMCI is a very in heterogeneous group that possibly couldconvert rapidly to AD or be stable for many years beforeconversionOther studies have investigated the change locally in
the hippocampus Wang et al [13] have used large-deformation diffeomorphic high-dimensional brain map-ping to quantify and compare changes in the hippocampalshape as well as volume They found that shape changeswere largely confined to the head of hippocampus andsubiculum for normal controls (NC) Other studies have
Lillemark et al BMCMedical Imaging 2014 1421 Page 9 of 12httpwwwbiomedcentralcom1471-23421421
Figure 3 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
confirmed these shape changes for the hippocampus[14-16] based on shape models and local hippocampalatrophy patterns We have focused on investigating therelationship between the different regions of the brain andhow they differ between healthy subjects and AD patientsThis way of investigating the regions could make it pos-sible to incorporate different kind of knowledge into thesame model where one could go from the individual scaleof each region to the interaction between the regionsand finally to combined picture of the brain as one wholeregion
Table 7 The AUC and corresponding p-values for theclassification of MCI-c andMCI-nc
Markers AUC pminusvalue
HPICV 0466 0516
WBICV 0512 0823
Surface all 0542 0416
Surface func 0624 0017
Surface potato 0603 0048
Procrustes all 0465 0486
Procrustes func 0498 0964
Procrustes potato 0534 0501
Only the surface connectivity markers was able to significantly discriminate thetwo groups functional and potato-shaped
An alternative use of MRI images for early predictionof AD is by using texture analysis where different texturesfeatures is used to construct a computational frameworkwhich have been able to discriminate AD MCI and NCwith a separability of up to 95 [234048] This indicatesthat one can combine the three different kinds of mark-ers volume texture and shapeproximity markers to get amore sophisticated picture of the disease progressionOther image modalities such as single-photon emis-
sion computed tomography (SPECT) functional MRI andMR spectroscopy (MRS) positron emission tomography(PET) and molecular imaging have been used for investi-gation of brain changes related to AD SPECT combinedwith MRI images can give additional information aboutdisease progression when combined [49] Functional MRIand MR spectroscopy (MRS) have shown changes inmetabolic levels even prior to symptom onset in ADbut are difficult to implement in clinical settings due totechnical support [5051] PET metabolic imaging withradioactive glucose has also been used to examined thefunctional change and tracking of the AD disease progres-sion [5253] Due to the invasiveness radiation dose limi-tation requiring lumbar punctures and high cost PET isunsuitable for repeated measurements of a single patientor screening programs for large populations Molecularimaging with amyloid tracers have showed great potential
Lillemark et al BMCMedical Imaging 2014 1421 Page 10 of 12httpwwwbiomedcentralcom1471-23421421
as to be accurate markers for early diagnosis of AD but donot show progression in established disease [5455] whichis our object of interestTo conclude structural MRI is an suitable image modal-
ity for detection of AD and AD progression Our mark-ers have shown promising results in capturing how theproximity of different regions in the brain can aid inAD diagnosis and prognosis The proximity analysis cap-tures additional information about the whole brain com-pared to atrophy scores This additional information cancontribute to the refinement of the AD markers andmay be able to give a more detailed picture of ADprogression
Competing interestsThe authors declare that they have no competing interests
Authorsrsquo contributionsLL have contributed in study design data analysis and interpretation preparedand submitted the manuscript LS and AP performed study design and datacollection EBD and MN participated in design and reviewed manuscript Allauthors have read and approved the final manuscript
AcknowledgementsWe gratefully acknowledge the funding from the Danish Research Foundation(Den Danske Forskningsfond) and The Danish National Advanced TechnologyFoundation supporting this work and FreeSurfer for providing the softwareused for the segmentations in this paper Data collection and sharing for thisproject was funded by the Alzheimerrsquos Disease Neuroimaging Initiative (ADNI)(National Institutes of Health Grant U01 AG024904) and DOD ADNI(Department of Defense award number W81XWH-12-2-0012) ADNI is fundedby the National Institute on Aging the National Institute of Bio medicalImaging and Bioengineering and through generous contributions from thefollowing Alzheimerrsquos Association Alzheimerrsquos Drug Discovery FoundationBioClinica Inc Biogen Idec Inc Bristol-Myers Squibb Company Eisai Inc ElanPharmaceuticals Inc Eli Lilly and Company F Hoffmann-La Roche Ltd and itsaffiliated company Genentech Inc GE Healthcare Innogenetics NV IXICOLtd Janssen Alzheimer Immunotherapy Research amp Development LLCJohnson amp Johnson Pharmaceutical Research amp Development LLC MedpaceInc Merck amp Co Inc Meso Scale Diagnostics LLC NeuroRx Research NovartisPharmaceuticals Corporation Pfizer Inc Piramal Imaging Servier Synarc Incand Takeda Pharmaceutical Company The Canadian Institutes of HealthResearch is providing funds to Rev December 5 2013 support ADNI clinicalsites in Canada Private sector contributions are facilitated by the Foundationfor the National Institutes of Health (wwwfnihorg) The grantee organizationis the Northern California Institute for Research and Education and the study iscoordinated by the Alzheimerrsquos Disease Cooperative Study at the University ofCalifornia San Diego ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern CaliforniaData used in preparation of this article were obtained from the AlzheimerrsquosDisease Neuroimaging Initiative (ADNI) database (adniloniuscedu) As suchthe investigators within the ADNI contributed to the design andimplementation of ADNI andor provided data but did not participate inanalysis or writing of this report A complete listing of ADNI investigators canbe found at httpadniloniusceduwp-contentuploadshow_to_applyADNI_Acknowledgement_Listpdf
Author details1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash Denmark 2Biomediq Fruebjergvej3 2100 Copenhagen Oslash Denmark
Received 2 January 2014 Accepted 9 May 2014Published 2 June 2014
References1 Alzheimerrsquos association 2011
[httpwwwalzorgdownloadsFacts_Figures_2011pdf]2 Braskie MN Klunder AD Hayashi KM Protas H Kepe V Miller KJ Huang SC
Barrio JR Ercoli LM Siddarth P Satyamurthy N Liu J Toga AWBookheimer SY Small GW Thompson PM Plaque and tangle imagingand cognition in normal aging and Alzheimerrsquos disease NeurobiolAging 2010 311669ndash1678
3 Braak H Braak E Neuropathological stageing of alzheimer-relatedchanges Acta neuropathologica 1991 82(4)239ndash259
4 West MJ Coleman PD Flood DG Troncoso JC Differences in thepattern of hippocampal neuronal loss in normal ageing andAlzheimerrsquos disease Lancet 1994 344769ndash772
5 Apostolova LG Mosconi L Thompson PM Green AE Hwang KS RamirezA Mistur R Tsui WH de Leon MJ Subregional hippocampal atrophypredicts alzheimerrsquos dementia in the cognitively normal NeurobiolAging 2010 31(7)1077ndash1088
6 Tondelli M Wilcock GK Nichelli P De Jager CA Jenkinson M Zamboni GStructural mri changes detectable up to ten years before clinicalalzheimerrsquos disease Neurobiol Aging 2012 33(4)825ndash25
7 Bernard C Helmer C Dilharreguy B Amieva H Auriacombe S DartiguesJ-F Allard M Catheline G Time course of brain volume changes in thepreclinical phase of alzheimerrsquos disease Alzheimerrsquos Dementia 201410(2)143ndash151
8 Dickerson B Stoub T Shah R Sperling R Killiany R Albert M Hyman BBlacker D deToledo-Morrell L Alzheimer-signature mri biomarkerpredicts ad dementia in cognitively normal adults Neurology 201176(16)1395ndash1402
9 Hansson O Zetterberg H Buchhave P Londos E Blennow K Minthon LAssociation between csf biomarkers and incipient alzheimerrsquosdisease in patients with mild cognitive impairment a follow-upstudy Lancet Neurol 2006 5(3)228ndash234
10 Leung KK Shen K-K Barnes J Ridgway GR Clarkson MJ Fripp JSalvado O Meriaudeau F Fox NC Bourgeat P Ourselin S Increasingpower to predict mild cognitive impairment conversion toalzheimerrsquos disease using hippocampal atrophy rate andstatistical shape models In Proceedings of the 13th InternationalConference onMedical Image Computing and Computer-assistedIntervention Part II MICCAIrsquo10 Berlin Heidelberg Springer2010125ndash132
11 Holland D Dale AM Nonlinear registration of longitudinal imagesandmeasurement of change in regions of interestMed Image Anal2011 15(4)489ndash497
12 Smith SM Zhang Y Jenkinson M Chen J Matthews P Federico ADe Stefano N Accurate robust and automated longitudinal andcross-sectional brain change analysis Neuroimage 200217(1)479ndash489
13 Wang L Swank JS Glick IE Gado MH Miller MI Morris JC Csernansky JGChanges in hippocampal volume and shape across time distinguishdementia of the Alzheimer type from healthy aging Neuroimage2003 20667ndash682
14 Li S Shi F Pu F Li X Jiang T Xie S Wang Y Hippocampal shape analysisof Alzheimer disease based onmachine learning methods AJNR AmJ Neuroradiol 2007 281339ndash1345
15 Costafreda SG Dinov ID Tu Z Shi Y Liu CY Kloszewska I Mecocci PSoininen H Tsolakif M Vellasg B Wahlundh L-O Spengerh C Togab AWLovestonea S Simmonsa A Automated hippocampal shape analysispredicts the onset of dementia in mild cognitive impairmentNeuroImage 2011
16 Scher AI Xu Y Korf ES White LR Scheltens P Toga AW Thompson PMHartley SW Witter MP Valentino DJ Launer LJ Hippocampal shapeanalysis in Alzheimerrsquos disease a population-based studyNeuroimage 2007 368ndash18
17 Klein S Loog M van der Lijn F den Heijer T Hammers A de Bruijne Mvan der Lugt A Duin RPW Breteler MMB Niessen WJ Early diagnosis ofdementia based on intersubject whole-brain dissimilarities InProceedings of the 2010 IEEE International Conference on BiomedicalImaging fromNano toMacro ISBIrsquo10 Piscataway NJ USA IEEE Press2010249ndash252
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 7 of 12httpwwwbiomedcentralcom1471-23421421
Figure 2 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
The large variabilityrsquos in the brain regions is related toAlzheimerrsquos Disease [171925-27] and this have moti-vated our two markers describing the proximity betweenthe regions in the brain Both our markers were ableto significantly differentiate between AD and NC alsowhen adjusted for whole brain and hippocampus volumeThe surface connectivity marker was comparable to hip-pocampus volume which is one of known most effect fullmarkers from MRI Also after adjustment for volumes wehad a significant classification results this indicates thatour markers hold additional information about the devel-opment of the brain in relation to progression of ADWe believe that our markers capture an individual shrink-age due to pathological alterations In subjects with ADthe cerebral cortex is shrinking the sulcirsquos is widenedthe cortical ribbon may be thinned and ventricles aredilated [24344] Our surface connectivity markers maycapture some of these pathological alterations in measur-ing the proximity between regionsWe have evaluated our markers over a 1 - year period
where we have investigated the change in the Procrustesaligned positions and the change in surface connectivityIn this case we were also able to significantly discrim-inate between the classes although the signal was lessstrong The weakened signal can be due to noise in thesegmentation of the data Our markers were not taken
from registered brains but normalized within the samebrain so they captured comparable information acrosstime and study population The segmentation of the indi-vidual regions at two time steps can still be quite differ-ent and when we were using the difference between thescore values it can introduce noise in our markers Thisis also visible in the values for hippocampus and wholebrain volume in the longitudinal part of our study whichshowed lower results for classification than other reportedresults [1745]Our surface connectivity marker performed the best
indicating that it captured how the cell death caused byAD minimizes the surface connectivity between regionsThis was most visible in the functional regions The func-tional group were limited to functional regions of thebrain and the good performance of this grouping is in linewith the knowledge that AD affect the network aroundand including the medial temporal lobe and disruption inthis region contributes to memory impairment [46] Thelower performance of our Procrustes marker could be dueto the captured information is closer to volume and thatno particular regions moves related to the others but allregions moved due to general volume lossCuingnet et al [18] have made a comparison study
for classification of NC versus AD NC versus MCI-converters (MCI-c) and MCI-c versus MCI-non-
Lillemark et al BMCMedical Imaging 2014 1421 Page 8 of 12httpwwwbiomedcentralcom1471-23421421
Table 6 Classification result for NC-AD NC-MCI andMCI-AD for the difference between the bl andmonth 12makers 6(a)is the not adjusted case 6(b) is adjusted for bl whole brain volume and 6(c) is adjusted for baseline hippocampus volume
Our markers was still able to significantly discriminate between the three groups Our surface connectivity markers for the two subgroups functional and potatoperformed the best
converters (MCI-nc) based on 81 NC 67 MCI-nc 39MCI-c and 69 AD subjects from the ADNI databaseThey investigated voxel based segmented tissue regionsfor the whole brain in six different variants and for graymatter (GM) and GM white matter (WM) and cere-brospinal fluid (CSF) combined cortical thickness inthree different variants and finally hippocampus volumeand shape in three different variants a total of ten differ-ent methods They conclude that all methods were able toclassify NC vs AD with a sensitivity and specificity at therange from 59 - 81 and 77 - 98 respectively whichis comparable to our classification Other prediction stud-ies have shown better classification rates at 67 - 92 forcross-sectional studies [1417194547] and 69 - 815for longitudinal studies [19-21] The difference in theclassification accuracy between our method and the otherpapers can be explained by the tuning of methods and theuse of different data sets
Only our surface connectivity marker was able to clas-sify MCI-c fromMCI-nc and not with a highly significantresult This is in line with Cuignet et al comparison studyfor AD classification where they found that only fourmethods managed to predict MCI-c vs MCI-nc betterthan a random classifier and none of those got signifi-cantly better results [18] The main reason for the lowresult in the conversion case could be due to the fact thatMCI is a very in heterogeneous group that possibly couldconvert rapidly to AD or be stable for many years beforeconversionOther studies have investigated the change locally in
the hippocampus Wang et al [13] have used large-deformation diffeomorphic high-dimensional brain map-ping to quantify and compare changes in the hippocampalshape as well as volume They found that shape changeswere largely confined to the head of hippocampus andsubiculum for normal controls (NC) Other studies have
Lillemark et al BMCMedical Imaging 2014 1421 Page 9 of 12httpwwwbiomedcentralcom1471-23421421
Figure 3 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
confirmed these shape changes for the hippocampus[14-16] based on shape models and local hippocampalatrophy patterns We have focused on investigating therelationship between the different regions of the brain andhow they differ between healthy subjects and AD patientsThis way of investigating the regions could make it pos-sible to incorporate different kind of knowledge into thesame model where one could go from the individual scaleof each region to the interaction between the regionsand finally to combined picture of the brain as one wholeregion
Table 7 The AUC and corresponding p-values for theclassification of MCI-c andMCI-nc
Markers AUC pminusvalue
HPICV 0466 0516
WBICV 0512 0823
Surface all 0542 0416
Surface func 0624 0017
Surface potato 0603 0048
Procrustes all 0465 0486
Procrustes func 0498 0964
Procrustes potato 0534 0501
Only the surface connectivity markers was able to significantly discriminate thetwo groups functional and potato-shaped
An alternative use of MRI images for early predictionof AD is by using texture analysis where different texturesfeatures is used to construct a computational frameworkwhich have been able to discriminate AD MCI and NCwith a separability of up to 95 [234048] This indicatesthat one can combine the three different kinds of mark-ers volume texture and shapeproximity markers to get amore sophisticated picture of the disease progressionOther image modalities such as single-photon emis-
sion computed tomography (SPECT) functional MRI andMR spectroscopy (MRS) positron emission tomography(PET) and molecular imaging have been used for investi-gation of brain changes related to AD SPECT combinedwith MRI images can give additional information aboutdisease progression when combined [49] Functional MRIand MR spectroscopy (MRS) have shown changes inmetabolic levels even prior to symptom onset in ADbut are difficult to implement in clinical settings due totechnical support [5051] PET metabolic imaging withradioactive glucose has also been used to examined thefunctional change and tracking of the AD disease progres-sion [5253] Due to the invasiveness radiation dose limi-tation requiring lumbar punctures and high cost PET isunsuitable for repeated measurements of a single patientor screening programs for large populations Molecularimaging with amyloid tracers have showed great potential
Lillemark et al BMCMedical Imaging 2014 1421 Page 10 of 12httpwwwbiomedcentralcom1471-23421421
as to be accurate markers for early diagnosis of AD but donot show progression in established disease [5455] whichis our object of interestTo conclude structural MRI is an suitable image modal-
ity for detection of AD and AD progression Our mark-ers have shown promising results in capturing how theproximity of different regions in the brain can aid inAD diagnosis and prognosis The proximity analysis cap-tures additional information about the whole brain com-pared to atrophy scores This additional information cancontribute to the refinement of the AD markers andmay be able to give a more detailed picture of ADprogression
Competing interestsThe authors declare that they have no competing interests
Authorsrsquo contributionsLL have contributed in study design data analysis and interpretation preparedand submitted the manuscript LS and AP performed study design and datacollection EBD and MN participated in design and reviewed manuscript Allauthors have read and approved the final manuscript
AcknowledgementsWe gratefully acknowledge the funding from the Danish Research Foundation(Den Danske Forskningsfond) and The Danish National Advanced TechnologyFoundation supporting this work and FreeSurfer for providing the softwareused for the segmentations in this paper Data collection and sharing for thisproject was funded by the Alzheimerrsquos Disease Neuroimaging Initiative (ADNI)(National Institutes of Health Grant U01 AG024904) and DOD ADNI(Department of Defense award number W81XWH-12-2-0012) ADNI is fundedby the National Institute on Aging the National Institute of Bio medicalImaging and Bioengineering and through generous contributions from thefollowing Alzheimerrsquos Association Alzheimerrsquos Drug Discovery FoundationBioClinica Inc Biogen Idec Inc Bristol-Myers Squibb Company Eisai Inc ElanPharmaceuticals Inc Eli Lilly and Company F Hoffmann-La Roche Ltd and itsaffiliated company Genentech Inc GE Healthcare Innogenetics NV IXICOLtd Janssen Alzheimer Immunotherapy Research amp Development LLCJohnson amp Johnson Pharmaceutical Research amp Development LLC MedpaceInc Merck amp Co Inc Meso Scale Diagnostics LLC NeuroRx Research NovartisPharmaceuticals Corporation Pfizer Inc Piramal Imaging Servier Synarc Incand Takeda Pharmaceutical Company The Canadian Institutes of HealthResearch is providing funds to Rev December 5 2013 support ADNI clinicalsites in Canada Private sector contributions are facilitated by the Foundationfor the National Institutes of Health (wwwfnihorg) The grantee organizationis the Northern California Institute for Research and Education and the study iscoordinated by the Alzheimerrsquos Disease Cooperative Study at the University ofCalifornia San Diego ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern CaliforniaData used in preparation of this article were obtained from the AlzheimerrsquosDisease Neuroimaging Initiative (ADNI) database (adniloniuscedu) As suchthe investigators within the ADNI contributed to the design andimplementation of ADNI andor provided data but did not participate inanalysis or writing of this report A complete listing of ADNI investigators canbe found at httpadniloniusceduwp-contentuploadshow_to_applyADNI_Acknowledgement_Listpdf
Author details1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash Denmark 2Biomediq Fruebjergvej3 2100 Copenhagen Oslash Denmark
Received 2 January 2014 Accepted 9 May 2014Published 2 June 2014
References1 Alzheimerrsquos association 2011
[httpwwwalzorgdownloadsFacts_Figures_2011pdf]2 Braskie MN Klunder AD Hayashi KM Protas H Kepe V Miller KJ Huang SC
Barrio JR Ercoli LM Siddarth P Satyamurthy N Liu J Toga AWBookheimer SY Small GW Thompson PM Plaque and tangle imagingand cognition in normal aging and Alzheimerrsquos disease NeurobiolAging 2010 311669ndash1678
3 Braak H Braak E Neuropathological stageing of alzheimer-relatedchanges Acta neuropathologica 1991 82(4)239ndash259
4 West MJ Coleman PD Flood DG Troncoso JC Differences in thepattern of hippocampal neuronal loss in normal ageing andAlzheimerrsquos disease Lancet 1994 344769ndash772
5 Apostolova LG Mosconi L Thompson PM Green AE Hwang KS RamirezA Mistur R Tsui WH de Leon MJ Subregional hippocampal atrophypredicts alzheimerrsquos dementia in the cognitively normal NeurobiolAging 2010 31(7)1077ndash1088
6 Tondelli M Wilcock GK Nichelli P De Jager CA Jenkinson M Zamboni GStructural mri changes detectable up to ten years before clinicalalzheimerrsquos disease Neurobiol Aging 2012 33(4)825ndash25
7 Bernard C Helmer C Dilharreguy B Amieva H Auriacombe S DartiguesJ-F Allard M Catheline G Time course of brain volume changes in thepreclinical phase of alzheimerrsquos disease Alzheimerrsquos Dementia 201410(2)143ndash151
8 Dickerson B Stoub T Shah R Sperling R Killiany R Albert M Hyman BBlacker D deToledo-Morrell L Alzheimer-signature mri biomarkerpredicts ad dementia in cognitively normal adults Neurology 201176(16)1395ndash1402
9 Hansson O Zetterberg H Buchhave P Londos E Blennow K Minthon LAssociation between csf biomarkers and incipient alzheimerrsquosdisease in patients with mild cognitive impairment a follow-upstudy Lancet Neurol 2006 5(3)228ndash234
10 Leung KK Shen K-K Barnes J Ridgway GR Clarkson MJ Fripp JSalvado O Meriaudeau F Fox NC Bourgeat P Ourselin S Increasingpower to predict mild cognitive impairment conversion toalzheimerrsquos disease using hippocampal atrophy rate andstatistical shape models In Proceedings of the 13th InternationalConference onMedical Image Computing and Computer-assistedIntervention Part II MICCAIrsquo10 Berlin Heidelberg Springer2010125ndash132
11 Holland D Dale AM Nonlinear registration of longitudinal imagesandmeasurement of change in regions of interestMed Image Anal2011 15(4)489ndash497
12 Smith SM Zhang Y Jenkinson M Chen J Matthews P Federico ADe Stefano N Accurate robust and automated longitudinal andcross-sectional brain change analysis Neuroimage 200217(1)479ndash489
13 Wang L Swank JS Glick IE Gado MH Miller MI Morris JC Csernansky JGChanges in hippocampal volume and shape across time distinguishdementia of the Alzheimer type from healthy aging Neuroimage2003 20667ndash682
14 Li S Shi F Pu F Li X Jiang T Xie S Wang Y Hippocampal shape analysisof Alzheimer disease based onmachine learning methods AJNR AmJ Neuroradiol 2007 281339ndash1345
15 Costafreda SG Dinov ID Tu Z Shi Y Liu CY Kloszewska I Mecocci PSoininen H Tsolakif M Vellasg B Wahlundh L-O Spengerh C Togab AWLovestonea S Simmonsa A Automated hippocampal shape analysispredicts the onset of dementia in mild cognitive impairmentNeuroImage 2011
16 Scher AI Xu Y Korf ES White LR Scheltens P Toga AW Thompson PMHartley SW Witter MP Valentino DJ Launer LJ Hippocampal shapeanalysis in Alzheimerrsquos disease a population-based studyNeuroimage 2007 368ndash18
17 Klein S Loog M van der Lijn F den Heijer T Hammers A de Bruijne Mvan der Lugt A Duin RPW Breteler MMB Niessen WJ Early diagnosis ofdementia based on intersubject whole-brain dissimilarities InProceedings of the 2010 IEEE International Conference on BiomedicalImaging fromNano toMacro ISBIrsquo10 Piscataway NJ USA IEEE Press2010249ndash252
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 8 of 12httpwwwbiomedcentralcom1471-23421421
Table 6 Classification result for NC-AD NC-MCI andMCI-AD for the difference between the bl andmonth 12makers 6(a)is the not adjusted case 6(b) is adjusted for bl whole brain volume and 6(c) is adjusted for baseline hippocampus volume
Our markers was still able to significantly discriminate between the three groups Our surface connectivity markers for the two subgroups functional and potatoperformed the best
converters (MCI-nc) based on 81 NC 67 MCI-nc 39MCI-c and 69 AD subjects from the ADNI databaseThey investigated voxel based segmented tissue regionsfor the whole brain in six different variants and for graymatter (GM) and GM white matter (WM) and cere-brospinal fluid (CSF) combined cortical thickness inthree different variants and finally hippocampus volumeand shape in three different variants a total of ten differ-ent methods They conclude that all methods were able toclassify NC vs AD with a sensitivity and specificity at therange from 59 - 81 and 77 - 98 respectively whichis comparable to our classification Other prediction stud-ies have shown better classification rates at 67 - 92 forcross-sectional studies [1417194547] and 69 - 815for longitudinal studies [19-21] The difference in theclassification accuracy between our method and the otherpapers can be explained by the tuning of methods and theuse of different data sets
Only our surface connectivity marker was able to clas-sify MCI-c fromMCI-nc and not with a highly significantresult This is in line with Cuignet et al comparison studyfor AD classification where they found that only fourmethods managed to predict MCI-c vs MCI-nc betterthan a random classifier and none of those got signifi-cantly better results [18] The main reason for the lowresult in the conversion case could be due to the fact thatMCI is a very in heterogeneous group that possibly couldconvert rapidly to AD or be stable for many years beforeconversionOther studies have investigated the change locally in
the hippocampus Wang et al [13] have used large-deformation diffeomorphic high-dimensional brain map-ping to quantify and compare changes in the hippocampalshape as well as volume They found that shape changeswere largely confined to the head of hippocampus andsubiculum for normal controls (NC) Other studies have
Lillemark et al BMCMedical Imaging 2014 1421 Page 9 of 12httpwwwbiomedcentralcom1471-23421421
Figure 3 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
confirmed these shape changes for the hippocampus[14-16] based on shape models and local hippocampalatrophy patterns We have focused on investigating therelationship between the different regions of the brain andhow they differ between healthy subjects and AD patientsThis way of investigating the regions could make it pos-sible to incorporate different kind of knowledge into thesame model where one could go from the individual scaleof each region to the interaction between the regionsand finally to combined picture of the brain as one wholeregion
Table 7 The AUC and corresponding p-values for theclassification of MCI-c andMCI-nc
Markers AUC pminusvalue
HPICV 0466 0516
WBICV 0512 0823
Surface all 0542 0416
Surface func 0624 0017
Surface potato 0603 0048
Procrustes all 0465 0486
Procrustes func 0498 0964
Procrustes potato 0534 0501
Only the surface connectivity markers was able to significantly discriminate thetwo groups functional and potato-shaped
An alternative use of MRI images for early predictionof AD is by using texture analysis where different texturesfeatures is used to construct a computational frameworkwhich have been able to discriminate AD MCI and NCwith a separability of up to 95 [234048] This indicatesthat one can combine the three different kinds of mark-ers volume texture and shapeproximity markers to get amore sophisticated picture of the disease progressionOther image modalities such as single-photon emis-
sion computed tomography (SPECT) functional MRI andMR spectroscopy (MRS) positron emission tomography(PET) and molecular imaging have been used for investi-gation of brain changes related to AD SPECT combinedwith MRI images can give additional information aboutdisease progression when combined [49] Functional MRIand MR spectroscopy (MRS) have shown changes inmetabolic levels even prior to symptom onset in ADbut are difficult to implement in clinical settings due totechnical support [5051] PET metabolic imaging withradioactive glucose has also been used to examined thefunctional change and tracking of the AD disease progres-sion [5253] Due to the invasiveness radiation dose limi-tation requiring lumbar punctures and high cost PET isunsuitable for repeated measurements of a single patientor screening programs for large populations Molecularimaging with amyloid tracers have showed great potential
Lillemark et al BMCMedical Imaging 2014 1421 Page 10 of 12httpwwwbiomedcentralcom1471-23421421
as to be accurate markers for early diagnosis of AD but donot show progression in established disease [5455] whichis our object of interestTo conclude structural MRI is an suitable image modal-
ity for detection of AD and AD progression Our mark-ers have shown promising results in capturing how theproximity of different regions in the brain can aid inAD diagnosis and prognosis The proximity analysis cap-tures additional information about the whole brain com-pared to atrophy scores This additional information cancontribute to the refinement of the AD markers andmay be able to give a more detailed picture of ADprogression
Competing interestsThe authors declare that they have no competing interests
Authorsrsquo contributionsLL have contributed in study design data analysis and interpretation preparedand submitted the manuscript LS and AP performed study design and datacollection EBD and MN participated in design and reviewed manuscript Allauthors have read and approved the final manuscript
AcknowledgementsWe gratefully acknowledge the funding from the Danish Research Foundation(Den Danske Forskningsfond) and The Danish National Advanced TechnologyFoundation supporting this work and FreeSurfer for providing the softwareused for the segmentations in this paper Data collection and sharing for thisproject was funded by the Alzheimerrsquos Disease Neuroimaging Initiative (ADNI)(National Institutes of Health Grant U01 AG024904) and DOD ADNI(Department of Defense award number W81XWH-12-2-0012) ADNI is fundedby the National Institute on Aging the National Institute of Bio medicalImaging and Bioengineering and through generous contributions from thefollowing Alzheimerrsquos Association Alzheimerrsquos Drug Discovery FoundationBioClinica Inc Biogen Idec Inc Bristol-Myers Squibb Company Eisai Inc ElanPharmaceuticals Inc Eli Lilly and Company F Hoffmann-La Roche Ltd and itsaffiliated company Genentech Inc GE Healthcare Innogenetics NV IXICOLtd Janssen Alzheimer Immunotherapy Research amp Development LLCJohnson amp Johnson Pharmaceutical Research amp Development LLC MedpaceInc Merck amp Co Inc Meso Scale Diagnostics LLC NeuroRx Research NovartisPharmaceuticals Corporation Pfizer Inc Piramal Imaging Servier Synarc Incand Takeda Pharmaceutical Company The Canadian Institutes of HealthResearch is providing funds to Rev December 5 2013 support ADNI clinicalsites in Canada Private sector contributions are facilitated by the Foundationfor the National Institutes of Health (wwwfnihorg) The grantee organizationis the Northern California Institute for Research and Education and the study iscoordinated by the Alzheimerrsquos Disease Cooperative Study at the University ofCalifornia San Diego ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern CaliforniaData used in preparation of this article were obtained from the AlzheimerrsquosDisease Neuroimaging Initiative (ADNI) database (adniloniuscedu) As suchthe investigators within the ADNI contributed to the design andimplementation of ADNI andor provided data but did not participate inanalysis or writing of this report A complete listing of ADNI investigators canbe found at httpadniloniusceduwp-contentuploadshow_to_applyADNI_Acknowledgement_Listpdf
Author details1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash Denmark 2Biomediq Fruebjergvej3 2100 Copenhagen Oslash Denmark
Received 2 January 2014 Accepted 9 May 2014Published 2 June 2014
References1 Alzheimerrsquos association 2011
[httpwwwalzorgdownloadsFacts_Figures_2011pdf]2 Braskie MN Klunder AD Hayashi KM Protas H Kepe V Miller KJ Huang SC
Barrio JR Ercoli LM Siddarth P Satyamurthy N Liu J Toga AWBookheimer SY Small GW Thompson PM Plaque and tangle imagingand cognition in normal aging and Alzheimerrsquos disease NeurobiolAging 2010 311669ndash1678
3 Braak H Braak E Neuropathological stageing of alzheimer-relatedchanges Acta neuropathologica 1991 82(4)239ndash259
4 West MJ Coleman PD Flood DG Troncoso JC Differences in thepattern of hippocampal neuronal loss in normal ageing andAlzheimerrsquos disease Lancet 1994 344769ndash772
5 Apostolova LG Mosconi L Thompson PM Green AE Hwang KS RamirezA Mistur R Tsui WH de Leon MJ Subregional hippocampal atrophypredicts alzheimerrsquos dementia in the cognitively normal NeurobiolAging 2010 31(7)1077ndash1088
6 Tondelli M Wilcock GK Nichelli P De Jager CA Jenkinson M Zamboni GStructural mri changes detectable up to ten years before clinicalalzheimerrsquos disease Neurobiol Aging 2012 33(4)825ndash25
7 Bernard C Helmer C Dilharreguy B Amieva H Auriacombe S DartiguesJ-F Allard M Catheline G Time course of brain volume changes in thepreclinical phase of alzheimerrsquos disease Alzheimerrsquos Dementia 201410(2)143ndash151
8 Dickerson B Stoub T Shah R Sperling R Killiany R Albert M Hyman BBlacker D deToledo-Morrell L Alzheimer-signature mri biomarkerpredicts ad dementia in cognitively normal adults Neurology 201176(16)1395ndash1402
9 Hansson O Zetterberg H Buchhave P Londos E Blennow K Minthon LAssociation between csf biomarkers and incipient alzheimerrsquosdisease in patients with mild cognitive impairment a follow-upstudy Lancet Neurol 2006 5(3)228ndash234
10 Leung KK Shen K-K Barnes J Ridgway GR Clarkson MJ Fripp JSalvado O Meriaudeau F Fox NC Bourgeat P Ourselin S Increasingpower to predict mild cognitive impairment conversion toalzheimerrsquos disease using hippocampal atrophy rate andstatistical shape models In Proceedings of the 13th InternationalConference onMedical Image Computing and Computer-assistedIntervention Part II MICCAIrsquo10 Berlin Heidelberg Springer2010125ndash132
11 Holland D Dale AM Nonlinear registration of longitudinal imagesandmeasurement of change in regions of interestMed Image Anal2011 15(4)489ndash497
12 Smith SM Zhang Y Jenkinson M Chen J Matthews P Federico ADe Stefano N Accurate robust and automated longitudinal andcross-sectional brain change analysis Neuroimage 200217(1)479ndash489
13 Wang L Swank JS Glick IE Gado MH Miller MI Morris JC Csernansky JGChanges in hippocampal volume and shape across time distinguishdementia of the Alzheimer type from healthy aging Neuroimage2003 20667ndash682
14 Li S Shi F Pu F Li X Jiang T Xie S Wang Y Hippocampal shape analysisof Alzheimer disease based onmachine learning methods AJNR AmJ Neuroradiol 2007 281339ndash1345
15 Costafreda SG Dinov ID Tu Z Shi Y Liu CY Kloszewska I Mecocci PSoininen H Tsolakif M Vellasg B Wahlundh L-O Spengerh C Togab AWLovestonea S Simmonsa A Automated hippocampal shape analysispredicts the onset of dementia in mild cognitive impairmentNeuroImage 2011
16 Scher AI Xu Y Korf ES White LR Scheltens P Toga AW Thompson PMHartley SW Witter MP Valentino DJ Launer LJ Hippocampal shapeanalysis in Alzheimerrsquos disease a population-based studyNeuroimage 2007 368ndash18
17 Klein S Loog M van der Lijn F den Heijer T Hammers A de Bruijne Mvan der Lugt A Duin RPW Breteler MMB Niessen WJ Early diagnosis ofdementia based on intersubject whole-brain dissimilarities InProceedings of the 2010 IEEE International Conference on BiomedicalImaging fromNano toMacro ISBIrsquo10 Piscataway NJ USA IEEE Press2010249ndash252
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 9 of 12httpwwwbiomedcentralcom1471-23421421
Figure 3 (a) show the ROC for AD vs NC (b) shows the ROC for NC vs MCI and (c) shows the ROC for MCI vs AD
confirmed these shape changes for the hippocampus[14-16] based on shape models and local hippocampalatrophy patterns We have focused on investigating therelationship between the different regions of the brain andhow they differ between healthy subjects and AD patientsThis way of investigating the regions could make it pos-sible to incorporate different kind of knowledge into thesame model where one could go from the individual scaleof each region to the interaction between the regionsand finally to combined picture of the brain as one wholeregion
Table 7 The AUC and corresponding p-values for theclassification of MCI-c andMCI-nc
Markers AUC pminusvalue
HPICV 0466 0516
WBICV 0512 0823
Surface all 0542 0416
Surface func 0624 0017
Surface potato 0603 0048
Procrustes all 0465 0486
Procrustes func 0498 0964
Procrustes potato 0534 0501
Only the surface connectivity markers was able to significantly discriminate thetwo groups functional and potato-shaped
An alternative use of MRI images for early predictionof AD is by using texture analysis where different texturesfeatures is used to construct a computational frameworkwhich have been able to discriminate AD MCI and NCwith a separability of up to 95 [234048] This indicatesthat one can combine the three different kinds of mark-ers volume texture and shapeproximity markers to get amore sophisticated picture of the disease progressionOther image modalities such as single-photon emis-
sion computed tomography (SPECT) functional MRI andMR spectroscopy (MRS) positron emission tomography(PET) and molecular imaging have been used for investi-gation of brain changes related to AD SPECT combinedwith MRI images can give additional information aboutdisease progression when combined [49] Functional MRIand MR spectroscopy (MRS) have shown changes inmetabolic levels even prior to symptom onset in ADbut are difficult to implement in clinical settings due totechnical support [5051] PET metabolic imaging withradioactive glucose has also been used to examined thefunctional change and tracking of the AD disease progres-sion [5253] Due to the invasiveness radiation dose limi-tation requiring lumbar punctures and high cost PET isunsuitable for repeated measurements of a single patientor screening programs for large populations Molecularimaging with amyloid tracers have showed great potential
Lillemark et al BMCMedical Imaging 2014 1421 Page 10 of 12httpwwwbiomedcentralcom1471-23421421
as to be accurate markers for early diagnosis of AD but donot show progression in established disease [5455] whichis our object of interestTo conclude structural MRI is an suitable image modal-
ity for detection of AD and AD progression Our mark-ers have shown promising results in capturing how theproximity of different regions in the brain can aid inAD diagnosis and prognosis The proximity analysis cap-tures additional information about the whole brain com-pared to atrophy scores This additional information cancontribute to the refinement of the AD markers andmay be able to give a more detailed picture of ADprogression
Competing interestsThe authors declare that they have no competing interests
Authorsrsquo contributionsLL have contributed in study design data analysis and interpretation preparedand submitted the manuscript LS and AP performed study design and datacollection EBD and MN participated in design and reviewed manuscript Allauthors have read and approved the final manuscript
AcknowledgementsWe gratefully acknowledge the funding from the Danish Research Foundation(Den Danske Forskningsfond) and The Danish National Advanced TechnologyFoundation supporting this work and FreeSurfer for providing the softwareused for the segmentations in this paper Data collection and sharing for thisproject was funded by the Alzheimerrsquos Disease Neuroimaging Initiative (ADNI)(National Institutes of Health Grant U01 AG024904) and DOD ADNI(Department of Defense award number W81XWH-12-2-0012) ADNI is fundedby the National Institute on Aging the National Institute of Bio medicalImaging and Bioengineering and through generous contributions from thefollowing Alzheimerrsquos Association Alzheimerrsquos Drug Discovery FoundationBioClinica Inc Biogen Idec Inc Bristol-Myers Squibb Company Eisai Inc ElanPharmaceuticals Inc Eli Lilly and Company F Hoffmann-La Roche Ltd and itsaffiliated company Genentech Inc GE Healthcare Innogenetics NV IXICOLtd Janssen Alzheimer Immunotherapy Research amp Development LLCJohnson amp Johnson Pharmaceutical Research amp Development LLC MedpaceInc Merck amp Co Inc Meso Scale Diagnostics LLC NeuroRx Research NovartisPharmaceuticals Corporation Pfizer Inc Piramal Imaging Servier Synarc Incand Takeda Pharmaceutical Company The Canadian Institutes of HealthResearch is providing funds to Rev December 5 2013 support ADNI clinicalsites in Canada Private sector contributions are facilitated by the Foundationfor the National Institutes of Health (wwwfnihorg) The grantee organizationis the Northern California Institute for Research and Education and the study iscoordinated by the Alzheimerrsquos Disease Cooperative Study at the University ofCalifornia San Diego ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern CaliforniaData used in preparation of this article were obtained from the AlzheimerrsquosDisease Neuroimaging Initiative (ADNI) database (adniloniuscedu) As suchthe investigators within the ADNI contributed to the design andimplementation of ADNI andor provided data but did not participate inanalysis or writing of this report A complete listing of ADNI investigators canbe found at httpadniloniusceduwp-contentuploadshow_to_applyADNI_Acknowledgement_Listpdf
Author details1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash Denmark 2Biomediq Fruebjergvej3 2100 Copenhagen Oslash Denmark
Received 2 January 2014 Accepted 9 May 2014Published 2 June 2014
References1 Alzheimerrsquos association 2011
[httpwwwalzorgdownloadsFacts_Figures_2011pdf]2 Braskie MN Klunder AD Hayashi KM Protas H Kepe V Miller KJ Huang SC
Barrio JR Ercoli LM Siddarth P Satyamurthy N Liu J Toga AWBookheimer SY Small GW Thompson PM Plaque and tangle imagingand cognition in normal aging and Alzheimerrsquos disease NeurobiolAging 2010 311669ndash1678
3 Braak H Braak E Neuropathological stageing of alzheimer-relatedchanges Acta neuropathologica 1991 82(4)239ndash259
4 West MJ Coleman PD Flood DG Troncoso JC Differences in thepattern of hippocampal neuronal loss in normal ageing andAlzheimerrsquos disease Lancet 1994 344769ndash772
5 Apostolova LG Mosconi L Thompson PM Green AE Hwang KS RamirezA Mistur R Tsui WH de Leon MJ Subregional hippocampal atrophypredicts alzheimerrsquos dementia in the cognitively normal NeurobiolAging 2010 31(7)1077ndash1088
6 Tondelli M Wilcock GK Nichelli P De Jager CA Jenkinson M Zamboni GStructural mri changes detectable up to ten years before clinicalalzheimerrsquos disease Neurobiol Aging 2012 33(4)825ndash25
7 Bernard C Helmer C Dilharreguy B Amieva H Auriacombe S DartiguesJ-F Allard M Catheline G Time course of brain volume changes in thepreclinical phase of alzheimerrsquos disease Alzheimerrsquos Dementia 201410(2)143ndash151
8 Dickerson B Stoub T Shah R Sperling R Killiany R Albert M Hyman BBlacker D deToledo-Morrell L Alzheimer-signature mri biomarkerpredicts ad dementia in cognitively normal adults Neurology 201176(16)1395ndash1402
9 Hansson O Zetterberg H Buchhave P Londos E Blennow K Minthon LAssociation between csf biomarkers and incipient alzheimerrsquosdisease in patients with mild cognitive impairment a follow-upstudy Lancet Neurol 2006 5(3)228ndash234
10 Leung KK Shen K-K Barnes J Ridgway GR Clarkson MJ Fripp JSalvado O Meriaudeau F Fox NC Bourgeat P Ourselin S Increasingpower to predict mild cognitive impairment conversion toalzheimerrsquos disease using hippocampal atrophy rate andstatistical shape models In Proceedings of the 13th InternationalConference onMedical Image Computing and Computer-assistedIntervention Part II MICCAIrsquo10 Berlin Heidelberg Springer2010125ndash132
11 Holland D Dale AM Nonlinear registration of longitudinal imagesandmeasurement of change in regions of interestMed Image Anal2011 15(4)489ndash497
12 Smith SM Zhang Y Jenkinson M Chen J Matthews P Federico ADe Stefano N Accurate robust and automated longitudinal andcross-sectional brain change analysis Neuroimage 200217(1)479ndash489
13 Wang L Swank JS Glick IE Gado MH Miller MI Morris JC Csernansky JGChanges in hippocampal volume and shape across time distinguishdementia of the Alzheimer type from healthy aging Neuroimage2003 20667ndash682
14 Li S Shi F Pu F Li X Jiang T Xie S Wang Y Hippocampal shape analysisof Alzheimer disease based onmachine learning methods AJNR AmJ Neuroradiol 2007 281339ndash1345
15 Costafreda SG Dinov ID Tu Z Shi Y Liu CY Kloszewska I Mecocci PSoininen H Tsolakif M Vellasg B Wahlundh L-O Spengerh C Togab AWLovestonea S Simmonsa A Automated hippocampal shape analysispredicts the onset of dementia in mild cognitive impairmentNeuroImage 2011
16 Scher AI Xu Y Korf ES White LR Scheltens P Toga AW Thompson PMHartley SW Witter MP Valentino DJ Launer LJ Hippocampal shapeanalysis in Alzheimerrsquos disease a population-based studyNeuroimage 2007 368ndash18
17 Klein S Loog M van der Lijn F den Heijer T Hammers A de Bruijne Mvan der Lugt A Duin RPW Breteler MMB Niessen WJ Early diagnosis ofdementia based on intersubject whole-brain dissimilarities InProceedings of the 2010 IEEE International Conference on BiomedicalImaging fromNano toMacro ISBIrsquo10 Piscataway NJ USA IEEE Press2010249ndash252
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 10 of 12httpwwwbiomedcentralcom1471-23421421
as to be accurate markers for early diagnosis of AD but donot show progression in established disease [5455] whichis our object of interestTo conclude structural MRI is an suitable image modal-
ity for detection of AD and AD progression Our mark-ers have shown promising results in capturing how theproximity of different regions in the brain can aid inAD diagnosis and prognosis The proximity analysis cap-tures additional information about the whole brain com-pared to atrophy scores This additional information cancontribute to the refinement of the AD markers andmay be able to give a more detailed picture of ADprogression
Competing interestsThe authors declare that they have no competing interests
Authorsrsquo contributionsLL have contributed in study design data analysis and interpretation preparedand submitted the manuscript LS and AP performed study design and datacollection EBD and MN participated in design and reviewed manuscript Allauthors have read and approved the final manuscript
AcknowledgementsWe gratefully acknowledge the funding from the Danish Research Foundation(Den Danske Forskningsfond) and The Danish National Advanced TechnologyFoundation supporting this work and FreeSurfer for providing the softwareused for the segmentations in this paper Data collection and sharing for thisproject was funded by the Alzheimerrsquos Disease Neuroimaging Initiative (ADNI)(National Institutes of Health Grant U01 AG024904) and DOD ADNI(Department of Defense award number W81XWH-12-2-0012) ADNI is fundedby the National Institute on Aging the National Institute of Bio medicalImaging and Bioengineering and through generous contributions from thefollowing Alzheimerrsquos Association Alzheimerrsquos Drug Discovery FoundationBioClinica Inc Biogen Idec Inc Bristol-Myers Squibb Company Eisai Inc ElanPharmaceuticals Inc Eli Lilly and Company F Hoffmann-La Roche Ltd and itsaffiliated company Genentech Inc GE Healthcare Innogenetics NV IXICOLtd Janssen Alzheimer Immunotherapy Research amp Development LLCJohnson amp Johnson Pharmaceutical Research amp Development LLC MedpaceInc Merck amp Co Inc Meso Scale Diagnostics LLC NeuroRx Research NovartisPharmaceuticals Corporation Pfizer Inc Piramal Imaging Servier Synarc Incand Takeda Pharmaceutical Company The Canadian Institutes of HealthResearch is providing funds to Rev December 5 2013 support ADNI clinicalsites in Canada Private sector contributions are facilitated by the Foundationfor the National Institutes of Health (wwwfnihorg) The grantee organizationis the Northern California Institute for Research and Education and the study iscoordinated by the Alzheimerrsquos Disease Cooperative Study at the University ofCalifornia San Diego ADNI data are disseminated by the Laboratory for NeuroImaging at the University of Southern CaliforniaData used in preparation of this article were obtained from the AlzheimerrsquosDisease Neuroimaging Initiative (ADNI) database (adniloniuscedu) As suchthe investigators within the ADNI contributed to the design andimplementation of ADNI andor provided data but did not participate inanalysis or writing of this report A complete listing of ADNI investigators canbe found at httpadniloniusceduwp-contentuploadshow_to_applyADNI_Acknowledgement_Listpdf
Author details1Department of Computer Science University of CopenhagenUniversitetsparken 1 2100 Copenhagen Oslash Denmark 2Biomediq Fruebjergvej3 2100 Copenhagen Oslash Denmark
Received 2 January 2014 Accepted 9 May 2014Published 2 June 2014
References1 Alzheimerrsquos association 2011
[httpwwwalzorgdownloadsFacts_Figures_2011pdf]2 Braskie MN Klunder AD Hayashi KM Protas H Kepe V Miller KJ Huang SC
Barrio JR Ercoli LM Siddarth P Satyamurthy N Liu J Toga AWBookheimer SY Small GW Thompson PM Plaque and tangle imagingand cognition in normal aging and Alzheimerrsquos disease NeurobiolAging 2010 311669ndash1678
3 Braak H Braak E Neuropathological stageing of alzheimer-relatedchanges Acta neuropathologica 1991 82(4)239ndash259
4 West MJ Coleman PD Flood DG Troncoso JC Differences in thepattern of hippocampal neuronal loss in normal ageing andAlzheimerrsquos disease Lancet 1994 344769ndash772
5 Apostolova LG Mosconi L Thompson PM Green AE Hwang KS RamirezA Mistur R Tsui WH de Leon MJ Subregional hippocampal atrophypredicts alzheimerrsquos dementia in the cognitively normal NeurobiolAging 2010 31(7)1077ndash1088
6 Tondelli M Wilcock GK Nichelli P De Jager CA Jenkinson M Zamboni GStructural mri changes detectable up to ten years before clinicalalzheimerrsquos disease Neurobiol Aging 2012 33(4)825ndash25
7 Bernard C Helmer C Dilharreguy B Amieva H Auriacombe S DartiguesJ-F Allard M Catheline G Time course of brain volume changes in thepreclinical phase of alzheimerrsquos disease Alzheimerrsquos Dementia 201410(2)143ndash151
8 Dickerson B Stoub T Shah R Sperling R Killiany R Albert M Hyman BBlacker D deToledo-Morrell L Alzheimer-signature mri biomarkerpredicts ad dementia in cognitively normal adults Neurology 201176(16)1395ndash1402
9 Hansson O Zetterberg H Buchhave P Londos E Blennow K Minthon LAssociation between csf biomarkers and incipient alzheimerrsquosdisease in patients with mild cognitive impairment a follow-upstudy Lancet Neurol 2006 5(3)228ndash234
10 Leung KK Shen K-K Barnes J Ridgway GR Clarkson MJ Fripp JSalvado O Meriaudeau F Fox NC Bourgeat P Ourselin S Increasingpower to predict mild cognitive impairment conversion toalzheimerrsquos disease using hippocampal atrophy rate andstatistical shape models In Proceedings of the 13th InternationalConference onMedical Image Computing and Computer-assistedIntervention Part II MICCAIrsquo10 Berlin Heidelberg Springer2010125ndash132
11 Holland D Dale AM Nonlinear registration of longitudinal imagesandmeasurement of change in regions of interestMed Image Anal2011 15(4)489ndash497
12 Smith SM Zhang Y Jenkinson M Chen J Matthews P Federico ADe Stefano N Accurate robust and automated longitudinal andcross-sectional brain change analysis Neuroimage 200217(1)479ndash489
13 Wang L Swank JS Glick IE Gado MH Miller MI Morris JC Csernansky JGChanges in hippocampal volume and shape across time distinguishdementia of the Alzheimer type from healthy aging Neuroimage2003 20667ndash682
14 Li S Shi F Pu F Li X Jiang T Xie S Wang Y Hippocampal shape analysisof Alzheimer disease based onmachine learning methods AJNR AmJ Neuroradiol 2007 281339ndash1345
15 Costafreda SG Dinov ID Tu Z Shi Y Liu CY Kloszewska I Mecocci PSoininen H Tsolakif M Vellasg B Wahlundh L-O Spengerh C Togab AWLovestonea S Simmonsa A Automated hippocampal shape analysispredicts the onset of dementia in mild cognitive impairmentNeuroImage 2011
16 Scher AI Xu Y Korf ES White LR Scheltens P Toga AW Thompson PMHartley SW Witter MP Valentino DJ Launer LJ Hippocampal shapeanalysis in Alzheimerrsquos disease a population-based studyNeuroimage 2007 368ndash18
17 Klein S Loog M van der Lijn F den Heijer T Hammers A de Bruijne Mvan der Lugt A Duin RPW Breteler MMB Niessen WJ Early diagnosis ofdementia based on intersubject whole-brain dissimilarities InProceedings of the 2010 IEEE International Conference on BiomedicalImaging fromNano toMacro ISBIrsquo10 Piscataway NJ USA IEEE Press2010249ndash252
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 11 of 12httpwwwbiomedcentralcom1471-23421421
18 Cuingnet R Gerardin E Tessieras J Auzias G Leheacutericy S Habert MOChupin M Benali H Colliot O Automatic classification of patients withalzheimerrsquos disease from structural mri A comparison of tenmethods using the adni database Neuroimage 201156(2)766ndash781
19 Ferrarini L Frisoni GB Pievani M Reiber JHC Ganzola R Milles JMorphological hippocampal markers for automated detection ofalzheimerrsquos disease andmild cognitive impairment converters inmagnetic resonance images J Alzheimerrsquos Dis 200917(3)643ndash659
20 Achterberg HC Van Der Lijn F Den Heijer T Van Der Lugt A BretelerMMB Niessen WJ De Bruijne M Prediction of dementia byhippocampal shape analysis In Proceedings of the First InternationalConference onMachine Learning in Medical Imaging MLMIrsquo10 BerlinHeidelberg Springer 201042ndash49
21 Misra C Fan Y Davatzikos C Baseline and longitudinal patterns ofbrain atrophy in MCI patients and their use in prediction ofshort-term conversion to AD results from ADNI Neuroimage 2009441415ndash1422
22 Apostolova LG Dutton RA Dinov ID Hayashi KM Toga AW Cummings JLThompson PM Conversion of mild cognitive impairment toalzheimer disease predicted by hippocampal atrophy maps ArchNeurol 2006 63(5)693
23 Liu X Shi Y Thompson P Mio W Amodel of volumetric shape for theanalysis of longitudinal alzheimerrsquos disease data In Proceedings of the11th European Conference on Computer Vision Conference on ComputerVision Part III ECCVrsquo10 Berlin Heidelberg Springer 2010594ndash606
24 Thompson PM Hayashi KM De Zubicaray GI Janke AL Rose SE Semple JHong MS Herman DH Gravano D Doddrell DM Toga AWMappinghippocampal and ventricular change in Alzheimer diseaseNeuroimage 2004 221754ndash1766
25 den Heijer T Geerlings MI Hoebeek FE Hofman A Koudstaal PJ BretelerM Use of hippocampal and amygdalar volumes onmagneticresonance imaging to predict dementia in cognitively intact elderlypeople Arch Gen Psychiatry 2006 63(1)57
26 De Jong L Van Der Hiele K Veer I Houwing J Westendorp R Bollen EDe Bruin P Middelkoop H Van Buchem M Van Der Grond J Stronglyreduced volumes of putamen and thalamus in alzheimerrsquos diseasean mri study Brain 2008 131(12)3277ndash3285
27 Ferrarini L PalmWM Olofsen H van der Landen R van BuchemMA ReiberJH Admiraal-Behloul F Ventricular shape biomarkers for alzheimerrsquosdisease in clinical mr imagesMagn ResonMed 2008 59(2)260ndash267
28 Jack CR Bernstein MA Fox NC Thompson P Alexander G Harvey DBorowski B Britson PJ L Whitwell J Ward C Dale AM Felmlee JP GunterJL Hill DL Killiany R Schuff N Fox-Bosetti S Lin C Studholme C DeCarliCS Krueger G Ward HA Metzger GJ Scott KT Mallozzi R Blezek D Levy JDebbins JP Fleisher AS Albert M et al The Alzheimerrsquos Diseaseneuroimaging initiative (ADNI) MRI methods J Magn Reson Imaging JMRI 2008 27(4)685ndash691
29 McKhann G Drachman D Folstein M Katzman R Price D Stadlan EMClinical diagnosis of alzheimerrsquos disease report of the nincds-adrdawork group under the auspices of department of health andhuman services task force on alzheimerrsquos disease Neurology 198434(7)939ndash939
30 Wechsler D A standardized memory scale for clinical use J Psychol1945 19(1)87ndash95
31 Wyman BT Harvey DJ Crawford K Bernstein MA Carmichael O Cole PECrane PK DeCarli C Fox NC Gunter JL Hilli D Killianyj RJ Pachaik CSchwarzl AJ Schuffm N Senjemd ML Suhyn J Thompsonc PM WeineroM Jack Jr CR Standardization of analysis sets for reporting resultsfrom adni mri data Alzheimerrsquos Dementia 2012 9(3)332ndash337
32 Blennow K de Leon MJ Zetterberg H Alzheimerrsquos disease The Lancet2006 368(9533)387ndash403
33 Fischl B Salat DH Busa E Albert M Dieterich M Haselgrove C van derKouwe A Killiany R Kennedy D Klaveness S Montillo A Makris N Rosen BDale AMWhole brain segmentation automated labeling ofneuroanatomical structures in the human brain Neuron 200233341ndash355
34 Talairach J Tournoux P Co-planar Stereotaxic Atlas of the Human Brain3-Dimensional Proportional System an Approach to Cerebral ImagingStuttgart George Thieme 1988
35 Sled JG Zijdenbos AP Evans AC A nonparametric method forautomatic correction of intensity nonuniformity in mri dataMedImaging IEEE Trans on 1998 17(1)87ndash97
36 Narayana P Brey W Kulkarni M Sievenpiper C Compensation forsurface coil sensitivity variation in magnetic resonance imagingMagn Reson Imaging 1988 6(3)271ndash274
37 Sabuncu MR Yeo BT Van Leemput K Fischl B Golland P A generativemodel for image segmentation based on label fusionMed ImagingIEEE Trans on 2010 29(10)1714ndash1729
38 Krzyzanowska A Carro E Pathological alteration in the choroid plexusof alzheimerrsquos diseaseimplication for new therapy approaches FrontPharmacol 2012 31ndash5
40 Liu Y Teverovskiy L Carmichael O Kikinis R Shenton M Carter C StengerV Davis S Aizenstein H Becker J Lopez OL Meltzer CC Discriminativemr image feature analysis for automatic schizophrenia andalzheimerrsquos disease classificationMed Image Comput Comput AssistIntervndashMICCAI 2004 3216393ndash401
41 Geladi P Kowalski BR Partial least-squares regression a tutorial AnalChim Acta 1986 1851ndash17
42 Mika S Ratsch G Weston J Scholkopf B Mullers K Fisher discriminantanalysis with kernels In Neural Networks for Signal Processing IX 1999Proceedings of the 1999 IEEE Signal Processing Society Workshop IEEE199941ndash48
43 Braak H Braak E Neuropathological stageing of Alzheimer-relatedchanges Acta Neuropathol 1991 82239ndash259
44 Price JL Ko AI Wade MJ Tsou SK McKeel DW Morris JC Neuron numberin the entorhinal cortex and CA1 in preclinical Alzheimer diseaseArch Neurol 2001 581395ndash1402
45 Duchesne S Caroli A Geroldi C Barillot C Frisoni GB Collins DLMri-based automated computer classification of probablead versus normal controlsMed Imaging IEEE Trans on 200827(4)509ndash520
46 Buckner RL Snyder AZ Shannon BJ LaRossa G Sachs R Fotenos AFSheline YI Klunk WE Mathis CA Morris JC Mintun MAMolecularstructural and functional characterization of alzheimerrsquos diseaseevidence for a relationship between default activity amyloid andmemory J Neurosci 2005 25(34)7709ndash7717
47 Wang L Beg F Ratnanather T Ceritoglu C Younes L Morris JCCsernansky JG Miller MI Large deformation diffeomorphism andmomentum based hippocampal shape discrimination in dementiaof the alzheimer type IEEE Trans Med Imag 2007 26(4)462ndash470
48 Zhou X Liu Z Zhou Z Xia H Study on texture characteristics ofhippocampus in mr images of patients with alzheimerrsquos disease InBiomedical Engineering and Informatics (BMEI) 2010 3rd InternationalConference On Volume 2 Yantai China IEEE 2010593ndash596
49 Bonte FJ Weiner MF Bigio EH White CL Spect imaging in dementias JNuclear Med 2001 42(7)1131ndash1133
50 Johnson SC Saykin AJ Baxter LC Flashman LA Santulli RB McAllister TWMamourian AC The relationship between fmri activation andcerebral atrophy comparison of normal aging and alzheimerdisease Neuroimage 2000 11(3)179ndash187
51 Kantarci K Jack Jr C Xu Y Campeau N OrsquoBrien P Smith G Ivnik R Boeve BKokmen E Tangalos EG Petersen RC Regional metabolic patterns inmild cognitive impairment and alzheimerrsquos disease a 1hmrs studyNeurology 2000 55(2)210
52 Herholz K Salmon E Perani D Baron J Holthoff V Froumllich L SchoumlnknechtP Ito K Mielke R Kalbe E Zuumlndorfa G Delbeuckb X Pelatic O Anchisic DFazioc F Kerrouched N Desgrangesd B Eustached F Beuthien-BaumanniB Menzelk JC Schroumlderg J Katoh T Arahatah Y Henzel M Heissa W-DDiscrimination between alzheimer dementia and controls byautomated analysis of multicenter fdg pet Neuroimage 200217(1)302ndash316
53 De Leon M Convit A Wolf O Tarshish C DeSanti S Rusinek H Tsui WKandil E Scherer A Roche A Imossi A Thorn E Bobinski M Caraos CLesbre P Schlyer D Poirier J Reisberg B Fowler J Prediction ofcognitive decline in normal elderly subjects with 2-[18f]fluoro-2-deoxy-d-glucosepositron-emission tomography (fdgpet)Proc Nat Acad Sci 2001 98(19)10966
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
Dimensionality reduction and classification
Result
Discussion and conclusion
Competing interests
Authors contributions
Acknowledgements
Author details
References
Lillemark et al BMCMedical Imaging 2014 1421 Page 12 of 12httpwwwbiomedcentralcom1471-23421421
55 Klunk WE Engler H Nordberg A Wang Y Blomqvist G Holt DPBergstroumlm M Savitcheva I Huang GF Estrada S Auseacuten B Debnath MLBarletta J Price JC Sandell J Lopresti BJ Wall A Koivisto P Antoni GMathis CA Laringngstroumlm B Imaging brain amyloid in alzheimerrsquosdisease with pittsburgh compound-b Ann Neurol 200455(3)306ndash319
doi1011861471-2342-14-21Cite this article as Lillemark et al Brain regionrsquos relative proximity asmarker for Alzheimerrsquos disease based on structural MRI BMCMedicalImaging 2014 1421
Submit your next manuscript to BioMed Centraland take full advantage of
bull Convenient online submission
bull Thorough peer review
bull No space constraints or color figure charges
bull Immediate publication on acceptance
bull Inclusion in PubMed CAS Scopus and Google Scholar
bull Research which is freely available for redistribution
Submit your manuscript at wwwbiomedcentralcomsubmit
Abstract
Background
Methods
Results
Conclusion
Keywords
Background
Methods
ADNI brain MRI and preprocessing
MRI acquisition
Participants
Freesurfer segmentation
Grouping of the segmented regions
Surface connectivity marker procrustes marker and volume marker
