k-space and q-space: Combining ultra-high spatial

and angular resolution in diffusion imaging

using ZOOPPA at 7T

Robin M. Heidemann a, Alfred Anwander a,

Thorsten Feiweier b, Thomas R Knösche a, Robert Turner a

a Max Planck Institute for Human Cognitive and Brain Sciences, Leipzig, Germany.

b Siemens AG, Healthcare Sector, Allee am Roethelheimpark 2, 91052 Erlangen,

Germany

Corresponding author:

Dr. Robin M. Heidemann

Max Planck Institute for Human Cognitive and Brain Sciences,

Department of Neurophysics,

Stephanstrasse 1a

04103 Leipzig

Germany

http://www.cbs.mpg.de/~heidemann

Preprint submitted to Neuroimage (2011). Final draft: November 2011

Au

tho

r Ma

nu

sc

ript A

uth

or M

an

us

crip

t Au

tho

r Ma

nu

scrip

t

Author Manuscript Preprint submitted to Neuroimage (2012).

Published in final edited form as:

Neuroimage. Vol 60, Iss 2, Pages 967–978, 2012 DOI: 10.1016/j.neuroimage.2011.12.081

2

Abstract

There is ongoing debate whether using a higher spatial resolution (sampling k-

space) or a higher angular resolution (sampling q-space angles) is the better way to

improve diffusion MRI (dMRI) based tractography results in living humans. In both

cases, the limiting factor is the signal-to-noise ratio (SNR), due to the restricted

acquisition time. One possible way to increase the spatial resolution without

sacrificing either SNR or angular resolution is to move to a higher magnetic field

strength. Nevertheless, dMRI has not been the preferred application for ultra-high

field strength (7 Tesla). This is because single-shot echo-planar imaging (EPI) has

been the method of choice for human in vivo dMRI. EPI faces several challenges

related to the use of a high resolution at high field strength, for example, distortions

and image blurring. These problems can easily compromise the expected SNR gain

with field strength. In the current study, we introduce an adapted EPI sequence in

conjunction with a combination of ZOOmed imaging and Partially Parallel Acquisition

(ZOOPPA). We demonstrate that the method can produce high quality diffusion-

weighted images with high spatial and angular resolution at 7 Tesla. We provide

examples of in vivo human dMRI with isotropic resolutions of 1 mm and 800 µm.

These data sets are particularly suitable for resolving complex and subtle fiber

architectures, including fiber crossings in the white matter, anisotropy in the cortex

and fibers entering the cortex.

Keywords: Diffusion MRI, fiber-tracking, ultra-high field MRI, parallel imaging,

zoomed imaging

3

1. Introduction

Diffusion magnetic resonance imaging (dMRI) is currently the most important

tool for investigating white matter structures within the living human brain; see for

example (Jones, 2008). It has been shown that at a voxel resolution of around 2-3

mm, a simple tensor model can be used to track the major white matter pathways in

the human brain (Mori et al., 2002). However, this way it is impossible to accurately

track through regions with more complex fiber arrangements such as crossing,

fanning and branching. One way to address this issue would be to increase the

spatial resolution (extending the acquisition in k-space), thereby reducing the number

of concurrent fiber populations per voxel. However, increasing the resolution also

means decreasing the signal-to-noise ratio (SNR), and thus impairing the robustness

of fiber tract reconstruction.. Therefore, high spatial resolution has so far been

achieved at the expense of; (1) applicability to living humans (Roebroeck et al., 2008;

McNab et al., 2009; Miller et al., 2011), (2) isotropy, by using thick slices (Karampinos

et al., 2009; Finsterbusch 2009; Yassa et al., 2010), or (3) angular resolution (Sarlls

and Pierpaoli 2009). Alternatively, one can apply more sophisticated local models,

which are able to model multiple fiber populations and rely on higher angular

resolution and/or multiple b-values (extending the acquisition in q-space) (e.g. Tuch

et al., 2002; Tuch, 2004; Wedeen et al., 2005; Behrens et al., 2003; Tournier et al.,

2004; Jansons and Alexander, 2003). For example, in diffusion spectrum imaging

(DSI), a low spatial resolution of about 3 mm is used to achieve sufficient SNR to

enable the acquisition of a high number of diffusion directions and multiple b-values,

thereby resolving crossing fiber directions.

Since the MR signal scales with the main magnetic field strength, one possible

way to increase the spatial resolution without sacrificing SNR or angular resolution is

to move to a higher magnetic field strength. In theory, a gain of more than a factor of

two in SNR can be expected when moving from a 3 Tesla to a 7 Tesla MR scanner.

Even though applications such as high resolution anatomical imaging and functional

MRI are already benefitting from the higher field strength, dMRI has not been seen as

4

a preferred application for ultra-high field imaging yet. This is due to inherent

problems associated with high resolution diffusion-weighted single-shot EPI (ssh-EPI)

at high field strength. The two major challenges, which scale with the field strength

and the resolution, are susceptibility induced distortions and image blurring caused by

T2* relaxation. These effects can easily compromise the expected SNR gain for dMRI.

Even though these negative aspects have been identified and are well understood,

ssh-EPI is still the most frequently used methodology for human in vivo dMRI.

The above-described challenges can be addressed by using parallel imaging

methods, as was first shown by Griswold et al. (1999). In parallel imaging, also known

as Partially Parallel Acquisition (PPA), the MR data is acquired in parallel by multiple

independent receiver coils. This allows the acquisition to be accelerated by skipping a

certain fraction of MR data, followed by a reconstruction procedure. Typically, the

acceleration factor (AF) ranges between two and three, meaning that the echo train

length is reduced accordingly. In the reconstructed data set, after the parallel image

reconstruction, the time between consecutive echoes, the acquired and the

reconstructed echo, is the effective echo spacing. This is the echo spacing of the

acquired data divided by the AF. Hence, because EPI distortions linearly depend on

the time between consecutive echoes, they are reduced by parallel imaging.

Additionally, the shortened EPI readout duration reduces image blurring, allowing a

higher spatial resolution (Griswold et al., 1999) and improves the SNR. However, this

gain in SNR is reduced by an intrinsic SNR loss due to the reduced data sampling,

amounting to the square root of the AF. Furthermore, an additional SNR loss related

to the parallel image reconstruction procedure using receiver coils with sub-optimal

coil geometries has to be taken into account. This additional SNR loss is described by

the g-factor (Pruessmann et al., 1999), which is always greater than or equal to one.

More specifically, the SNR is decreased by 1/(g-factor AF). Please note that the g-

factor is in turn a non-linear function of the AF. For higher acceleration factors, in

particular, the g-factor increases drastically.

5

More than a decade before parallel imaging was used to accelerate in vivo MRI,

inner volume imaging (Feinberg et al., 1985) was introduced to speed up the

acquisition by acquiring a reduced field-of-view (FOV). A reduced FOV corresponds

to a reduced sampling density in k-space. In other words, a certain fraction of time-

consuming phase-encoding steps can be skipped, resulting in accelerated

acquisition. When the reduced FOV is smaller than the object to be imaged,

wraparound artifacts, also known as aliasing artifacts, will affect the resulting image.

In inner volume imaging, aliasing artifacts are avoided by exciting and refocusing only

the reduced FOV covering the region of interest.

A few years later, in zoomed imaging, the inner volume approach was applied to

EPI to achieve higher spatial resolution (Mansfield et al., 1988) instead of reducing

the acquisition time. This was achieved by keeping the number of phase encoding

steps constant. Similar to parallel imaging, reduced FOV imaging is affected by an

intrinsic SNR loss due to reduced data sampling, which is related to the AF. However,

reduced FOV imaging does not suffer from the additional losses due to the g-factor

penalty. The major disadvantage of reduced FOV imaging is the limited imaging

region. When large AFs are required, a very small FOV has to be chosen.

In the current work, we use a combination of a reduced FOV acquisition

(ZOOmed imaging) and Partially Parallel Acquisition (PPA), named ZOOPPA, to

benefit from the advantages of both methods and to improve the image quality of

single-shot EPI (Heidemann et al., 2008). For this purpose, we implemented

ZOOPPA in a diffusion-weighted EPI sequence (Heidemann et al., 2009), which is

used to perform isotropic high resolution diffusion MRI (dMRI) at ultra-high field

strength (7 Tesla). In conjunction with a high performance gradient system, this

approach enables dMRI with an isotropic resolution down to 800 µm. The dMRI data

acquired have sufficient SNR to resolve complex fiber configurations in white matter.

In addition, the diffusion signal in the cortex shows clear radial anisotropy and the

high spatial resolution allows the detection of subcomponents of white matter fiber

bundles, such as fibers entering the cortex, which are difficult to observe with MRI.

6

2. Theory and Methods

2.1 Accelerated MRI

As mentioned in the introduction, reduced FOV imaging can be used to

accelerate the MR data acquisition. Parallel imaging is also used to speed up the

acquisition. With regard to the k-space trajectory or sampling density, reduced FOV

imaging and parallel imaging are identical. In both cases, a reduced FOV is acquired

by reducing the k-space sampling density. The major difference lies in the way

aliasing artifacts, resulting from undersampling, are avoided. In parallel imaging, after

data acquisition, a reconstruction procedure is applied to the undersampled data in

order to obtain a final image without aliasing. This is done to obtain a full FOV image.

In comparison, for reduced FOV imaging, a pre-experiment is used to ensure that no

signal is received from the region outside the reduced FOV before the data

acquisition. For this pre-experiment, there are two main options. Besides the inner-

volume imaging methods including localized excitation (Feinberg et al., 1985;

Mansfield et al., 1988; Pauly et al., 1989), outer-volume suppression (OVS) (Le Roux

et al., 1998) can be used. For outer-volume suppression, tissue located outside the

reduced FOV is excited and subsequently suppressed. The OVS method is less

sensitive to B1 inhomogeneities than inner-volume excitation, a feature that is

especially advantageous for ultra-high field imaging (Pfeuffer et al., 2002). As in

parallel imaging, the reduced FOV approach allows a reduction of the EPI read-out

time, resulting in reduced distortions and T2* blurring. However, reduced FOV

imaging can be affected by remaining signal from the region outside the reduced

FOV, which will result in aliasing artifacts. Furthermore, the low coverage, especially

with high AF, is a disadvantage.

In fMRI, a high in-plane resolution has been achieved with both inner-volume

imaging (Duong et al., 2002) and outer volume suppression (Pfeuffer et al., 2002).

7

For diffusion MRI, inner-volume imaging has been used for ADC mapping of the

human optic nerve (Wheeler-Kingshott et al., 2002) and later, OVS was used for

spinal cord dMRI (Wilm et al., 2007). In both studies, the FOV was very small, 30 mm

and below, a low number of slices were acquired and highly anisotropic voxel sizes

were used. This was acceptable because the objects of interest, the optic nerve and

the spinal cord, fit within this small FOV and the structure does not change abruptly

along the main axis of the object. In this case, thick slices will not be significantly

affected by partial volume effects and a low number of slices is sufficient to cover the

main structure.

2.2 Zoomed Partially Parallel Acquisition (ZOOPPA)

To overcome the weakness of each separate technique, reduced FOV imaging

can be combined with partially parallel acquisition methods. This technique, given the

name ZOOPPA (ZOOmed Partially Parallel Acquisition), was originally developed

and implemented for fMRI (Heidemann et al., 2008) with gradient-echo EPI. We have

adapted ZOOPPA for diffusion-weighted spin-echo EPI (Heidemann et al., 2009). In

the current study, we use ZOOPPA to achieve high acceleration factors which enable

the acquisition of high resolution single-shot EPI data with a relatively large FOV.

Here, part of the overall acceleration comes from the zoomed approach, e.g. by

acquiring 50% FOV with OVS (see Fig 1B) resulting in an acceleration factor due to

zoomed imaging AFZOOM = 2. In a next step, this reduced FOV is further reduced by

parallel imaging, e.g. acquiring only one third of the reduced FOV (see Fig 1C)

resulting in an acceleration factor due to parallel imaging AFPPA = 3. In total, only a

sixth of the FOV is acquired. In this example, the overall acceleration factor would be

AF = AFZOOM * AFPPA = 2*3 = 6. Due to the parallel image reconstruction applied, 50%

of the FOV will be obtained (as in Fig. 1B) with AF = 6, instead of only 20% of the

FOV when only the zoomed approach is used.

8

Figure 2 depicts the experimental realization of this study. An OVS band is

placed over one half of the brain (here the posterior part), allowing a reduction of the

FOV by 50% (AFZOOM = 2) without aliasing. Within this reduced FOV, parallel imaging

is applied with AFPPA = 3, resulting in an overall acceleration factor AF = 6. This setup

allows imaging of half the brain with AF = 6, but with the g-factor penalty for AFPPA =

3. For comparison, in Fig. 2C a conventional GRAPPA acquisition with overall AF = 3

is shown. For the current OVS implementation, an asymmetric adiabatic full-passage

radio-frequency (RF) pulse is used, as proposed in Hwang et al. (1999) and Pfeuffer

et al. (2002). This “skewed” RF pulse1 provides a sharp edge to the region of interest.

In the remaining in vivo examples presented in this paper, AFZOOM < 2 and AFPPA = 3.

For parallel imaging, GeneRalized Autocalibrating Partially Parallel Acquisitions -

GRAPPA (Griswold at al., 2002) - was employed with a 2D convolution kernel

(Griswold, 2004) using three source points along the readout direction and two

source points along the phase encoding direction.

As described in Pruessmann et al. (1999), the SNR of an acquisition only

accelerated with parallel imaging can be derived by taking the g-factor penalty into

account:

AFg

SNRSNR

full

PPA (1)

For a reduced FOV acquisition without parallel image reconstruction, the g-

factor is one. In other words, we do not expect an additional g-factor penalty in the

resulting SNR for zooming. In the case of ZOOPPA, we have two independent

mechanisms for the SNR loss.

ZOOMPPAPPA

full

ZOOPPAAFAFAFg

SNRSNR

)( (2)

In Eq. 2, we expect an SNR loss as given in Eq. 1, due to the acceleration

achieved with parallel imaging. Here, it is important to note that the g-factor depends

1 Hyperbolic Secant HS1/2 with time-bandwidth product R = 0.9, pulse duration Tp = 0.9; tanh/tan with

time-bandwidth product R = 100, pulse duration Tp = 0.1

9

on AFPPA. The acceleration achieved with the zoomed approach results in an SNR

loss given by AFZOOM. When we compare GRAPPA and ZOOPPA, both with overall

AF = 4, but for ZOOPPA AFPPA = 2 and AFZOOM = 2, we will have the following

situation:

)2(

)4(4)4(

22)2(

PPA

PPA

full

PPA

PPA

full

GRAPPA

ZOOPPA

AFg

AFg

SNR

AFg

AFg

SNR

SNR

SNR (3)

We expect to achieve a higher SNR with ZOOPPA compared to GRAPPA, with

a maximum SNR difference determined by the ratio of the g-factors for ZOOPPA and

GRAPPA.

2.3 Phantom measurements

We performed phantom studies using a spherical phantom filled with

dimethylpolysiloxane oil to evaluate and compare SNR and g-factor maps for

conventional GRAPPA and ZOOPPA acquisitions. The SNR is estimated as

described in Firbank et al. (1999), by using the subtraction of two consecutive

acquisitions as noise map. The subtraction of two images has the advantage that coil

sensitivity variations and potentially remaining aliasing artifacts due to errors in the

GRAPPA reconstruction are canceled out and do not affect the SNR calculation. The

g-factor maps are derived using the method described by Breuer et al. (2009). For the

conventional GRAPPA acquisition of the phantom, an AFPPA between one and four is

used. For the ZOOPPA phantom experiments, the OVS is used to suppress signal

and the FOV is reduced by a factor of two. This results in an acceleration due to the

zoomed approach of AFZOOM = 2. GRAPPA is then applied to this already reduced

FOV with AFPPA = 2 and AFPPA = 3, resulting in an overall AF = 4 and 6. For

conventional GRAPPA as well as for ZOOPPA, identical imaging protocol parameters

were used.

10

2.4 Experimental setup for human dMRI

We acquired dMRI data from five healthy young volunteers (age 25 3 years).

Written informed consent was obtained from all participants in accordance with the

ethical approval from the University of Leipzig. Imaging was performed on a human

whole-body 7 Tesla MR scanner (MAGNETOM 7T, Siemens Healthcare, Erlangen,

Germany) with a gradient system achieving a maximum gradient amplitude of 70

mT/m with a maximum slew rate of 200 T/m/s (SC72, Siemens Healthcare, Erlangen,

Germany). With this gradient coil, 95% of the maximum gradient amplitude (67 mT/m)

was used for all dMRI acquisitions. Two participants were scanned with a different

gradient coil (AS95 DS: maximum gradient amplitude 40 mT/m, slew rate 200 T/m/s).

The AS95 gradient system has a lower maximum gradient amplitude and was

replaced by the newer SC72 gradients in our 7 Tesla system. For signal reception, a

24-element phased array head coil (Nova Medical, Wilmington, MA, USA) was used.

This coil is equipped with a single channel transmit coil for excitation and 24

independent receive-elements. Phase-encoding was chosen anterior-posterior to

obtain less pronounced, symmetric distortions. For the head geometry of the

volunteers examined in this study, in anterior-posterior direction, a minimum FOV of

200 – 218 mm is necessary to avoid aliasing. With the zoomed approach using OVS,

we reduced this FOV down to 141 – 144 mm. Based on the individual minimum

required FOV for each participant, the acceleration due to the zoomed approach

AFZOOM is between 1.39 and 1.53. Since the parallel imaging AFPPA is three for all in

vivo examples here, the resulting overall AF is in the range between 4.2 and 4.6 (see

Table 1).

For all acquisitions, an optimized monopolar Stejskal-Tanner sequence (Morelli

et al., 2010) was used in conjunction with ZOOPPA with the imaging protocol

parameters as listed in Table 1. In all experiments, dMRI data with b = 1000 s/mm2

and 60 directions (interspersed with 7 b0 images used for motion correction) were

obtained with 4-6 averages, resulting in a total acquisition time of around one hour.

For participants 1 and 2 (AS95 gradient system), a fat suppression method as

11

proposed by Ivanov et al. (2010) was used, which is well-suited to ultra-high field

dMRI. For the remaining participants, a combination of the above-mentioned method

and a gradient reversal fat saturation method (Park et al., 1987) was used. This

approach results in a more effective fat saturation as gradient reversal alone with a

reduced duration of the refocusing pulse compared to the method by Ivanov. Due to

this, a shorter TE can be used, resulting in a slightly higher SNR, as shown in Eichner

et al. (2011). For the parameter settings and hardware used, please refer to Table 1.

The data of the first two participants were corrected for subject motion in FSL

using the interspersed b0 images and linearly registered to a T1-weighted anatomical

scan. Then a diffusion tensor was fitted in each voxel and streamline trajectories were

computed using the tensor-line approach (Weinstein et al., 1999; Fillard et al., 2007)

implemented in MedINRIA2.

For the remaining data sets, we additionally applied a two-stage hybrid image

restoration procedure to the dMRI data prior to motion correction as described in

Lohmann et al. (2010) and implemented in Lipsia3. In each voxel, multiple fiber

orientations (significance level F > 0.05) (Behrens et al., 2007) were computed using

FSL4. In addition, the fiber orientation density function (fODF) was derived by

constrained spherical deconvolution (Tournier et al., 2007) based on spherical

harmonics of order 6, followed by whole-brain streamline-tracking using MRtrix5.

3. Results

3.1. Phantom study

The results of the phantom study are summarized in Fig 3. Here, we compare a

conventional GRAPPA acquisition with AFPPA = 4 (see Fig. 3A) and a ZOOPPA

acquisition with overall AF = 4 (see Fig. 3B). A similar ZOOPPA acceleration scheme

2 www-sop.inria.fr/asclepios/software/medinria

3 www.cbs.mpg.de/institute/software/lipsia

4 www.fmrib.ox.ac.uk/fsl

5 www.brain.org.au/software/mrtrix

12

was used for the following in vivo acquisitions. For both acquisition schemes, the

same overall acceleration was used and all imaging parameters were kept the same.

Corresponding g-factor maps are shown for conventional GRAPPA in Fig. 3C and for

ZOOPPA in Fig. 3D. The conventional GRAPPA g-factor map (Fig. 3C) has a

maximum g-factor of 3.2 and a mean g-factor within a region-of-interest (ROI) in the

phantom of 2.4. Using ZOOPPA with the same overall AF, the maximum g-factor is

1.9, while the mean g-factor in the same ROI is 1.7. This improvement is also visible

in the SNR evaluation, listed in Table 2. Here, SNR values are derived for different

ROIs of different size. ROI I is the upper half FOV of Fig. 3A, which corresponds to

the reduced FOV size of Fig. 3B. ROI II is indicated as a white box in Fig. 3A. ROI III

and IV are smaller and do not contain regions with background noise and no signal.

Even though the SNR values vary across different ROIs, which can be explained by

coil sensitivity variations, the ratios between the SNR values achieved with GRAPPA

and with ZOOPPA are very similar. This is reflected in the percentage gain in SNR for

ZOOPPA (numbers in brackets in Table 2), which is slightly above 25% for this

phantom study. Here, ROI I is excluded because it contains regions with background

noise and no signal and the SNR values might not be reliable in this region.

3.2 Human study

To demonstrate the effects of increased resolution and averaging, we show

color-coded direction maps in Fig. 4. A comparison is drawn between acquisitions

with isotropic resolutions of 1.5 mm (Fig. 4A) and 1.0 mm (Fig. 4B) by showing

enlarged sections of axial slices covering the occipital lobe. Obviously, due to the

increased resolution, more anatomical details are visible in the image obtained at 1.0

mm isotropic resolution. Furthermore, we see a better contrast between gray and

white matter. This can be seen in the enlarged sections of the fractional anisotropy

maps shown in Fig. 4C and in the profile lines through the sulcus indicated by the

white box (see middle of Fig. 4C). Compared to 1.5 mm isotropic resolution (green

13

profile), the cortex is clearly visible at 1.0 mm isotropic resolution (black profile).

However, with a very high isotropic resolution SNR becomes a major issue making

averaging and post processing of the raw data important, which is demonstrated in

Fig. 4D. Here, whole axial slices are shown from data with of 800 µm isotropic

resolution without averaging (Fig. 4D left), with 4 averages (Fig. 4D middle) and with

4 averages and the image restoration procedure (Fig. 4D right).

Figure 5 shows trace-weighted images as an example of the coverage

achievable using ZOOPPA with 1 mm isotropic resolution. It comprises the posterior

two thirds of the brain and misses structures like the prefrontal cortex and the anterior

temporal lobe due to the OVS. Even though we acquired data with a large number of

slices (see Table 1), due to the high voxel resolution, the coverage is limited. In the

example shown, it leads to the omission of brain stem and cerebellum. The left

column of Fig. 5 shows the trace weighted images and the middle column shows the

trace weighted images overlaid with the gray matter white matter boundaries derived

from a corresponding anatomical scan and the right column shows additional axial

slices from different positions in the brain.

In the following, we present a number of examples selected to highlight specific

image properties of the ZOOPPA method at 7 Tesla and the resulting opportunities

for imaging fine anatomical details.

First, we explored the possibilities of the increased resolution achieved by

ZOOPPA in classical tensor-based streamline tractography. For this purpose, we

chose the pons. This brainstem region is known to contain several interdigitated small

fiber bundles. The acquired dMRI data with 1 mm isotropic resolution show minimal

distortions even in such a basal region, as can be seen in Fig. 6A and B. Color-coded

direction maps (Fig. 6C) and streamline tractography (Fig. 6D) demonstrate the

resolution of complex fiber crossings. Qualitative comparison with MR microscopy of

the same region taken from Duvernoy’s Atlas of the Human Brain Stem and

14

Cerebellum confirms that dMRI extracts realistic and relevant features of the fiber

architecture.

Second, we investigated whether and to what extent our method is capable of

imaging anisotropy in gray matter. For this purpose, we focused on imaging diffusion

anisotropy in the cortex using different local reconstruction methods. The diffusion

direction in the cortex clearly differs from the direction within the white matter. At an

isotropic resolution of 1 mm, both diffusion tensor (Fig. 7C) and ball-and-stick model

(Fig. 7D) yield clear radial anisotropy in most cortical areas with fractional anisotropy

values of up to 0.4 (see a myelin stained cortex section taken from Braitenberg 1962

in Fig. 7B for qualitative comparison). Voxels at the gray-white matter interface show

reduced anisotropy (highlighted in one position with an arrow in Fig. 7C), possibly due

to partial volume effects. Note the resolution of fiber crossings in the white matter by

the ball-and-stick model (Fig. 7D). Fiber orientation density functions (Fig. 7E)

demonstrate even finer detail of intracortical diffusion patterns across different layers

within the cortex.

Third, we demonstrated the ability of the ZOOPPA technique to resolve complex

fiber crossings within the white matter. Here, not only the ultra-high spatial resolution

(800 µm isotropic resolution), but also the high angular resolution (60 directions) is of

relevance. We focus on the triple crossing area in the internal capsule, where fibers

of the corpus callosum (CC), the cortico-spinal tract (CST) and the superior

longitudinal fasciculus (SLF) intersect in a complex way. Figure 8 highlights the fact

that all three fiber bundles can be followed through the crossing area showing fine

details of interdigitated sub-bundles and the fan-like arrangement of the fiber

streamlines.

Finally, we studied the possibility of reconstructing fiber tracts forming the

interface between gray and white matter. These fibers are particularly important as

they determine the connectivity between particular gray matter areas and the white

matter fiber bundles. In Fig. 9, we show that sub-millimeter (800 µm) ZOOPPA data

15

can be used to track white matter fibers into the cortex. For a qualitative comparison,

see a myelin-stained section of a gyrus (inset in Fig. 9A, taken from Braitenberg

1962).

4. Discussion

4.1 ZOOPPA – substantial improvement in image quality

Fast MRI acquisition techniques in conjunction with substantial progress in the

development of MR hardware, such as stronger and faster gradients, have reduced

the acquisition time of a two dimensional image of the human body from several

minutes to a few seconds, or in some cases, fractions of a second. This corresponds

to an acceleration factor of several hundred. Compared to this, the acceleration

factors achievable with parallel imaging seem to be small. However, one has to keep

in mind two facts: First, a factor of two faster imaging due to improved gradient

performance requires double the gradient strength and a gradient rise time that is four

times faster. Second, parallel imaging is used in addition to the performance of the

gradient system. In this study, we used 95% of the maximum gradient amplitude for

dMRI. Even with the high performance gradient system used, without the acceleration

due to ZOOPPA, the echo time of the dMRI acquisition with 0.8 mm isotropic

resolution would increase from 65 ms to 149 ms. Such a long echo time would result

in severe signal losses due to the shortened T2 relaxation time at 7 Tesla compared

to 3 Tesla, spoiling any SNR benefit obtained from the higher field strength.

Therefore, the main objective for dMRI at ultra-high field strength is to shorten the

echo time. The other challenge with high resolution EPI is to minimize susceptibility-

induced geometric distortions. Since this effect scales with the field strength, we need

higher AFs at 7 Tesla as compared to 3 Tesla in order to compensate for this. Thus, it

is obvious that acceleration techniques such as ZOOPPA are playing an even more

important role in imaging at ultra-high field strength.

16

Wiesinger et al. (2004) showed that under certain assumptions, higher parallel

image AFs can be achieved at higher field strength. However, EPI acquisitions at 7

Tesla with acceleration factors greater than three are still problematic and tend not to

be useful, due to the increasing g-factor penalty, resulting in severe SNR losses. As

shown in Fig. 3, an AF = 4, achieved using GRAPPA only (Fig. 3A), is sub-optimal.

Conventional GRAPPA acquisitions suffer from a higher g-factor compared to the

ZOOPPA approach (compare Figs. 3C and 3D). The reduced g-factor of the

ZOOPPA approach is also reflected in an increased SNR. The ZOOPPA acquisition

shows a 25% higher SNR compared to conventional GRAPPA for the phantom

acquisitions performed in this study (see Table 2). In general, as shown in Eq. 3, we

expect a maximum SNR gain with ZOOPPA compared to conventional parallel

imaging, given by the ratio of the g-factors.

For practical application of the proposed method it is relevant that the voxels are

cubic. Although dMRI studies with highly anisotropic image voxels have been

performed, this strategy is not suited to tractography in most parts of the brain. Only

in regions like the hippocampus for example (as in Yassa et al., 2010), it might be

advantageous to use thicker slices when the slices are exactly orthogonal to the

structure, and major changes are only expected along in-plane directions. However,

the convoluted structure of the cerebral cortex and inhomogeneous white matter

structures (white matter voxels contain more than one population of axonal fibers)

make it important to use high resolution isotropic voxels. Besides the reduced partial

volume effect when thinner slices are used, an additional advantage of thinner slices

is that intravoxel dephasing, due to through-plane gradients, is reduced (Howseman

et al., 1999, Weiskopf et al., 2007). These dephasing effects can cause severe signal

dropouts in EPI.

A further important aspect of the in vivo examples shown here is the acquisition

time. With about one hour total acquisition time, the ZOOPPA dMRI experiments

could be affected by hardware issues as well as physiological issues. A test run of the

800 µm isotropic resolution ZOOPPA protocol with the SC72 gradient system for 90

17

minutes did not show any hardware constraints, such as measurement interrupts due

to heating of the gradient coil. The one hour protocol used in this work is tolerable for

healthy volunteers. However, a shorter acquisition time would be beneficial not only

for larger studies or patient studies, but also to enable the acquisition of other

applications, such as high resolution anatomical and functional scans within the same

scan session. One reason for the long scan time is simply the resolution. Due to the

very high resolution, averaging is necessary to address the SNR issue which is

demonstrated in Fig. 4D. Besides averaging, another reason for the long scan time is

the relatively long repetition time (TR). Due to SAR limitations TR has to be increased

from a minimum TR of 12 s to 14 s (1 mm protocol as listed in Table 2 middle). The

OVS pulse adds essentially the same amount of SAR than a 180° pulse.

Furthermore, the duration of the OVS pulse adds another 3 s to the TR. In summary,

the use of OVS prolongs the total scan time (assuming 4 averages) by about 24 min.

In the current implementation of ZOOPPA, for each slice an OVS pulse is played out.

Here, SAR could be reduced by a factor of two by using OVS only for every second

slice. Another potential solution would be to combine the ZOOPPA approach with a

simultaneous multi-slice excitation and readout technique, such as CAIPIRINHIA

(Breuer et al., 2005). This method allows a reduction of the repetition time, and

therefore the acquisition time, by a factor of three, while showing very promising

results for EPI acquisitions (Setsompop et al., 2011).

The proposed ZOOPPA approach of combining parallel imaging and zoomed

imaging is not limited to the use of GRAPPA and OVS with adiabatic RF pulses, as

used in this study. In fact, other combinations as for example GRAPPA with 2D RF

pulses may have advantages in terms of a more homogeneous excitation at 7 Tesla.

In general, for ZOOPPA, any combination of a parallel imaging technique with a

reduced FOV approach can be used.

4.2 Tractography with high resolution ZOOPPA data – what can be gained?

18

In tractography, it is assumed that the diffusion MR signal accurately reflects the

true fiber orientations at each point. This assumption is clearly a simplification, as its

literal fulfillment would require a number of very strict conditions to be met: (1) the

diffusion time is sufficiently long for the molecules to hit the relevant boundaries and,

if no such boundaries exist, travel far enough to dephase and cause signal loss; (2)

the sampling of the signal is sufficiently dense, in angular space (diffusion directions),

in radial space (diffusion lengths) and in time; and (3) the voxel size is in the order of

magnitude of the relevant structures, i.e., the axon diameter. It is evident that these

conditions cannot be met in practice, leading to imperfections in fiber track

reconstruction. Regarding condition (1), diffusion times and gradient areas are usually

chosen such that in the absence of boundaries, significant signal loss occurs, and this

signal loss is reduced by boundaries in distances that correspond to typical axonal

diameters (approximately between 0.1 and 5 µm; personal communication, Almut

Schüz, Tübingen, Germany) and packing densities (approx. 380000 fibers per mm²,

Aboitiz et al., 1992). For example, free diffusion in water at body temperature leads to

an average displacement of 25 µm for a diffusion time of 30 ms. Much less perfectly

condition (2) can be fulfilled: in human in vivo dMRI, we often measure with a number

of uniformly distributed gradient directions and sample the diffusion propagator with a

single combination of gradient area and diffusion time (b-value). In specialized

settings, in particular in animal or post-mortem studies, multiple b-values (Assaf and

Basser, 2005; Wedeen et al., 2005) or even independent variation of diffusion time

and gradient strength (Assaf et al., 2008) can also be applied. However, even with

these settings, it remains a discrete sampling of the diffusion propagator. This leads

to significant blurring of the angular profile of fiber directions. With respect to

condition (3), we realize that, even if the other two conditions were perfectly met, the

finite voxel size remains a serious obstacle to accurate fiber reconstruction. Even if

every fiber orientation within a voxel is identified with complete accuracy, we do not

know which points on the voxel boundaries these fibers connect. In other words, the

angular profile of fiber orientations is averaged over the voxel volume. Partial volume

19

effects with other structures with completely different diffusion properties, for example

ventricles, can also cause severe errors.

Any gain in echo time and therefore SNR can be invested in improvements of the

diffusion propagator sampling (more b-values and/or gradient directions) or

refinement of the voxel resolution, or some tradeoff between both. It has been shown

that using high q-space sampling (515 gradients) in combination with relatively low

spatial resolution (2.75 mm isotropic) enables the resolution of fine details, in

particular fiber crossings in both white and gray matter areas (Wedeen et al., 2008).

On the other hand, in this work we show that more modest sampling of the q-space

(only one radius, 60 directions) combined with very high voxel resolution (1.0 and 0.8

mm isotropic) yields similar results, at least in crossings of large fiber bundles.

Therefore, it might be instructive to consider what could be achieved by pushing

further in either direction, assuming, for the sake of argument, that SNR does not

impose limitations. Even perfect (infinite) sampling of diffusion direction and b-value

with finite voxel size would still lead to an average propagator. In contrast, if the voxel

size were reduced to the typical axon diameter, even very crude sampling of the

diffusion propagator would result in an accurate detection of the fiber direction (as

there is only one fiber left in the voxel) and would enable, in principle, perfect

tractography. Obviously, it is not clear how these extreme (and unattainable) cases

translate into more realistic situations and, beyond doubt, optimal imaging schemes

will in practice always include some compromise between k-space and q-space

sampling. However, our results seem to indicate that improvement in voxel resolution

is a fairly potent means of improving the reconstruction power of tractography. This

might be related to the fact that, in many parts of the brain, fiber architecture is more

complicated than simple crossing of major fiber tracts. Instead, fiber bundles and

separating gray matter structures often have a very small spatial extent. When using

large voxels, even with dense sampling of the diffusion propagator, severe

misidentifications of fiber bundles can be the result. For example, the fiber systems of

the external and extreme capsules are separated by the claustrum with a thickness of

20

little more than 1 mm in many places. With voxels sizes of 2 mm or more, partial

volume effects would lead to a mixture of connections of the two capsule systems, no

matter how accurately the mean fODF of the voxel is reconstructed. Similar problems

may occur everywhere in the brain, where the extent of relevant structures is in or

below the order of magnitude of the voxels, which is demonstrated with the example

shown in Fig. 4A-C. Obviously, more anatomical details are visible and a better

delineation of the cortex is possible in the color coded direction maps based on the

data with 1.0 mm compared to 1.5 mm isotropic resolution.

The methodology employed for local modeling as well as for fiber tractography in this

work comprises standard state-of-the-art techniques. We used streamline tracking on

the basis of either single tensors or multi-compartment models (ball-and-stick) or

fODFs computed using spherical deconvolution. In this way, it is possible to highlight

the specific benefit of the imaging method.

The triple crossing between the three major fiber bundles of the corona radiata, the

superior longitudinal fascicle and the projections of the corpus callosum has been

used by many authors to demonstrate the ability of their algorithms to resolve fiber

crossings (Tuch et al., 2003; Kaden et al., 2007; Wedeen et al; 2008; Kaden et al.,

2008; Descoteaux et al., 2009). Here, we show that in spite of the small voxel size,

the data have sufficient SNR to reconstruct this crossing, using streamline

tractography based on spherical deconvolution fODFs (Fig. 7). Comparable to other

techniques, we were not only able to follow all three fiber bundles through the

crossing area, but also to see how the corpus callosum fibers start to fan out in the

crossing area.

A more significant challenge is posed by crossings and interdigitations of small fiber

bundles, such as in the brain stem. By using the simple tensor as a basis for the

reconstruction of the pontine fiber bundles, we could demonstrate that with a

sufficiently small voxel size, it is possible to resolve fiber crossings composed of small

interdigitated bundles, even with this most simple approximation of the diffusion

21

propagator (Fig. 6). When using larger voxels, more complex local models, such as

DSI (see Wedeen et al., 2008; Fig. 8), are needed to achieve comparable results

(although it is not even clear if the fine structure of the interdigitated bundles can be

reconstructed at all). Please note that the insets in Figs. 6, 7 and 9 showing examples

of MR microscopy (Fig. 6C) and Weigert stain (Figs. 7B and 9A) only serve to

demonstrate the correspondence between the dMRI results and to the gross

anatomy.

An important test ground for tractography algorithms is the identification of fiber

pathways in gray matter and at the interface between gray and white matter. Both are

particularly important for the estimation of anatomical connectivity. Here, we

demonstrate that the radial structure of the cortex as well as the fibers entering the

cortex can be imaged in fine detail. While using a simple tensor already allows the

identification of the radial structure in the cortex (Fig. 7C), using more sophisticated

local modeling allows us to distinguish between different layers in the cortex with

different properties (Fig. 7E) as well as to track fibers from the white matter into the

cortex (Fig. 9). This requires high spatial resolution. Note that for reconstruction of

fine fiber crossings in the cortex, a mere increase in spatial resolution is not sufficient

(at least within practical bounds), as fibers do not usually cross as interdigitated

bundles, but at a much finer spatial scale (Wedeen et al., 2008).

Although our results must be seen as a “low resolution” approximation of the true

fiber architecture in white and gray matter, they clearly indicate that using the gain in

SNR due to the use of ultra-high field strength in conjunction with ZOOPPA is well

invested in the increase of voxel resolution.

Conclusion

The synergetic effect of using diffusion MRI with ZOOPPA at ultra-high field

strength with high performance gradients enables accelerated EPI acquisitions with

minimal artifacts and a high SNR. As a result, high angular resolution data with an

22

isotropic resolution down to 800 µm can be obtained. We show that the method is

powerful enough to reconstruct large scale fiber crossings (as many other methods

have likewise proved to be able to do), to resolve fine interdigitated fiber bundles in

the brain stem, to resolve consistent fiber orientation as well as layered structure in

the cortex and to resolve the gray-white matter interface by tracking white matter

fibers into the cortex. By virtue of our results as well as theoretical considerations, we

argue that using the gain in SNR provided by our method is well invested in the

increase of voxel resolution.

6. Acknowledgements

We would like to thank Felix Breuer of the Research Center Magnetic Resonance

Bavaria (MRB) in Würzburg and Josef Pfeuffer, David Porter, Keith Heberlein, Stefan

Huwer and Heiko Meyer of Siemens Healthcare in Erlangen for their support and

technical contributions. Furthermore, we wish to thank the following people at the Max

Planck Institute in Leipzig: Robert Trampel, Dimo Ivanov, Cornelius Eichner, Elisabeth

Wladimirow and Domenica Wilfling. Finally, we wish to thank Fabrizio Fasano of the

University of Parma.

Part of this work is supported by the FET project CONNECT of the European Union

(www.brain-connect.eu).

23

REFERENCES

Aboitiz, F., Scheibel, A. B., Fisher, R. S., Zaidel, E., 1992. Fiber composition of the human corpus callosum. Brain. Res. 598, 143-153.

Assaf, Y., Basser, P. J., 2005. Composite hindered and restricted model of diffusion (CHARMED) MR imaging of the human brain. Neuroimage 27, 48-58.

Assaf, Y., Blumenfeld-Katzir, T., Yovel, Y., Basser, P. J., 2008. AxCaliber: A method for measuring axon diameter distribution from diffusion MRI. Magn. Reson. Med. 59, 1347-1354.

Behrens, T. E. J., Woolrich, M. W., Jenkinson, M., Johansen-Berg, H., Nunes, R. G., Clare, S., Matthews, P. M., Brady, J. M., Smith, S. M., 2003. Characterization and propagation of uncertainty in diffusion-weighted MR imaging. Magn. Reson. Med. 50, 1077-1088.

Behrens, T.E.J., Johannsen Berg, H., Jbabdi, S., Rushworth, M.F.S., Woolrich. M.W., 2007. Probabilistic diffusion tractography with multiple fibre orientations: What can we gain? Neuroimage 34, 144-155.

Braitenberg, V., 1962. A note on myeloarchitectonics. J. Comp. Neurol. 118, 141-156.

Breuer, F.A., Kannengiesser, S.A.R., Blaimer, M., Seiberlich, N., Jakob, P.M., Griswold, M.A., 2009. General formulation for quantitative G-factor calculation in GRAPPA reconstructions. Magn. Reson. Med. 62, 739-746.

Descoteaux, M., Deriche, R., Knösche, T. R., Anwander, A., 2009. Deterministic and

probabilistic tractography based on complex fibre orientation distributions. IEEE Trans Med Imaging 28, 269-286.

Duong, T.Q., Yacoub, E., Adriany, G., Hu, X., Ugurbil, K., Vaughan, T., Merkle, H., Kim, S-G., 2002. High-resolution spin-echo BOLD, and CBF fMRI at 4 and 7 T. Magen. Reson. Med. 48, 589-593.

Eichner, C., Ivanov, D., Feiweier, T., Anwander, A., Knösche, T., Turner, R., Heidemann. R.M., 2011. High-Resolution Diffusion-Weighted Imaging at 7T – Issues of Slice Thickness and Fat Suppression. In Proceedings of the 17th Scientific Meeting OHBM, Quebec, #560.

Feinberg, D.A., Hoenninger, J.C., Crooks, L.E., Kaufman, L., Watts, J.C., Arakawa, M., 1985. Inner volume MR imaging: technical concepts and their application. Radiology 156, 743–747.

Fillard, P., Pennec, X., Arsigny, V., Ayache, N., 2007. Clinical DT-MRI estimation, smoothing, and fiber tracking with log-Euclidean metrics. IEEE Trans. Med. Imaging 26, 1472-1482.

24

Finsterbusch, J., 2009. High-resolution diffusion tensor imaging with inner field-of-view EPI. J. Magn. Reson. Imaging 29, 987-993.

Firbank, M.J., Coulthard, A., Harrison, R.M., Williams, E.D., 1999. A comparison of two methods for measuring the signal to noise ratio on MR images. Phys. Med. Biol. 44, N261-264.

Griswold, M.A., Jakob, P.M., Chen, Q., Goldfarb, J., Manning, W.J., Edelman, R.R., Sodickson, D.K., 1999. Resolution enhancement in single-shot imaging using simultaneous acquisition of spatial harmonics (SMASH). Magn. Reson. Med. 41, 1236-1245.

Griswold, M.A., Jakob, P.M., Heidemann, R.M., Nittka, M., Jellus, V., Wang, J., Kiefer, B., Haase, A., 2002. GeneRalized Autocalibrating Partially Parallel Acquisitions (GRAPPA). Magn. Reson. Med. 47, 1202-1210.

Griswold, M.A., 2004. Advanced k-space techniques. In: Book of abstract: 2nd International Workshop on Parallel Imaging, Zuerich, Switzerland, 16-18.

Heidemann, R.M., Fasano, F., Vogler, M., Leuze, C., Pfeuffer, J., Turner, R., 2008. Improving Image Quality by Combining Outer Volume Suppression and Parallel Imaging: zoomed EPI with GRAPPA at 7T. In Proceedings of the 16th Scientific Meeting ISMRM, Toronto, p. 1284.

Heidemann, R.M., Feiweier, T., Anwander, A., Fasano, F., Pfeuffer, J., Turner, R., 2009. High Resolution Single-Shot Diffusion Weighted Imaging with a Combination of Zoomed EPI and Parallel Imaging. In Proceedings of the 17th Scientific Meeting ISMRM, Honululu, p. 2736.

Howseman, A.M., Grootoonk, S., Porter, D.A., Ramdeen, J., Holmes, A.P., Turner, R., 1999. The effect of slice order and thickness on fMRI activation data using multislice echo-planar imaging. Neuroimage 9, 363-376.

Hwang, T.S., van Zijl, P.C.M., Garwood, M., 1999. Asymetric Adiabatic Pulses for NH Selection. J. Magn. Reson. 138, 173-177.

Ivanov, D., Schäfer, A., Streicher, M., Heidemann, R.M., Trampel, R., Turner, R., 2010. A Simple Low-SAR Technique for Chemical-Shift Selection with High Field Spin-Echo Imaging. Magn. Reson. Med. 64, 319-326.

Jansons, K. M., Alexander, D. C., 2003. Persistent angular structure: new insights from diffusion magnetic resonance imaging data. Inverse Problems 19, 1031-1046.

Jones, D.K., 2008. Studying connections in the living human brain with diffusion MRI. Cortex 44, 936-952.

Kaden, E., Knösche, T. R., Anwander, A., 2007. Parametric spherical deconvolution: inferring anatomical connectivity using diffusion MR imaging. Neuroimage 37, 474-488.

25

Kaden, E., Anwander, A., Knösche, T. R., 2008. Variational inference of the fiber orientation density using diffusion MR imaging. Neuroimage 42, 1366-1380.

Karampinos, D. C., Van, A. T., Olivero, W. C., Georgiadis, J. G., Sutton, B. P., 2009. High-resolution diffusion tensor imaging of the human pons with a reduced field-of-view, multishot, variable-density, spiral acquisition at 3 T. Magn. Reson. Med. 62, 1007-1016.

Le Roux, P., Gilles, R.J., McKinnon, G.C., Carlier, P.G., 1998. Optimized outer volume suppression for single-shot fast spin-echo cardiac imaging. J. Magn. Reson. Imaging 8, 1022–1032.

Lohmann, G., Bohn, S., Müller, K., Trampel, R., Turner, R., 2010. Image restoration and spatial resolution in 7-tesla magnetic resonance imaging. Magn. Reson. Med. 64, 15-22.

Mansfield, P., Ordidge, R.J., Coxon, R., 1988. Zonally magnified EPI in real time by NMR. J. Phys. E.: Sci. Instrum. 21, 275–280.

McNab, J. A., Jbabdi, S., Deoni, S. C. L., Douaud, G., Behrens, T. E. J., Miller, K. L., 2009. High resolution diffusion-weighted imaging in fixed human brain using diffusion-weighted steady state free precession. Neuroimage 46, 775-785.

Miller, K. L., Stagg, C. J., Douaud, G., Jbabdi, S., Smith, S. M., Behrens, T. E. J., Jenkinson, M., Chance, S. A., Esiri, M. M., Voets, N. L., Jenkinson, N., Aziz, T. Z., Turner, M. R., Johansen-Berg, H., McNab, J. A., 2011. Diffusion imaging of whole, post-mortem human brains on a clinical MRI scanner. Neuroimage 57, 167-181.

Morelli, J.N., Runge, V.M., Feiweier, T., Kirsch, J.E., Williams, K.W., Attenberger, U.I., 2010. Evaluation of a Modified Stejskal-Tanner Diffusion Encoding Scheme, Permitting a Marked Reduction in TE, in Diffusion-Weighted Imaging of Stroke Patients at 3 T. Invest. Radiol. 45, 29-35.

Mori, S., van Zijl, P. C. M., 2002. Fiber tracking: principles and strategies - a technical review. NMR Biomed 15, 468-480.

Park, H.W., Kim, D.J., Cho, Z.H., 1987. Gradient reversal technique and its applications to chemical-shift-related NMR imaging. Magn. Res. Med. 4, 526-536.

Pauly, J., Nishimura, D., Macovski, A., 1989. A k-Space Analysis of Small-Tip-Angle Excitation. J. Magn. Reson. 81, 43-56.

Pfeuffer, J., van de Moortele, P.F., Yacoub, E., Shmuel, A., Adriany, G., Andersen, P., Merkle, H., Garwood, M., Ugurbil, K., Hu, X., 2002. Zoomed Functional Imaging in the Human Brain at 7 Tesla with Simultaneous High Spatial and High Temporal Resolution. Neuroimage 17, 272-286.

Pruessmann, K.P., Weiger, M., Scheidegger, B., Boesiger, P., 1999. SENSE: sensitivity encoding for fast MRI. Magn. Reson. Med. 42, 952-962.

26

Roebroeck, A., Galuske, R., Formisano, E., Chiry, O., Bratzke, H., Ronen, I., Shik Kim, D., Goebel, R., 2008. High-resolution diffusion tensor imaging and tractography of the human optic chiasm at 9.4 T. Neuroimage 39, 157-168.

Sarlls, J. E., Pierpaoli, C., 2009. In vivo diffusion tensor imaging of the human optic chiasm at sub-millimeter resolution. Neuroimage 47, 1244-1251.

Setsompop, K., Gagoski, G.A., Polimeni, J., Witzel, T., Wedeen, V.J., Wald, L.L., 2011. Blipped-Controlled Aliasing in Parallel Imaging for Simultaneous Multislice Echo Planar Imaging With Reduced g-Factor Penalty. Magn. Reson. Med. DOI: 10.1002/mrm.23097.

Tournier, J.-D., Calamante, F., Gadian, D. G., Connelly, A., 2004. Direct estimation of the fiber orientation density function from diffusion-weighted MRI data using spherical deconvolution. Neuroimage 23, 1176-1185.

Tournier, J-D., Calamante, F., Connelly, A., 2007. Robust determination of the fibre orientation distribution in diffusion MRI: Non-negativity constrained super-resolved spherical deconvolution. Neuroimage 35, 1459-1472.

Tuch, D. S., Reese, T. G., Wiegell, M. R., Makris, N., Belliveau, J. W., Wedeen, V. J., 2002. High angular resolution diffusion imaging reveals intravoxel white matter fiber heterogeneity. Magn. Reson. Med. 48, 577-582.

Tuch, D. S., Reese, T. G., Wiegell, M. R., Wedeen, V. J., 2003. Diffusion MRI of complex neural architecture. Neuron 40, 885-889.

Tuch, D. S., 2004. Q-ball imaging. Magn. Reson. Med. 52, 1358-1372.

Wedeen, V. J., Hagmann, P., Tseng, W.-Y. I., Reese, T. G., Weisskoff, R. M., 2005. Mapping complex tissue architecture with diffusion spectrum magnetic resonance imaging. Magn. Reson. Med. 54, 1377-1386.

Wedeen, V. J., Wang, R. P., Schmahmann, J. D., Benner, T., Tseng, W. Y. I., Dai, G., Pandya, D. N., Hagmann, P., D'Arceuil, H., de Crespigny, A. J., 2008. Diffusion spectrum magnetic resonance imaging (DSI) tractography of crossing fibers. Neuroimage 41, 1267-1277.

Weinstein, D., Kindlmann, G., Lundberg, E., 1999. Tensorlines: advection-diffusion based propagation through diffusion tensor fields. Proceedings of the IEEE Visualization Conference, 249-530.

Weiskopf, N., Hutton, C., Josephs, O., Turner, R., Deichmann, R., 2007. Optimized EPI for fMRI studies of the orbitofrontal cortex: compensation of susceptibility-induced gradients in the readout direction. MAGMA 20, 39-49.

Wheeler-Kingshott, C.A.M., Parker, G.J.M., Symms, M.R., Hickman, S.J., Tofts, P.S., Miller, D.H., Barker, G.J., 2002. ADC Mapping of the Human Optic Nerve: Increased Resolution, Coverage, and Reliability With CSF-Suppressed ZOOM-EPI. Magn. Reson. Med. 47, 24-31.

27

Wiesinger, F., Van de Moortele, P.-F., Adriany, G., De Zanche, N., Ugurbil, K., Pruessmann, K.P., 2004. Parallel imaging performance as a function of field strength--an experimental investigation using electrodynamic scaling. Magn. Reson. Med. 52, 953-964.

Wilm, B.J., Svensson, J., Henning, A., Pruessmann, K.P., Boesiger, P., Kollias, S.S., 2007. Reduced Field-of-View MRI Using Outer Volume Suppression for Spinal Cord Diffusion Imaging. Magn. Reson. Med. 57, 625-630.

Yassa, M. A., Muftuler, L. T., Stark, C. E. L., 2010. Ultrahigh-resolution microstructural diffusion tensor imaging reveals perforant path degradation in aged humans in vivo. Proc. Natl. Acad. Sci. USA 107, 12687-12691.

28

FIGURE CAPTIONS

Fig. 1: Depiction of how zoomed imaging and parallel imaging are combined (ZOOPPA) to achieve a high acceleration factor: (A) Non-accelerated, 100% FOV. (B) Zoomed accelerated, 50% FOV with OVS, AFZOOM = 2 without aliasing. (C) ZOOPPA accelerated, 17% FOV. The reduced FOV is further reduced by a factor of three. The overall AF is AFZOOM multiplied by AFPPA, which results in overall AF = 6. The parallel image reconstruction with AFPPA = 3 will unfold this image to 50% FOV as in (B). Fig. 2: ZOOPPA acquisition and final image: (A) An OVS band is placed to null the signal outside the region of interest (indicated by an orange box). The FOV can be reduced to 50% FOV (AFZOOM = 2) without aliasing. The FOV is further reduced by AFPPA = 3, resulting in a FOV of 17% (indicated by the gray box). This translates into overall AF = 6. (B) The ZOOPPA image: The parallel image reconstruction with AFPPA = 3 unfolds the aliased image to 50% FOV. (C) For comparison, an image obtained with overall AF = 3 using GRAPPA is shown. This image is cropped to match the coverage of the ZOOPPA image. Fig. 3: Phantom study to compare SNR and g-factor maps of conventional GRAPPA to GRAPPA with zoomed imaging (ZOOPPA). (A) Factor of four accelerated GRAPPA acquisition. (B) Factor of four accelerated ZOOPPA acquisition. Here, AF

PPA = 2 and

AFZOOM

= 2, resulting in overall AF = 4. Corresponding g-factor maps for (C) GRAPPA

and (D) ZOOPPA, both with overall AF = 4. Fig. 4: Comparison between different resolutions and number of averages: (A and C) Sections of an axial color coded directional map with 1.5 mm isotropic resolution (A) and with 1.0 mm isotropic resolution (B). (C) Corresponding enlarged sections, as indicated by the white boxes in (A) and (B), showing the fractional anisotropy values. Profiles, indicated by white lines are plotted in the middle of (C). (D) The effect of averaging on the results obtained from data with 0.8 mm isotropic resolution. (left) Without averaging, (middle) 4-times averaging and (right) 4-times averaging and images restoration procedure. Fig. 5: (left column) Trace-weighted images of a ZOOPPA acquisition with 1 mm isotropic resolution. The coronal (top), sagital (middle) and axial (bottom) view show the coverage which can be achieved with this protocol acquiring 94 slices. (middle column) Trace-weighted images overlaid with the gray/white matter boundaries from a corresponding anatomical scan. (right column) Three additional axial slices from different regions of the brain. Fig. 6: Imaging of the pons and cerebellar region. (A and B) Main diffusion directions overlaid onto the T1 anatomy in a sagittal slice. The images demonstrate the low

29

distortion level of the ZOOPPA images. Note the fine level of anatomical details in the image, for example, the radial diffusivity in the cortex and the separate lamina in the pons. (C) Color-coded fractional anisotropy map in the same slice. The texture indicates the fiber orientation. This map resolves sub-parts of the fine laminar medio-lateral oriented structures (red) in the pons dividing ventro-dorsal oriented fibers (blue) into the fronto-pontine tract (F), the cortico-spinal tract (P) and the temporo-parieto-pontine fibers (PT). For qualitative comparison, see inset in (C), showing a similar slice in MR microscopy at 9.4 T (Duvernoy’s Atlas of the Human Brain Stem and Cerebellum, 2009, Fig. 8.46, modified). (D) Streamline tractography showing the separation of interdigitated medial-lateral and inferior-superior fiber bundles. Fig. 7: Radial cortical anisotropy in the trans-occipital sulcus shown in three different participants with three different methods at the same nominal resolution. (A) Anatomical reference showing the coronal slice (Talairach y=-73) with the depicted region. (B) Myeloarchitecture of human cortex (Weigert stain, from Braitenberg 1962) showing radial anisotropy. (C) Principle diffusion directions based on the tensor model overlaid on the color-coded fractional anisotropy map. (D) Diffusion directions computed by a multiple compartment model (ball and two sticks). E: Fiber orientation density functions computed by spherical deconvolution. In all three examples, a clear radial asymmetry in the cortex is observed. Fig. 8: Streamline tracking in fiber crossing area with sub-millimeter isotropic resolution (800 µm) on top of fiber orientation density plots. The streamlines are color-coded according to their local orientation. (A) Coronal section of the brain with crossings of corpus callosum (CC, red), cortico-spinal tract (CST, blue) and superior longitudinal fascicle (SLF, green). (B) Enlarged section of the region indicated in (A). Fig. 9: Fiber orientation density plots (A) and streamline tracking (B) of sub-millimeter isotropic resolution (800 µm) dMRI data in a horizontal slice through parietal lobe, showing white matter fiber tracts entering the cortex. See inset for myeloarchitecture of a gyrus for qualitative comparison (Weigert stain, from Braitenberg 1962). Fig. 10: Supplementary material: A single diffusion direction after motion correction and 4-times averaged. Here, only every second slice is shown.

Isotr. resol. [mm]

Grad. system

TR [ms]

TE [ms]

Partial-Fourier

FOV [mm2]

Slices Av. Scan time [min]

AF=AFZOOM *AFPPA

1.0 AS95 9500 72 5/8 141x191 71 6 69 4.4 = 1.45 * 3.0

1.0 SC72 14000 55 6/8 144x150 109 4 66 4.2 = 1.39 * 3.0

0.8 SC72 14100 65 6/8 143x147 91 4 65 4.6 = 1.53 * 3.0

Table 1: List of imaging protocol parameters for the in vivo acquisitions with 60 diffusion directions: Protocols with 1 mm isotropic resolution have been obtained with an earlier gradient system (AS95) and with a new gradient system (SC72). Sub-millimeter isotropic resolution has been obtained with the new gradient system.

Table 1

Pixels within

ROI GRAPPA AFPPA = 4

ZOOPPA AF = 4 (AFPPA = 2 and AFZOOM = 2)

SNR ROI I 18870 37 48 (+30%)

SNR ROI II 13200 56 71 (+27%)

SNR ROI III 7000 66 84 (+27%)

SNR ROI IV 4500 61 77 (+26%)

Table 2: Mean SNR values of different regions of interest (ROIs) averaged over 4 slices obtained in an oil phantom. Conventional GRAPPA is compared to GRAPPA combined with zoomed imaging (ZOOPPA). The imaging parameters are kept the same for both experiments. ROI I is exactly the region covered with the ZOOPPA approach (half the FOV, as shown in Fig. 3B) including the background. ROIs II-IV are regions within the phantom with shrinking areas. Due to coil sensitivity variations, the SNR is locally dependent. However, for all positions investigated, the ratio is similar, indicating an SNR gain of more than 25% when ZOOPPA is used (percentage gain compared to conventional GRAPPA in brackets).

Table 2

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Supplementary Material. Figure 10