Non-Invasive Glucose Measurement Technologies: AnUpdate from 1999 to the Dawn of the New Millennium
OMAR S. KHALIL, Ph.D.
ABSTRACT
There are three main issues in non-invasive (NI) glucose measurements: namely, specificity,compartmentalization of glucose values, and calibration. There has been progress in the use ofnear-infrared and mid-infrared spectroscopy. Recently new glucose measurement methods havebeen developed, exploiting the effect of glucose on erythrocyte scattering, new photoacousticphenomenon, optical coherence tomography, thermo-optical studies on human skin, Ramanspectroscopy studies, fluorescence measurements, and use of photonic crystals. In addition tooptical methods, in vivo electrical impedance results have been reported. Some of these meth-ods measure intrinsic properties of glucose; others deal with its effect on tissue or blood prop-erties. Recent studies on skin from individuals with diabetes and its response to stimuli, skinthermo-optical response, peripheral blood flow, and red blood cell rheology in diabetes shednew light on physical and physiological changes resulting from the disease that can affect NIglucose measurements. There have been advances in understanding compartmentalization ofglucose values by targeting certain regions of human tissue. Calibration of NI measurementsand devices is still an open question. More studies are needed to understand the specific glu-cose signals and signals that are due to the effect of glucose on blood and tissue properties. Thesestudies should be performed under normal physiological conditions and in the presence of otherco-morbidities.
660
INTRODUCTION
Non-invasive (NI) glucose measurements
NI DIAGNOSIS AND MONITORING of diabetes at-tracted tremendous attention in the past 2
decades because of the emergence of diabetes asa major epidemic, especially when associatedwith the increased overall obesity of the popu-lation. NI determination of glucose will promotemore frequent testing, allows tighter control ofdiabetes, and delays the onset of diabetes com-plications and their associated health care costs.
It is now more than 15 years since the firstreports of the feasibility of NI glucose, andclose to 5 years since the spectroscopic aspectsof NI glucose determinations were reviewed,1yet there is no commercialized NI glucoseproduct yet. This paper reviews advances in NIglucose testing methods that were suggested ortested over the period between 1999 and thedawn of the New Millennium. Previously re-viewed methods will not be discussed unlessnew whole blood, animal model, or human ex-periments have been published. Other reviewarticles have covered various aspects of NI
Diagnostics Division, Abbott Laboratories, Abbott Park, Illinois.
Review
5264_e12_p660-697 9/29/04 12:48 PM Page 660
sensing of glucose.2,3 In addition to the previ-ously reviewed detection technologies,1–3 newmethods that have appeared over the past 4years will be discussed. The review will con-centrate on published, peer-reviewed papers.Discussion of methods disclosed only inpatents will be limited. This is because patentshave limited data sets and lack the detailed in-formation on experimental design and dataanalysis that are presented in peer-reviewedpapers. Methods that show high correlationswith glucose and low estimation errors will beparticularly scrutinized as they describe per-formance close to that of a viable commercialproduct.
The U.S. Food and Drug Administration mayconsider NI and minimally invasive devicesthat are intended to measure, monitor, or pre-dict blood glucose levels in persons with dia-betes to be high-risk medical devices subject topre-market approval.4 Gutman et al.4 saw thatthe technology is not yet well understood, andthe information obtained from these devices isoften different from the information that hasbeen the traditional base for the managementof diabetes; consequently the Food and DrugAdministration will require both analytical andclinical studies to support the intended claimsfor these new NI devices. Analytical studies forin vitro diagnostic devices generally includestudy of dose–response relations, determina-tion of factors affecting accuracy and precision,recovery studies, and interference studies inserum and blood samples. Analytical calibra-tion of a home glucose monitor implies deter-mination of glucose concentration from a stan-dard curve generated by using calibrators andstandards. It will be difficult to perform thesestudies for NI glucose devices where calibra-tion will be mainly based on clinical data. It isimportant to initiate discussions on the possi-bility of using data obtained on tissue-simulat-ing phantoms and simulation study results toaugment clinical data on human volunteers inproving performance of NI glucose testing de-vices. At least these studies could be used toexplain the theory of operation of the device.
There are three main issues in NI glucosemeasurements: specificity, compartmentaliza-tion of glucose values, and calibration. Three
questions arise regarding NI determination ofglucose:
1. What is actually being detected and deter-mined? Is it an intrinsic property of the glu-cose molecule, or is it the effect of change inglucose concentrations on tissue or bloodproperties?
2. In what body compartment is this glucosevalue determined? How does the deter-mined concentration relate to arterial bloodglucose concentration?
3. How to calibrate the NI testing device? Is ita single-person calibration or multiple-per-son (universal) calibration? Will the NI test-ing device be calibrated at the factory, or canthe user calibrate it? What data inputs arerequired for the calibration?
Specificity of NI glucose measurements
Methods used for the NI determination ofglucose can be classified in two broad cate-gories: as methods tracking a molecular prop-erty of glucose, or methods tracking the effectof glucose on tissue and blood properties. Thefirst category depends on tracking an intrinsicmolecular property of glucose such as near-in-frared (NIR) absorption coefficient, mid-in-frared (IR) absorption coefficient, optical rota-tion, Raman shifts, NIR photoacoustic (PA)absorption, and the like. These methods as-sume the ability to detect glucose in tissue orblood independent of other body components,and also independent of the body’s physiolog-ical state. The second set of methods dependson measuring the effect of glucose on the opti-cal properties of tissue. These properties in-clude light scattering coefficient of tissue, re-fractive index of interstitial fluid (ISF), andsound propagation in tissue.
The ability to collect reliable experimental NIglucose data faces several obstacles: (a) theminute magnitude of the measured signals, (b)repositioning error of the measuring probewith respect to the body part, (c) temperaturevariations, (d) variations in tissue physicalproperties at the probe/body interface, (e) ef-ficiency of optical and thermal coupling be-tween the probe and tissue, and (f) effect of
NON-INVASIVE GLUCOSE MEASUREMENT 661
5264_e12_p660-697 9/29/04 12:48 PM Page 661
probe/tissue interaction on signal magnitude,etc. Advances in overcoming these obstacleswill be discussed in this article.
Compartmentalization of glucose values
Glucose in the human body is found in sev-eral body fluids, such as blood, tissue ISF, eyevitreous fluid, tears, and sweat. It is distributedin different body compartments. Current pa-tient care is based on determination of clinicallysignificant concentrations of glucose using invitro invasive reference methods, which in-volve sampling of venous blood or arterializedvenous blood. Glucose is detected either inwhole blood or in the separated serum. An en-zymatic method with optical or electrochemi-cal sensor is used for subsequent detection.Several NI methods propose determination ofglucose in tissue ISF. Glucose in the ISF, or anyother body fluid, can be used as a substitute forvenous or capillary blood glucose values onlyif changes in its magnitude and duration ofchange in the blood vessels and tissue are iden-tical.5 This may not be the case when changesin blood glucose values are sudden and are toolarge in magnitude to allow for equilibrationbetween vascular and interstitial levels of glu-cose.5,6 Even for blood glucose measurements,there are site-specific effects on the magnitudeof the glucose levels.6 Equilibration betweenblood glucose values and the glucose concen-tration in other body fluids, such as the eye vit-reous fluid, the ISF, and saliva, has been a con-troversial issue with widely different reportedlag times.
In a study on the human arms as a body site,Thennadil et al.7 used the suction blister tech-nique to investigate the relationship betweenglucose levels in dermal ISF and capillary ve-nous blood in subjects with diabetes whoseblood glucose levels were manipulated so as toinduce rapid changes in blood glucose levels.Glucose levels in the three compartments ex-hibited high correlations when individual vol-unteers were considered separately, or whendata from all volunteers were combined. Nosignificant time lag was observed between ISFand either capillary or venous blood glucoselevels during the glucose excursions.7 How-ever, it was argued that the condition in the
previously described experiment allowed forslow equilibration between blood glucose andISF glucose without large sudden surges in glu-cose values in either fluid.6 Lag times betweenblood glucose levels and ISF glucose levelswere reported for implantable glucose sen-sors.8,9 ISF glucose concentration lagged be-hind blood glucose values by 4.4 � 0.8 min.The relationship between blood and ISF glu-cose was not affected by insulin. Delays in ISFglucose equilibration could be corrected withdigital filters. The authors summarized lag timebetween ISF and blood glucose values for dif-ferent technologies: These lag times variedfrom �5 min to �30 min, depending on themeasurement technology.9
Several body sites were studied or suggestedfor the NI determination of glucose, raising thepossibility of different values in body com-partments for different sites. Forearm skin,10–12
fingers,13–16 ear lobe,17,18 tongue,19 abdominaltissue,20 inner lip mucosa,21 the eye,22 and theconjunctiva23 were suggested or used as sitesfor NI glucose measurements. Different delaytimes between blood glucose values and NI-de-termined glucose have been suggested, andmay be encountered, for different body sites.
Calibration of NI measurements
Multivariate analysis is generally used for de-termination of the concentration of a compo-nent in a complex mixture.24–26 It has been usedfor data analysis of NI glucose measurementsthat are based on NIR spectra.12,15,19,21 The qual-ity of an NI measurement is judged mainly bythe magnitude of the standard error of predic-tion (SEP) and the prediction correlation coef-ficient (Rp). A low SEP, when associated with ahigh correlation coefficient, is indicative of asuccessful determination. One must guardagainst the possibility of (a) small range of con-centration of analytes, (b) chance correlationwith other time-dependent events, and (c) over-fitting the experimental data. Extreme caremust be taken to avoid the effect of overfitting.The training set and the prediction set need tobe separated in time. The number of input NIdata points, such as spectra, temperatures, dis-tances, etc., must be larger than the number ofterms in the fitting equations. Partial least
KHALIL662
5264_e12_p660-697 9/29/04 12:48 PM Page 662
squares (PLS) calibration models require use ofseveral independent standard spectra for eachfactor used in the PLS analysis.
A high SEP can be interpreted in one of threeways: (a) The calibration model is void of glu-cose-specific information, (b) glucose predic-tions are limited by measurement noise, or (c)glucose predictions are limited by biologicalbackground noise in the body part and reposi-tioning errors of the probe with respect to thebody part. Noise sources, such as circulationparameters, skin surface condition, skin watercontent, circadian rhythm effects, and temper-ature changes, are examples of biological vari-ables that have not been yet considered in suf-ficient detail.
A simple method to establish a calibrationmodel is to fit the measured NI experimen-tal data to invasively determined glucose concentrations. Linear least squares (LLS) fit-ting or PLS fitting is used. For example, a lin-ear calibration relationship can have theform11,12,15,19,21:
[Glucose] � a0 � �i ai � P (k,l,m) (1)
where [Glucose] is the concentration of glucose,ai is a regression coefficient, P is a measuredparameter, and k, l, and m are various con-straints, which may be different wavelengths,wavelengths and distances, wavelengths andtemperatures, etc. The calibration correlationcoefficient (Rc) and the standard error of cali-bration (SEC) are then determined. The cali-bration model can then be validated by check-ing its ability to predict one data point usingthe rest of the data points as the calibration set.The process is repeated for each data point inthe set. This is known as the leave-one-out crossvalidation (LOOCV). The predicted values andthe reference values are then used to calculatethe standard error of cross-validation predic-tion (SECV) and the cross-validation correla-tion coefficient (RCV). A low SECV and a highRCV indicate the validity of the calibrationmodel. These RCV and SECV values can beused, with caution, to indicate the predictionability of the model. This should be especiallythe case when only a small number of datapoints are available. As the LOOCV procedureinvolves the use of data points from the same
experimental set, it is prone to effects of spuri-ous correlations with instrument or with exper-imental time-dependent events. True predictioncan only be achieved by using independent cal-ibration and prediction data sets that have beenobtained over different time periods. The twocalculated prediction parameters are the Rpand the SEP.
A major difference between in vitro and invivo calibration experiments is that the humanbody has time-dependent physiological effectsthat are apparent in the circadian rhythm. In aresting state the cardiac pulse, blood pressure,blood circulation, respiration, body tempera-ture, and cutaneous blood flow are subject tocircadian periodicity. These parameters are attheir highest values between 8 a.m. and 12noon, and again between 4 p.m. and 7 p.m.They are at their lowest values between 1 p.m.and 3 p.m., and again between 10 p.m. and 6a.m., next morning. This regular periodicity fornormal healthy individuals is perturbed bychanges in environmental conditions, and cer-tain diseases.27,28 In the meantime, the con-centration of glucose in subjects without dia-betes and patients with in-control diabetesvaries during the day and follows a regularprofile. Generally, the morning glucose valuesare low (fasting glucose), levels increase sub-stantially after the morning meal, decrease toa pre-mid-day nadir, increase again after mid-day and evening meals, and settle to lower val-ues at night. This profile will be distorted incases of advanced diabetes. Deviation fromthis pattern requires intervention. Tight glyce-mic control can return the daily glucose fluc-tuation pattern close to that of a healthy indi-vidual. Glucose injection produced bodytemperature (vital sign) changes.29 There is apotential for a coincidental correlation be-tween the circadian fluctuation of glucose con-centration in human blood and the circadianperiodicity of the body temperature and othervital signs. It is important to separate the ef-fects of circadian body temperature and bloodflow effects on the NI measurement so they donot confound the signals used to calculateblood glucose concentration.
Calibration of NI glucose devices requires theavailability of various invasively determinedconcentrations of glucose and the correspond-
NON-INVASIVE GLUCOSE MEASUREMENT 663
5264_e12_p660-697 9/29/04 12:48 PM Page 663
ing NI signals. One calibration approach is toinduce a change in glucose concentration span-ning a range over which the in vivo measuredsignal can be monitored. This is achieved by us-ing a glucose clamp procedure,30–32 an oral glu-cose tolerance test (OGTT), or a meal tolerancetest (MTT). In the glucose clamp experiment,the concentrations of injected glucose and in-sulin are manipulated to result in a steady con-centration of either glucose (hyperglycemic orhypoglycemic clamps) or insulin (hyperinsu-linemic clamp) over a period of time. In theOGTT a known load, typically 75 g of glucose,is given to a fasting subject, and the concen-tration of glucose is followed as a function oftime. The MTT is similar to the OGTT, exceptthat the carbohydrate load is administered as ameal rather than a glucose load. Data that aregenerated during the test period can be used topredict glucose concentration from subsequentNI measurements. As the response of an NI in-strument may embody non–glucose-relatedphysiological effects, relying on calibrationbased on correlating OGTT or MTT data withNI instrument response leads to a calibrationmodel that is unique to the individual tested.This calibration model will need to be period-ically updated, using an invasive test. Time-de-pendent artifacts can influence the results frommultivariate calibrations when randomizedsampling over time cannot be performed. Inaddition to instrument-related time variables,the previously discussed circadian rhythm ofthe human body can lead to spurious time-de-pendent biological background that will be su-perimposed on the sequential MTT data points.Although OGTT and MTT data are needed toprove that a particular NI signal varies with in-duced change in glucose concentration, theyare not sufficient to establish a reliable calibra-tion of signal versus blood glucose values. Ide-ally, it is preferable to have a self-calibratingtechnology that does not require multiple in-vasive data points to calibrate the device, thecalibration of which is not unique to an indi-vidual.
A calibration method that is more immuneto time-dependent effects is a random spot testthat is performed at randomized time slots dur-ing the day and over a long period of time (lon-gitudinal study).19,33,34 This approach allows the
use of data points that cover multiple physio-logical conditions and avoids circadian rhythm-related correlations. It will avoid most instru-ment-related chance correlation.33
Frequent recalibration of a NI device with aninvasive test is a major issue. The higher thefrequency of these invasive recalibrations, thelower is the value of the NI testing device forthe patient. An NI device that depends on fre-quent invasive recalibration, even when tech-nically successful, will doubtfully be acceptedin the market. The term “universal calibration”has been touted as an achievable goal, but cur-rent sensitivity and specificity of the availablemethods preclude this assertion. After estab-lishing specificity of the test method and un-derstanding the compartmentalization of glu-cose values, which are essential to the technicalfeasibility of a test device, researchers shouldaddress calibration frequency issues, as it is im-portant to the patient’s acceptance of an NI test-ing product.
OPTICAL PROPERTIES OF TISSUES ANDHUMAN SKIN
NIR tissue optical properties
The NIR spectral region is commonly usedin most reported methods. It has several spec-tral windows where hemoglobin, melanin, andwater absorption band intensities are lowenough to allow light to penetrate in the tissue,which enables NI spectral measurements.
Attenuation of light in tissue is described, ac-cording to light transport theory, by the effec-tive attenuation coefficient �eff
35:
I � I0 e��effl (2)
where I0 is the incident light intensity, I is thereflected light intensity, l is the optical path-length in tissue, and �eff is defined as:
�eff � [3 �a (�a � �s)]12 (3)
An exact solution of light transport equationin turbid media can be modeled by followingthe path of each individual photon and calcu-lating the probability of scattering or absorp-tion in a series of steps using Monte Carlo sim-
KHALIL664
5264_e12_p660-697 9/29/04 12:48 PM Page 664
ulation.35 This modeling is used to study thepath of photons in tissues and was used for op-timization of photodynamic therapy,36 laser-Doppler flowmetry (LDF),37 and optical mam-mography.38
Light propagation in tissue is expressed bythe absorption coefficient, �a, and the scatter-ing coefficient, �s. Absorption and scatteringcoefficients of tissue are determined by per-forming one or more of transmission, diffusereflectance, localized reflectance, time domain,or frequency domain measurements.35 The ab-sorption coefficient is related to the concentra-tion of a tissue chromophore by �a � 2.303 �Ccm�1, where � is the molar extinction coeffi-cient and C is the molar concentration. Changesin glucose concentration can influence �a of tis-sue through absorption corresponding to wa-ter displacement (absorption decreases as glu-cose concentration increases), or change in itsintrinsic absorption (absorption increases asglucose concentration increases). Change in �adue to water displacement is nonspecific. The�a of glucose in the NIR is low and is muchsmaller than that of water by virtue of the largedisparity in their respective concentrations. Inthe NIR the weak glucose overtone and com-bination spectral bands overlap with otherstronger overtone and combination bands ofwater, hemoglobin, protein, and fats.1
An example of the magnitude of NIR glucoseintrinsic absorption signals is illustrated by thevalues of the extinction coefficient of glucose.Youcef-Toumi and Saptari39 determined � ofglucose at the first overtone band 1,686 nm(5,930 cm�1) and the combination band 2,273nm (4,430 cm�1). The measured value at 1,689nm was 2.3 � 10�2 M�1 cm�1, and that mea-sured at 2,257 nm was 0.4 M�1 cm�1. The val-ues are far smaller than the 6.2 � 103 M�1 cm�1
� value of NADH at 340 nm, a compound thatis usually used for the determination of serumglucose values on automated blood analyzers.Using a 1-mm pathlength, a 10 mmol/L glu-cose concentration will generate an absorbancevalue of 2.3 � 10�5 absorbance units at 1,686nm and 4 � 10�4 absorbance units at 2,257 mm.A 1-mm pathlength is longer than the path-length encountered in diffuse reflectance mea-surements, and is of comparable magnitude tothe pathlength in some spatially resolved re-
flectance measurements.10,11 The intrinsic ex-tinction coefficient of glucose will be muchlower at the higher overtone bands between800 nm and 1,300 nm, thus requiring extremelysensitive detection and elimination of sourcesof background noise, and leading to extremedifficulty in interpreting the data in this spec-tral range.39
The scattering coefficient is �s � �� [1 � g]cm�1, where � is the scattering cross section, �is the number density of scattering centers, andg is the anisotropy factor.35 Change in glucoseconcentration affects the intensity of light scat-tered by tissue. The reduced scattering coeffi-cient of a tissue or can be expressed as �s � f(�, a, nscatterer/nmedium), where � is the numberdensity of scattering centers in the observationvolume, a is the mean diameter of scatteringcenters, nscatterer is their refractive index, andnmedium is the refractive index of the sur-rounding fluid.40 For the case of cutaneous tis-sue, connective tissue fibers are the scatteringcenters. Erythrocytes [red blood cells (RBCs)]are the scattering centers for the case of blood.The effect of a solute on the refractive index ofa medium, and hence �s, is nonspecific and iscommon to other soluble analytes.
Mid-IR properties of tissue and skin
Mid-IR lies in the spectral range between 2.5�m (4,000 cm�1) and 10 �m (1,000 cm�1).Bands in this spectral range correspond mainlyto frequencies of fundamental molecular vi-brations, which are characteristic of the specificchemical bonds. While the NIR spectral rangeencompasses combinations and overtonebands that are broad and weak, bands in themid-IR are sharp and have a higher absorptioncoefficient. The mid-IR spectral bands of glu-cose and other carbohydrate have been as-signed and are dominated by C-C, C-H, and O-H stretching and bending vibrations.41 The800–1,200 cm�1 fingerprint region of the IRspectrum of glucose has bands at 836, 911,1,011, 1,047, 1,076, and 1,250 cm�1 that havebeen assigned to C-H bending vibrations.41,42
A 1,026 cm�1 band corresponds to C-O-H bendvibration.42 Spectral measurements in this fre-quency interval were used to determine of glu-cose in serum and blood. IR emission in the
NON-INVASIVE GLUCOSE MEASUREMENT 665
5264_e12_p660-697 9/29/04 12:48 PM Page 665
same frequency range was used for NI mea-surement of glucose and will be discussed later.
Tissue optical properties in the mid-IR be-tween 2.5 �m (4,000 cm�1) to 10 �m (1,000 cm�1)differ from the NIR as scattering decreases atlonger wavelengths and light attenuation by wa-ter, proteins, and fats absorption spectra domi-nates. Water constitutes 70% of hydrated tissue,followed by the connective tissue proteins andlipids. The strong absorption limits light pene-tration depth in tissue to 0.2 �m at 2.5 �m (4,000cm�1) and to 1.2 �m at a wavelength of 10 �m(1,000 cm�1).43 Attenuated total reflectance zinc-selenide prisms have been used for studies ofthe mid-IR spectra of tissue.21,43–46 Light pene-tration depth limits the sampling depth to thestratum corneum (SC). Attenuated total re-flectance (ATR) mid-IR spectroscopy was usedto study properties of the SC such as its watercontent, and effect of cleansing, stripping the SCby adhesive tape, and partial occlusion on its op-tical properties. It was also used to study diffu-sion of topical medication in the SC, and ep-ithelium layers of oral mucosa and other tissues.The results of these studies are quite importantfor the NI determination of glucose as they re-veal several important factors about skin–probeinteraction, properties of different skin sites, andeffect of skin cleaning procedures on its opticalproperties.
The mid-IR spectrum of the skin as deter-mined by ATR accessories showed overlap be-tween IR bands of glucose and those in skincomponents.43–46 Table 1 summarizes the low-
frequency vibrations in the mid-IR band fre-quencies of water, glucose, and human skin.Although bands in the 10 �m range are specificto glucose in aqueous solutions, there is a highprobability of having C-C, C-H, and C-O bend-ing vibrations from other skin components co-inciding with them. The C-C bands in the spec-trum of skin do not relate only to glucose, as itis a minor component compared with proteinsand fats.
Skin mid-IR spectrum depended on its wa-ter content, showing increase in intensity in wa-ter bands intensity upon hydration.44–46 In-crease of skin hydration was observed upontreating the skin with 70% isopropanol, occlu-sion of the skin, or tape striping of the skin.Contact between the skin and the probe causedocclusion and increased water content by trap-ping the transdermal transpired water betweenthe stratum corneum and the probe. This effectshould be also present in NIR measurements.It is noticeable in mid-IR ATR studies of skinbecause of the ability of the technique to sam-ple the SC only without interference from theunderlying layers, and because of the sharp-ness of mid-IR absorption bands.
Diabetes and microcirculation changes
Diabetes mellitus affects microcirculationand leads to microvessel complications such asneuropathy, retinopathy, and nephropathy.LDF studies showed impaired circulation inpatients with diabetes, which was manifestedby a decrease in cutaneous blood flow47 and adifference in blood flow response to cooling orwarming.48,49 Vascular walls in the cutaneousmicrovasculature of patients with diabeteswere abnormally thick,50 and blood vessels re-sponded subnormally to heat, injury, and his-tamine.51–53 Persons with type 2 diabetes orborderline glucose intolerance have stiffer ar-teries than their counterparts with normal glu-cose tolerance.54 Patients with diabetes exhib-ited differences from those without diabetes incutaneous blood flow,52–55 reduced maximalhyperemia,56 impaired peripheral vasomo-tion,57 and response to contralateral cooling.58
Endothelium-dependent vasodilatation is im-paired during acute hyperglycemia,59 and bothmicrovascular and macrovascular reactivities
KHALIL666
TABLE 1. MID-IR FREQUENCY BANDS (IN CM�1) OF
WATER, GLUCOSE, AND HUMAN SKIN IN THE FINGERPRINT
REGION OF GLUCOSE IR SPECTRUM
Watera Glucoseb Skina
836852
911 9171,0111,026 1,0351,0471,076 1,077
1,150 1,1181,164
1,250 1,245
aFrom Lucassen et al.44
bFrom Vasko et al.41,42
5264_e12_p660-697 9/29/04 12:48 PM Page 666
are reduced in persons at risk for type 2 dia-betes.60 Diabetes causes changes in cutaneousvascular response to external stimuli, whichcan precede clinical symptoms of diabetes.61
In addition to the vascular effects of diabetes,both insulin and glucose are reported to havevascular activities. Acute hyperglycemia atten-uated endothelium-dependent vasodilatationin humans.62 Insulin is reported to cause va-sodilatation in human circulation.63 In anotherstudy, acute hyperglycemia (11.78 mmol/L,212 mg/dL glucose) and hyperinsulinemia(119 �U/L insulin) enhanced vasodilatation intype 1 diabetes over a standard of 4.78 mmol/L(86 mg/dL) glucose and 22 �U/L insulin asshown in Figure 1, which is calculated from thedata presented by Oomen et al.64 If one con-siders the glucose clamp part of the experi-mental data in Figure 1, i.e., the first two setsof bars show blood flow increases with in-creasing glucose concentration.64 Comparingthe third set of bars to the first set of bars showsthat insulin has a similar effect on cutaneous
blood flow as glucose does, and hence on theLDF signals.
The vasodilatory effect of hyperinsulinemiais well established by plethymography of theforearm in both healthy volunteers and subjectswith type 1 diabetes.63,64 These effects werefound to be temperature dependent and to varyfrom site to site in the human body. The re-sulting increase in light absorption associatedwith increased perfusion could be erroneouslyattributed to, and correlated with, blood glu-cose concentration. Indeed, one patent sug-gested use of LDF for NI determination of glu-cose without presenting any experimental datato show potential performance.65 Such a propo-sition is suspect in light of the experimentaldata on the clamp experiments.64 Hemoglobinelectronic absorption spectral bands and watervibration overtone bands dominate the shortwavelength NIR spectral window of 600–1,300nm. The NIR absorbance curve of a fingershifted downwards on restriction of bloodflow. It shifted upwards with thermal stimula-
NON-INVASIVE GLUCOSE MEASUREMENT 667
FIG. 1. Results of LDF study on human skin. Data are from Oomen et al.64 and are plotted as the mean values ofglucose concentration, insulin concentration, and LDF relative to the standard state. Glucose concentrations were con-verted to mg/dL for the purpose of graph scaling. The y-axis is the relative blood flow (BF), glucose concentration[G] (in mg/dL), or insulin concentration (in mU/L). The x-axis is standard insulin/standard glucose (SI-SG), stan-dard insulin/high glucose (SI-HG), high insulin/standard glucose (HI-SG), or high insulin/high glucose (HI-HG).Standard insulin concentration was 22 mU/L, standard glucose concentration was 86 mg/dL (4.78 mmol/L), high in-sulin concentration was 119 mU/L, and high glucose concentration was 212 mg/dL (11.78 mmol/L).
5264_e12_p660-697 9/29/04 12:48 PM Page 667
tion and with post-occlusion hyperemia.66 Theblood flow-induced shifts in the absorbancecurve were particularly pronounced in therange of 850–970 nm. LDF-determined skinblood flow correlated with the absorbance val-ues, Rc � 0.69.66 NIR signals between 590 and1,000 nm correlated with hemoglobin concen-tration or hematocrit values, resulting in goodcalibration and prediction errors for either ofthe two blood parameters.67,68
A sudden rise and fall of glucose or insulinconcentration, as in the case of the clamp ex-periment, will lead to vasodilatation and in-creased perfusion in the outer vascular bed.Blood perfusion will contribute to change intransmission and reflectance signals, especiallyin the short wavelength NIR range �1,300 nm.Blood perfusion effects can confound correlat-ing the NIR signal change with glucose con-centration. This assertion suggests that absorp-tion-based methods, which depend on directmeasurement of transmission and reflectancein the short wavelength NIR range, withoutconsidering perfusion, are not likely to presenta viable approach to NI glucose measurements.Correlating glucose concentration with the op-tical signal in an MTT experiment assumes thatglucose concentration is the only time-depen-dent variable. Other factors, such as tempera-ture, blood perfusion, tissue compression,blood oxygenation, cutaneous water, and othermetabolites or medications that affect tissueblood dynamics, are not usually considered inanalyzing NI signals, though they may domi-nantly affect the measured signal change.
Diabetes and change in tissue (skin) structuralproperties
In addition to circulation differences be-tween individuals with and without diabetes,a number of dermatological and skin structuralfactors are associated with diabetes.69 Amongthese dermal structural effects are diabetes“thick skin” and diabetes “yellow skin,” whichmay relate pathophysiologically to acceleratedcollagen aging, elastic fiber fraying, and in-creased cross-linkage glycosylation of collagenfibers. Monnier et al.70 reported change in der-mal collagen structure in patients with diabetesand showed that the level of collagen glycation
was associated with the level of glycated he-moglobin (HbA1c). X-ray diffraction studiesshowed structural changes in collagen skinfiber as a result of diabetes.71 Light reflectedfrom the skin of individuals with diabetes willhave a different intensity and a different re-sponse to glucose concentration depending onthe extent of structural differences induced byglycation.
Diabetes and blood cell morphology
RBC aggregation is a complex phenomenonthat has been widely studied. RBCs aggregateat low blood flow conditions. RBC aggregationis important to microcirculation and is a majorfactor that contributes to changes in blood flowproperties.72–75 There are reports of differencesin refractive index of RBCs and in their aggre-gation patterns between individuals with andwithout diabetes.76,77 This difference affectsblood flow and light scattering by RBCs. Thecombined effects of diabetes on the structureof cutaneous fibers and on RBC aggregationcan lead to a difference in transmission, diffuse reflectance, and localized reflectancesignals of blood-containing tissues. RBC aggregation is also affected by the concentra-tion of other blood components as triglyceridesand cholesterol.78–80 It depends on otherpathological conditions that are associatedwith low-flow states or change in cell mem-brane properties. These pathological condi-tions include diabetes, trauma, ischemia, elevated plasma fibrinogen, hemoglobino-pathies, oxidative stress, inflammation, unsta-ble angina, acute myocardial infarction, andbacterial infection.81,82
Temperature dependence of optical properties ofhuman skin
Our group recently studied the optothermalproperties of human skin within a 2-mm depthin tissue. Temperature change affected �a, �s,and light penetration depth in cutaneous tis-sue.83–87 The effect of temperature on �s wasattributed to changes in nISF. The effect of tem-perature on �a was attributed to its effect onblood perfusion to outer skin layers. Light pen-etration depth in skin increased by loweringskin temperature.85–87 Change in light penetra-
KHALIL668
5264_e12_p660-697 9/29/04 12:48 PM Page 668
tion depth with temperature and wavelengthcan help targeting a particular cutaneouslayer.85–87 Cooling the cutaneous tissue de-creases �s in a manner similar to increasingglucose concentration. Cooling decreases �a ina manner similar to decreasing insulin or glu-cose concentration (decreases cutaneous perfu-sion). Heating the cutaneous tissue increases�s in a manner similar to decreasing glucoseconcentration. Heating also increases �a in amanner similar to increasing glucose or insulinconcentration (increases cutaneous perfusion).
There were noted differences in the values ofthe optical parameters of skin from personswith and without diabetes. The fractionalchange in localized reflectance as a result oftemperature change was used in combinationwith a nonlinear discriminant function to clas-sify volunteers into those with and without di-abetes,85–87 which suggested dependence of tis-sue optical and thermal properties on thediabetes status of an individual. Light scatter-ing by capillary RBCs depends on nRBC/nplasmaand will vary with temperature, and can con-tribute to the observed thermo-optical behav-ior of skin from subjects with diabetes. The previously mentioned blood circulation differ-ences,47–58 the skin structural effects,69–71 andthe effect of diabetes on blood flow and rheol-ogy72–77 may have manifested themselves inthe observed discrimination between thosewith and without diabetes in those studies.85–87
Discrimination between individuals withand without diabetes on the basis of skinthermo-optical response can be used as an NIscreening method of the diabetes status or toassess the extent of complications. An NIscreening method can replace multiple glucosetests, HbA1c determination, and probably anMTT or OGTT. Temperature modulation of op-tical properties of human skin was used in aset of meal tolerance experiments to track glu-cose changes in human subjects.11
Nature of glucose NI measurement’s opticalwindow
The outermost layer of the body part is theoptical window for an NI measurement. Skinlayers, oral mucosa, and the eye lens are opti-cal windows through which an NI glucose
measurement is performed. Variation in theirproperties from person to person, and differ-ence from time to time for the same person, willaffect the reproducibility of the optical mea-surement. Change in the characteristics of theoptical widow as a result of interaction with themeasuring probe, effect of temperature, andvariations in optical and thermal coupling be-tween the probe and the window will affect thereproducibility of the optical measurement.Transepidermal water loss is a characteristic ofliving skin.88 Prolonged contact between probeand skin causes local occlusion and limits theescape of water molecules, and can lead to in-creased hydration of the SC. Hydration of theSC reaches a constant value of 70%43; excesswater may then appear as a thin layer betweenthe skin and the probe. Transepidermal waterloss may cause a time-dependent change or adrift in the measured optical signal, especiallyin diffuse reflectance measurements. Blank etal.89 used local occlusion to homogenize the ef-fects of different levels of skin dryness on theoptical signal.
The results of the thermo-optical studies onhuman skin suggest that because of blood cir-culation differences and skin structural effects,skin from an individual with diabetes contrib-utes to the measured signal in a manner dif-ferent from skin from an individual without di-abetes.83–87 Thus, skin does not act as a passiveoptical window for an NI glucose measure-ment. Differences in skin properties and theirdependence on disease state or environmentalconditions should be taken into considerationin analyzing NI optical signals.
OPTICAL METHODS FOR NI TISSUEMEASUREMENTS
NIR transmission and reflectance
NIR transmission and reflectance NI mea-surements of glucose are predicated upon thepremises that glucose-specific information isembedded within the NIR spectra and can beextracted by using multivariate analysis meth-ods. Study of several body sites (webbing,tongue, upper lip, lower lip, nasal septum, andcheek) for suitability for NIR transmission mea-
NON-INVASIVE GLUCOSE MEASUREMENT 669
5264_e12_p660-697 9/29/04 12:48 PM Page 669
surements based on the water and fat contentssuggested the tongue as the body part with thelowest fat content.90 Burmeister et al.19 studiedNIR (7,000 and 5,000 cm�1, 1,430–2,000 nm)transmission through the tongue of five sub-jects with type 1 diabetes over a period of 39days, performing five measurements per day.The tongue was selected because it is highlyvascularized, it does not have an SC, it has lit-tle fatty tissue, it is a homogeneous body part,it has a nearly constant temperature, and it of-fers an aqueous effective pathlength between 5and 6 mm. Spreading the data collection overseveral weeks (a random series of spot tests) in-stead of an OGTT or MTT alleviates contami-nation of the data set with temporal effects. PLScalibration models were generated for the dif-ferent individuals. Models were used to predictblind sample concentration (each fifth mea-surement) or the later part of the data with SEP�3 mmol/L and Rp varying from person toperson. Both sets of prediction data points wereobtained at different days.
Determination of glucose in oral mucosa us-ing a diffuse reflectance technique was exten-sively discussed in previous work and in morerecent work from Heise’s group.91,92 The wave-length range used was 1,111–1,835 nm. Lightpenetration in tissue in diffuse reflectance mea-surements is limited, leading to a short lightpathlength. The low extinction coefficient ofglucose in this region makes for a very smallsignal change. The oral mucosa was selectedbecause it is highly vascularized and it does nothave an SC layer. One may argue that in spiteof the short penetration depth, light is samplingthe vascular mucosal bed. Glucose was as-sumed to be determined in the blood vessels ofmucosal tissue. The lowest SEP was 2.1mmol/L. The medium through which light istransmitted and reflected differs from skin andtissue used in other studies. There is a poten-tial for a lag time between glucose concentra-tion in blood and saliva. Saliva componentsand residual food in the mouth present sourcesof interference in this measurement.1
Gabriely et al.15 used the NIR reflectance ofthe thumb, in the 400–1,700 nm range, for NIdetermination of glucose in a set of insulin andglucose infusion experiments. The thumb wasselected because it is highly vascularized. PLS
was used for data fitting. Exceptionally highcorrelation coefficients (Rp �0.95) and excep-tionally low SEP (�0.28 mmol/L) were re-ported.15 An objection was raised that the pre-diction set and the calibration set were notindependent.93 A rebuttal indicated that theprediction set had masked samples from the to-tal set that were not used in the calibrationmodel.94 The prediction error values are vastlysmaller than those reported for other NI-NIRtechniques, for in vitro glucose results in amuch simpler media, and even for some com-mercial invasive home glucose meters. Thesesmall SEP values are quite encouraging, unlessthere is overfitting in the data analysis, whichwarrants a thorough review of this work. Thereare some questions that are not clearly an-swered,15 for example:
1. Reference measurements. Did the referencemeasurement contribute to signal quality?
2. Body interface. Was the optical probe in con-tact with the thumb all the 5-h experimenttime? Were the measurements performedintermittently? What was the magnitude ofthe repositioning error of the thumb? Whatwas the applied pressure by the measure-ment fixture? Was there any occlusion dueto the applied pressure?
3. Wavelength. Were the 10–27 factors in thePLS regression analysis individual wave-lengths or spectral bands? Which wave-lengths in the 400–1,700 nm range were usedin the PLS fitting? How many independentspectra were used per PLS factor? Whichportion of the spectrum contributed to thecorrelation? Did the hemoglobin absorptionbands contribute to the correlation?
4. Timing. Did the experiment time overlapwith one segment of the circadian rhythmand one time window for the instrument?Chance correlation between instrument pa-rameters and time contributed to sequentialin vitro measurements of glucose.33
The issue of independence of prediction dataset from the calibration set was debated.93,94
The masked experiment data points that rep-resented the prediction set were not includedin the calibration set.94 They are still a part ofthe events in the same time sequence for each
KHALIL670
5264_e12_p660-697 9/29/04 12:48 PM Page 670
clamp experiment, which makes them followthe same time dependence of whatever pa-rameter that is changing during the experi-ment. Data sets used in the calibration modeland data sets used for testing prediction shouldbe obtained in separate clamp experiments. Anexample would be using the calibration modelfrom one day to predict clamp experiment re-sults from another day. A more direct approachto avoid chance correlation, and to account forrepositioning errors, is to test randomized setof spot tests.19 The small SEP obtained in thestudy by Gabriely et al.15 can be very encour-aging and warrants repeating the experimenton a number of individuals to confirm the pub-lished data and to answer the questions aboutexperimental design and data analysis. If theSEP values are still as low as reported,15 thiscan present an important step towards the NI-NIR optical detection of glucose.
Maruo et al.10,95 reported a NI glucose deter-mination method based on NIR absorption atthe overtone bands of glucose. A 200-�m illu-minating optical fiber in touch with the skin de-livered polychromatic light. Reflected light wascollected at a distance 0.65 mm from the illu-mination point using a second 200-�m fiber.Collected light was analyzed by a spectrome-ter as absorption spectra in the wavelengthrange 1,500–1,800 nm.10,95 This optical arrange-ment restricted the sampled depth to 0.5–2 mmto encompass the dermis. It avoids the SC andepidermis on the top, and the adipose tissue atthe bottom of the dermis layer. The measuredabsorbance (�log10 reflected intensity) was fit-ted to the in vitro glucose data to generate a cal-ibration model. Two subjects without diabeteswere tested, with six OGTTs performed. Fiveof these tests were used in the calibrationmodel, and the sixth set was predicted for eachperson from his own calibration set; the SEPwas 1.4 mmol/L. There was a run-to-run biasof up to 5.6 mmol/L. This bias was attributedto site structural differences.95
Using PLS fitting and the formalism given inEq. 1, Maruo et al.95 found that the largest valuefor the regression coefficient was at �1,600 nm.A similar positive regression coefficient wasfound at the same wavelength when glucosewas determined in serum using Fourier trans-form IR. Maruo et al.95 used this as an evidence
of specificity, i.e., the signal at 1,600 nm is dueto NIR absorption by glucose molecules. Onecontrol experiment was performed where thesubject drank cold water, and no glucose waspredicted by this subject’s measurement.
In a longitudinal study, Samann et al.34 re-ported the long-term accuracy and stability ofdiffuse reflectance of human finger over the800–1,350 nm spectral range. Spectra of 10 patients with diabetes were evaluated. Indi-vidual calibration models were calculated foreach patient from the spectra, which wererecorded at the beginning of the investigation.These models were then applied to calculateblood glucose values from subsequently ob-tained spectra, which were recorded 84–169days later. The long-term accuracy and stabil-ity of the calibration models, expressed as theroot mean squared error of prediction, variedfrom 3.1 to 35.9 mmol/L. The results show theneed to improve the long-term stability of thedetection and calibration method, and to un-derstand the underlying physiological pro-cesses over time.
NIR diffuse reflectance of the arm was mea-sured for several bands in the 1,050–2,450 nmspectral range and was correlated with glucoseconcentration.89 Precautions were made tominimize data variability due to skin hydra-tion, variations in tissue temperature, andprobe contact pressure. In a first approach,seven volunteers with diabetes were studiedover a 35-day period with random collection ofNIR spectra. The second approach involvedthree volunteers without diabetes and the useof OGTT over multiple days. Statistically validcalibration models were developed on three ofthe seven volunteers with diabetes. The meanSECV was 1.41 mmol/L. The results from theOGTT testing of the three volunteers withoutdiabetes yielded an SECV of 1.1 mmol/L. Val-idation of the calibration model with an inde-pendent test set produced a mean SEP equi-valent to 1.03 mmol/L. This group studiedcalibration transfer between instruments andbetween persons and the stability of calibrationover extended time periods. It was possible toestablish a standardized algorithm for nine outof 139 subjects.89
A summary of data on glucose NIR ab-sorbance measurements in human experiments
NON-INVASIVE GLUCOSE MEASUREMENT 671
5264_e12_p660-697 9/29/04 12:48 PM Page 671
is given in Table 2, which describes the speci-ficity, compartmentalization, and calibrationstatus of each method discussed.
Mid-IR spectroscopy
A different approach is to examine the fun-damental frequencies of glucose is the mid-IR.Janatsch et al.96 used mid-IR spectral analysisof human blood plasma with an ATR cell. Theconcentrations of total protein, glucose, triglyc-erides, total cholesterol, urea, and uric acidwere measured by chemical or enzymaticmethods. For these constituents a PLS algo-rithm was used for a multivariate calibration,including the IR fingerprint region of theplasma spectra. The average cross-validationprediction SECV for glucose was 1.2 mmol/L.96
Deissel et al.97 determined glucose using mid-IR spectroscopy and ATR measurement. Driedglucose films from small volume (100-nL)aqueous glucose solutions were deposited onan (ATR) accessory of a Fourier transform IRspectrometer equipped with a pyroelectric de-tector. Quantification of glucose was achievedbetween 0.6 and 33 mmol/L in samples withvolumes as low as 7 nL. The SEP for the con-centration range 0.6–5.6 mmol/L was 0.18mmol/L, with full interval data between 1,180and 940 cm�1. When all samples within thewhole concentration range to 33 mmol/L wereincluded, the SEP increased to 1.1 mmol/L be-cause of a nonlinear signal dependence on glu-cose concentration.97
Petrich et al.98 correlated the shape of themid-IR spectra (1,000–4,000 cm�1) of driedserum samples with the disease state of thosewith type 1 or type 2 diabetes and without di-abetes. These investigators applied clusteranalysis and discriminant analysis to the data.Approximately 80% sensitivities and specifici-ties of the disease state were achieved withintheir data set.98 Spectral features in the mid-IRspectrum 1,119–1,022 cm�1 (8.937–9.785 �m)were used for measurement of glucose inwhole blood.99 Another mid-IR determinationglucose in whole blood in the range 950–1,200cm�1 (8.333–10.526 �m) yielded an SEP of 0.59mmol/L.100
A comparison of the performance of NIR andmid-IR spectral measurements is presented inthe work of the Sandia Group. Haaland et al.101
used NIR spectroscopy to determine glucoseconcentration in whole blood using a 1-mm celland PLS for data analysis; PLS-SECV was 1.8mmol/L over a range of 0.17–41.3 mmol/L inblood and glucose-spiked blood. Ward et al.102
measured in the mid-IR with blood from sixsubjects with diabetes. In one experiment theblood was spiked with glucose in the range2.8–25 mmol/L. The second set of samples waspostprandial blood obtained from the samesubjects during an MTT that had a glucoserange between 5.87 and 20.9 mmol/L. PLS-SECV was 0.61 mmol/L for the spiked samplesand 0.72 mmol/L for the postprandial bloodsamples.102 These data indicate that mid-IRspectral measurements yield lower SECV val-
KHALIL672
TABLE 2. GLUCOSE NIR ABSORBANCE MEASUREMENTS IN HUMAN EXPERIMENTS
Arm diffuse Malin et al.12; Glucose NIR Epidermis and dermis OGTT, calibration,reflectance Blank et al.89 1,050–2,450 nm prediction SEP �
1.39 mmol/L
5264_e12_p660-697 9/29/04 12:48 PM Page 672
ues than the NIR counterpart for in vitro whole-blood samples.
Heise and Marbach21 used the ATR tech-nique to characterize the outmost epidermallayer of human oral mucosa in the spectralrange 800–4000 cm�1. For several test sub-jects, lip spectra were recorded during anOGTT. The individually varying blood glu-cose concentration was followed by means offrequent blood testing. Oral mucosal spectraare supposed to vary with glucose concentra-tion in the same way that aqueous solutionsof glucose do. There was no evidence for anunderlying glucose absorbance spectrumfrom analyzed data of this experiment. In an-other approach, difference spectra betweenthe mean spectra of a subject with diabeteshaving a mean glucose value of 13.3 mmol/Land of a subject without diabetes having amean value of 6.7 mmol/L showed a struc-tured band (peak absorbance of 0.01 ab-sorbance unit) in the 1,000 and 1,200 cm�1
range. Although there is no SC in the oral mu-cosa, the difference spectrum in this spectralinterval resembles the published spectrum ofthe SC.43,44 There was no clear evidence ofmid-IR ATR sensing of changes in blood glu-cose concentration using this difference-spec-tra data analysis approach.21
Several patents by Berman and Roe103–105
describe a method for the NI determination ofglucose using mid-IR and an ATR element intouch with a human finger. Broadband IR en-ergy is directed into the ATR element and sub-sequently into the skin. Spectra were recordedfor the measurement range 1,000–1,053 cm�1
(10–9.5 �m) and for a reference range of1,143–1,212 cm�1 (7.75–8.25 �m). The inventorsrealized that the light path does penetrate theSC layer only, and thus suggested that the skincompartment in which glucose is measured isthe sweat glands or that glucose is transportedfrom the sweat glands to the outer layer of theskin. The difference in the spectral signal, S, inthe measurement range and the reference rangeis determined, and glucose concentration is de-duced from a simple application of Beer’s law.An empirical equation is used without the needfor multivariate analysis:
[Glucose, mg/dL] � 1,950S � 17 (4)
Such simplicity is a welcome change to the NIglucose determination field, which is domi-nated by calculation-intensive methods.21 Ap-plying PLS regression to mid-IR ATR of theoral S value could not reliably relate the changein signal to changes in blood glucose concen-tration.21
Another aspect of the work of Berman andRoe103–105 is that they applied a skin-cleansingstep, prior to contacting the skin and the ele-ment. The steps involved in the cleaning arewashing with water to remove surface glucose,washing with isopropanol to remove water,and then applying mineral oil to the skin. Thereare no mechanistic data as to what is the roleof each of the washing steps or components onthe optical properties of skin. This is importantto understand, especially in light of the re-ported effect of isopropanol on the water con-tent of the skin and on skin mid-IR ATR spec-tra.45,46 Spectra published by Brancaleon et al.46
show that the 1,000–1,100 cm�1 range of thespectra was affected both by the application ofisopropanol and by tape-stripping of the SC.
Light scattering measurements
Measuring light scattering of tissues. NI deter-mination of glucose was attempted using lightscattering of tissue components measured bylocalized reflectance (spatially resolved diffusereflectance) or NIR frequency domain reflec-tance techniques.106–108 In localized reflectancemeasurements a narrow beam of light illumi-nates a restricted area on the surface of a bodypart, and reflected signals are measured at sev-eral distances from the illumination point. Thevalues of �a and �s for tissue can be deducedfrom the distribution of reflected light densityas a function of source–detector (S-D) dis-tances.35 Both localized reflectance measure-ments and frequency domain measurementsare based on change in glucose concentration,which affects the refractive index mismatch be-tween the ISF and tissue fibers, and hence�s.20,106–108
A glucose clamp experiment showed that��s at 650 nm qualitatively tracked changesblood glucose concentration for the volunteerwith diabetes studied.20 The distances betweenthe source fibers and detector fibers (S-D dis-
NON-INVASIVE GLUCOSE MEASUREMENT 673
5264_e12_p660-697 9/29/04 12:48 PM Page 673
tances) were in the range of 1–10 mm, whichcorresponds to the approximate depth in tissueupon which �s is determined. In a secondstudy, ��s at 804 nm qualitatively trackedchanges blood glucose concentration for thetested volunteers with diabetes. The S-D dis-tances in this second study were in the rangeof 0.8–5.2 mm. Drift in �s that was indepen-dent of glucose prevented statistical analysisand was attributed by the authors to otherphysiological processes contributing to ��s.106
Changes in �s did not exclusively result fromchanges in nmedium caused by increased glucoseconcentration. A third study using localized re-flectance and OGTT involved five healthy vol-unteers and 13 volunteers with type 2 diabetesusing two sensors placed on the abdomen.107
The S-D distances in this study were in therange of 0.8–10 mm, and the wavelength atwhich the scattering coefficient was measuredwas 800 nm. In the case of the volunteers with-out diabetes 80% of measurements showedtracking between ��s and blood glucose con-centration. Twenty percent of measurementson those without diabetes showed no correla-tion between ��s and blood glucose. Seventy-three percent of the localized reflectance mea-surements on those with diabetes resulted in acalibration models for �s versus blood glucoseconcentration. In localized reflectance studies,the probe was continuously attached to the ab-domen.20,106,107 These studies did not addressrepositioning errors.
In a separate measurement in the same ex-periment a microdialysis probe was inserted inthe abdominal tissue 10 cm away from the op-tical probe. Glucose concentrations in the ISFtracked blood glucose concentrations duringthe OGTT experiment, but lagged blood glu-cose values in stepwise clamp experiments. Itwas possible to establish a calibration modelbetween ��s and the concentrations of glucosein the ISF as determined by the microdialysismethod.107 The authors concluded that physi-ological changes in blood glucose could be mon-itored by determination of the ��s in volunteersand patients with diabetes in most experiments.
The determined values of ��s do not corre-late with glucose concentration all the time.Glucose-independent drift in �s has preventedstatistical analysis. There was no attempt to
study the cause of the inability to establish acalibration model when it happened. There areno new reported data on the NI determinationof glucose using ��s determined by frequencydomain measurements.108
A possible interpretation for the poor corre-lation between measured ��s and glucose con-centration in the localized reflectance mea-surements with large S-D distances is that thedepth in abdominal tissue spanned by lightbeams used in these measurements is �10 mm.This depth encompasses the SC, epidermis,dermis, subcutaneous adipose tissue, and ab-dominal muscles. Bulow et al.109 found thatblood flow varied in the corresponding com-partments in the human arm. Blood flow to theforearm and that to the subcutaneous tissueand skin in the forearm were measured bystrain gauge plethysmography, 133Xe-elimina-tion flowmetry, and LDF, respectively, duringan oral glucose load (1 g of glucose/kg of leanbody mass) and during control experiments.Arterial glucose increased from 5.1 � 0.3 to7.8 � 1.17 mmol/L at 30 min after initiating theglucose load, and decreased to 4.4 � 1.17mmol/L by the end of the experiment. Theforearm blood flow remained constant duringboth glucose load and control experiments.Glucose induced a twofold vasodilatation insubcutaneous adipose tissue, which remainedthat way for the duration of the 240-min ex-periment. In skin, glucose induced a 50% rela-tive increase in vasodilatation (over the aspar-tame control case) between 120 and 150 min.This tissue showed a relative vasoconstrictionduring the rest of the experiment. Muscle bloodflow was estimated to decrease by about20–30% during both glucose load and controlexperiments.
The vascular effect of glucose, and possiblythe glucose concentration in tissue that is car-ried by the blood bolus, varies in the differenttissue compartments. The optical signal willvary during the course of the glucose load (in-cluding glucose clamp, OGTT, or MTT) exper-iment. Although the data presented are for theforearm,109 these arguments suggests that mea-surement of optical signal over multiple tissuecompartments (layers) may contribute to thelack of quantitation in localized reflectancestudies that use large S-D distances.20,106,107
KHALIL674
5264_e12_p660-697 9/29/04 12:48 PM Page 674
A summary of the data on tissue scatteringcoefficient measurement of the abdomen isgiven in Table 3. No new data were reportedusing the frequency domain approach in theperiod covered by this review.
Measuring of light scattering by RBCs. An approach that is based on light scattering from RBCs was dubbed occlusion spectro-scopy.110–112 RBCs agglomerate in individualswith diabetes, especially upon occlusion, whenthe shear forces of blood flow are at a mini-mum. Finger blood vessels were occluded toslow blood flow and allow RBCs to aggregate.Change in light scattering upon occlusion wasmeasured. Occlusion will not affect the rest ofthe tissue components, while scattering prop-erties of agglomerated RBCs differ from thoseof the non-agglomerated ones and from the restof the tissue.110–112 It will thus be possible toseparate scattering due to agglomerated RBCsfrom the scattering by the rest of the tissue components. The scattering coefficient of RBCsdepends on the refractive index mismatch be-tween the RBCs and blood plasma nRBC/nplasma, similar to the case of refractive indexmismatch between connective tissue fibers andthe ISF.20,106–108 Change in glucose concentra-tion affects nplasma, and hence affects bloodlight scattering. Occlusion spectroscopy differsfrom that of localized reflectance and frequencydomain measurements in that it proposes mea-surements of glucose in blood rather than inthe ISF.106–108
The occlusion spectroscopy method wastested in a human study using a hyperinsu-linemic-hypoglycemic clamp.111 A linear cali-bration model was established with Rc � 0.836,but there were no reported error estimates.There was no reported interference from serumtriglycerides, catecholamines, and cortisol,which were considered as potential interfer-ents.113 RBC aggregation is a complex phe-nomenon that is observed in the case of dia-betes.77 It has also been reported for otherpathological conditions.80–82 This raises thequestion of specificity. Other diseases affectRBC aggregation.77–82 Clinical studies involv-ing other molecular or cellular interferencesand co-morbidities are needed to separate theeffect of glucose on measured RBC scatteringfrom other factors affecting shape and structureof RBCs. This technique addresses the com-partmentalization issue and offers the potentialof directly measuring the change in the refrac-tive index of blood plasma, nplasma.
Temperature-modulated localized reflectance mea-surement. Skin’s thermo-optical response is sen-sitive to status of diabetes.85 Blood glucose con-centrations alter thermally modulated opticalsignals through physiologic and physical ef-fects. Yeh et al.11 studied the relation betweenskin thermo-optical response and blood glu-cose values. Temperature changes affect cuta-neous vascular and refractive index responses,which in turn are affected by changes in glu-cose concentration.
NON-INVASIVE GLUCOSE MEASUREMENT 675
TABLE 3. TISSUE SCATTERING COEFFICIENT MEASUREMENT DATA ON THE ABDOMEN
Test type Reference Optical system Volunteers Results
Glucose clamp Bruulsema � � 650 nm With diabetes Qualitative tracking of ��s’et al.106 S-D distance 1–10 mm and glucose
Glucose clamp Heinemann � � 804 nm With diabetes Qualitative tracking of ��s’et al.20 S-D distance 0.8–5.2 mm and glucose in 73% of
experimentsOGTT Heinemann � � 800 nm 5 without diabetes Calibration model in 80% of
et al.107 S-D distance 0.8–10 mm experiments with Rc �0.75,interexperiment andinterperson bias, noprediction data
13 with diabetes Calibration model in 73% ofexperiments with Rc �0.75,interexperiment andinterperson bias, noprediction data
5264_e12_p660-697 9/29/04 12:48 PM Page 675
A device based on the thermo-optical re-sponse of human skin was used to collect sig-nal from the forearm of volunteers.83–87 Glucoseconcentrations were correlated with tempera-ture-modulated localized reflectance signals atwavelengths between 590 and 935 nm. Thereare no known NIR glucose absorption bands inthis range. �a corresponds to blood absorptionand thus reflects hemodynamic changes in cu-taneous tissue. �s is a measure of the refrac-tive index mismatch between the ISF and tis-sue connective fibers.
Localized reflectance data were collected con-tinuously over a 90-min period of probe–skincontact as temperature was repetitively steppedbetween 22°C and 38°C for 15 temperature mod-ulation cycles. Each cycle comprised the follow-ing steps: Skin was equilibrated for 2 min at aprobe temperature of 22°C, and the temperaturewas raised to 38°C over the course of 1 min, main-tained for 2 min, and then lowered to 22°C overa 1-min period. At each temperature limit (dur-ing the 2-min window), four optical data packetswere collected, and values of �a and �s were de-termined. The method was tested in a series ofexperiments involving MTT and control runs.The control conditions were no-meal, cold water,
or protein meals. Temperature modulation be-tween 38°C and 22°C caused a periodic set of cu-taneous refractive index and vascular changes,leading to periodic changes in skin reflectance.11
Yeh et al.11 used a four-term LLS fitting ofglucose to the reflectance data in the form:
[Glucose] � a0 � �i ai � Ri (r,�,T) (5)
The reflectance parameter R(r,�,T), as definedby Eq. 5, equals loge (measured localized re-flectance). Thirty-two sequences of R (atT22°C) � loge R(r,�,T22°C) and R (at T38°C) � logeR(r,�,T38°C) were used in the LLS correlation.For each MTT over the 2-h period, the temper-ature sequences encompassed 20 data points.11
Changes in glucose concentrations were pre-dicted using a model based on MTT calibrationfor each volunteer, with an SEP of �1.5mmol/L and Rp � 0.73 in 80% of the experi-ments. There were run-to-run differences in thewhole response curve of predicted glucose val-ues in the form of an upward or downwardshift of up to 5.6 mmol/L. These shifts were at-tributed to site-to-site structural differences.11
Figure 2 shows an example of the MTT resultsfor a volunteer with type 2 diabetes.
KHALIL676
FIG. 2. Results of mean-adjusted prediction of glucose on an MMT on a volunteer subject (Sbj) with diabetes using thethermo-optical response of human skin. Day 4 was used for predicting the glucose concentration in other days. The indi-vidual response curves are mean-adjusted because of the day-to-day shift. Reproduced with permission of The AmericanAssociation for Clinical Chemistry, from Yeh et al.11
5264_e12_p660-697 9/29/04 12:48 PM Page 676
The effect of temperature and change in glu-cose concentration on human skin are twofold:first affecting cutaneous vascular circulation(physiological effect) and second affecting cu-taneous light scattering (physical effect). Thecorrelation between glucose values and opticalsignals in this spectral range was attributed tothe effect of glucose on the cutaneous hemo-dynamic response and on the refractive in-dex.11 Table 4 lists the mean-adjusted four-termLLS prediction parameters for MTT experi-ments on the two volunteers with diabetes.11
An MTT calibration model detected changesin glucose concentration in 80% of the MTTruns. The day-to-day shift in the responsecurve (bias in the signal) was attributed toprobe positioning, cutaneous structural effects,or other physiological changes.11 Maruo et al.95
observed a similar bias in predicted glucosevalues in cutaneous NIR absorption experi-ment.
There was a fortuitous observation for vol-unteer B on run #8 in Table 4; the subject hadfever due to influenza infection. The Rp valuefor run #8 was negative (Rp � �0.69), and theSEP was �5.5 mmol/L. Experiments on thisvolunteer were repeated later as tabulated atthe bottom of Table 4.11 The effect of fever inthis case suggests that testing an NI glucose de-tection method should be performed under dif-ferent disease conditions, in addition to dia-betes.
Glucose concentrations in control runs (non-carbohydrate meals) showed considerable scat-ter when predicted by an MTT calibration
model. The calculated SEP was higher than themeasured glucose range in cases where the glu-cose range was �2 mmol/L. SEP values up to2.74 mmol/L were calculated for a protein mealexperiment where there was no change in glu-cose concentration. This raised the question ofwhat can be used as a control experiment forNI measurements. For the case of protein mealruns, cutaneous hemodynamic changes due todigestion affected the optical signal. The extent,rate, and direction of cutaneous hemodynamicresponse to protein meals may be differentfrom those caused by the change in glucoseconcentration. Further work is needed to testthis method in a spot test experiment.
The thermo-optical response method offerscertain compartmentalization advantages overlocalized reflectance measurements that uselarge S-D distances,20,106–108 as it limits samplingdepth to the dermis by virtue of the probe de-sign (short S-D distances) and the use of tem-perature control. The detection method incorpo-rates temperature effect on both NIR absorptionand scattering processes. No glucose specificityadvantage has been established yet over of theuse of �s determined by frequency domain orspatially resolved measurements.106–108
Optical coherence tomography (OCT). OCT is atissue imaging technique that allows depth res-olution of less than 10 �m.114,115 It allows de-termination of refractive index and scatteringcoefficient values in layered structures in skin,e.g., SC, epidermis, and dermis. The apparatusconsists of a low coherence light source such as
NON-INVASIVE GLUCOSE MEASUREMENT 677
TABLE 4. MEAN-ADJUSTED FOUR-TERM LLS PREDICTION PARAMETERS FOR CARBOHYDRATE-MEAL EXPERIMENTS
Volunteer A Volunteer B
Glucose range Mean- Mean-adjusted SEP Glucose range Mean- Mean-adjusted SEPDay (mmol/L) adjusted Rp (mmol/L) (mmol/L) adjusted Rp (mmol/L)
2 6.93 0.84 1.51 3.84 �0.67 1.153 6.67 a a 3.62 �0.88 0.544 6.3 0.86 1.17 3.49 a a
Reproduced with permission of The American Association for Clinical Chemistry, from Yeh et al.11
aData set used as a calibration model.
5264_e12_p660-697 9/29/04 12:48 PM Page 677
a superluminescent diode and an interferome-ter, which determines the depth of the back-scattering feature by measuring the delay cor-relation between backscattered light in thesample and the reflected light in the interfer-ometer arm. OCT of the palm of the hand andthe volar side of the arm (wavelength) showeda decrease in the refractive index of the SC, adrop in the �s of the epidermis, and a rise in�s of the SC over the 1–1.5-h period. The be-havior of the SC was interpreted by moistureuptake from sweating.116,117 The effect of mois-ture uptake and sweating as a result of inter-action of measuring probe with skin and theircontributions to signal variations has not beenestimated.
OCT was proposed for the NI detection of glucose with no experimental data pre-sented.118 The technique was applied afresh toNI determination of glucose by scanning a two-dimensional image of the skin and then con-verting it to a single curve to obtain a one-di-mensional distribution of light intensity as afunction of depth. The experimental set up isshown in Figure 3. The slope of the OCT sig-nal versus depth line is determined and is cor-related with the concentration of blood glu-cose.119–122 An example of the output of thedetection method is shown in Figure 4. First animage of the layers of skin determined by OCTis shown in Figure 4A. In the bottom part theOCT signal as a function of depth is shown. Thevarious skin layers are indicated in Arabic nu-merals. The slope of the portion of the plot in
the dermis layer is used to calculate �s. In ananesthetized animal skin experiment, OCT im-ages at 1,300 nm illumination wavelength dem-onstrate that glucose affects the refractive in-dex mismatch in skin and decreases �s.121 Theeffect of glucose on the �s values measured byOCT was similar to the effect measured by lo-calized reflectance or frequency domain mea-surements.20,106–108
The stability of the slope of the OCT signalwas dependent on tissue heterogeneity andmotion artifacts. Moderate skin temperaturefluctuations (�1°C) did not decrease accuracy,but substantial change (� several degrees Cel-sius) significantly affected the OCT signalslope, i.e., �s, an effect that is similar to thatobserved in the thermo-optical response stud-ies of human skin.11
An OCT system with a light source of 1,300nm was used to test 15 healthy volunteers in18 OGTT clinical experiments. OCT imageswere taken every 10–20 s from the left forearmover a total period of 3 h. Venous blood wassampled from the right arm vein every 5 or 15min. The slope of the OCT signals versus depthwas calculated for a depth of 200–600 �m be-low skin surface. Figure 5A shows OCT signaltracking of changes in glucose concentration.The OCT signal change is plotted in the reversedirection. The slope of OCT signals correlatedwith blood glucose concentrations throughoutthe duration of the experiments as is shown inFigure 5B, changing up to 2.8% per mmol/L ofplasma glucose values.119,122
KHALIL678
FIG. 3. Optical layout of the OCT system.Light from the 1,300 nm superluminescentdiode (SLD) is split between the reference andsample arms of the interferometer. Depth mea-surement is achieved by moving the mirror withrespect to the reference arm. Reproduced withpermission of the American Diabetes Associa-tion, from Larin et al.119
5264_e12_p660-697 9/29/04 12:48 PM Page 678
The OCT technique is promising as the opti-cal measurements targets a restricted area inthe skin. Unlike the localized reflectancemethod that uses large S-D distances and spansmultiple layers in tissues,20,106–108 OCT ad-dresses the issue of skin layers. It offers cer-tain compartmentalization advantages over lo-calized reflectance measurements, as it limitssampling depth to the upper dermis withoutunwanted signal from other layers. OCT mea-surement has been attempted only in an OGTTor clamp-like studies where there is a sequen-tial rise and fall of glucose concentration. It hasnot been attempted in spot-testing situations orin longitudinal studies. Like other scattering
techniques, the detected phenomenon is the ef-fect of glucose on the refractive index of the ISF.It does not address the effect of circulation andtemperature changes. No specificity advantagehas been established for OCT over other scat-tering studies.106–108
A summary of the human data of methodsbased on light scattering measurements isshown in Table 5. Localized reflectance on hu-man abdomen spans different tissue compart-ments down to 10 mm.106–108 These include the
NON-INVASIVE GLUCOSE MEASUREMENT 679
FIG. 4. A: Typical OCT image obtained from skin of avolunteer using 1,300 nm excitation. B: Correspondingone-dimensional OCT signal. The numbers on the graphrefer to the different cutaneous layers: (1) SC; (2) epider-mis layer; and (3) dermis. The slope of the OCT signalversus depth is close to linear in the dermis layer and isrelated to the scattering coefficient. It was used for cor-relation with glucose concentration. Reproduced withpermission of the American Diabetes Association, fromLarin et al.119
FIG. 5. A: Slope of OCT signals (plotted in the invertedscale) and corresponding blood glucose concentrationsobtained from a volunteer without diabetes. The bloodglucose concentration was measured every 5 min. Dotsrepresent the OCT signal slope (in arbitrary units), andthe black line represents the fit of the data points. Theblack small squares are the invasively measured bloodglucose concentrations. B: Slope of OCT signals versusblood glucose concentration (BGC) for the data shown inA. R, correlation coefficient. Dots represent the OCT sig-nal slope, and the line represents the linear fit of the OCTdata points. Reproduced with permission of the Ameri-can Diabetes Association, from Larin et al.119
0
B
A
100 200 300
3
21
Depth (µm)400 500 600 700
OC
T s
igna
l (ar
bitr
ary
units
)
−100
0.0
0.6
0.8
1.0
0.2
0.4
1
2
3
5264_e12_p660-697 9/29/04 12:48 PM Page 679
dermis, subcutaneous fat, muscles, and bloodvessels therein. OCT localized �s measure-ments to the dermis, RBC scattering localizedit to cutaneous blood vessels, and temperaturemodulation studies localized scattering andperfusion measurements to the dermis.
Optical activity and polarimetry
Polarimetry has been used for quantitativeanalysis of solutions of optically active (chi-ral) compounds such as glucose. When a beamof plane-polarized light transverses a solution,its plane of polarization rotates by an angle �,which is related to the concentration of the op-tically active solute. Optical rotation in the eyewas the earliest proposed NI methods for NIdetermination of glucose.123 Several advancesin the polarimetry technique have been re-ported. Rawer et al.124 discussed different ap-proaches to utilizing the polarizing propertiesof the aqueous humor (AH) for quantitativeglucose measurements. Cameron et al.125 pre-sented in vivo results from a laser-based opti-cal polarimetry system using the anteriorchamber of a rabbit eye. The time delay be-tween blood glucose and NI-measured glu-cose in AH was reported to be less than 5 minin rabbit eye.126 In a different study AH glu-cose in a rabbit’s eyeball lagged blood glucosevalues by a 30-min delay.127 Lane et al.128 stud-ied the acute effect of insulin on AH. Undereuglycemic conditions of high and relativelylow insulin concentrations, AH flow through
the anterior segment of the eye decreased inpatients with type 1 diabetes.128 Bockle et al.129
described a new Brewster-angle approach todetermination of change in degree of polar-ization. Polarimetry requires a body part withlow scattering such as cornea, appropriate cal-ibration, and understanding of lag time be-tween blood glucose and AH glucose. The eyeoffers an advantage over the skin for NI mea-surements of glucose because of the absenceof the SC. Corneal rotation, corneal birefrin-gence, and eye motion artifact are potentialsource of error in polarimetric ocular mea-surements.1
Raman scattering
Use of Raman spectroscopy for the detectionof glucose falls in the category of methods thatmeasure an intrinsic property of the glucosemolecule. Most recent Raman studies are invitro measurements. Raman bands at 900–1,200cm�1 are specific to the molecular structure ofglucose.130,131 Fundamental vibrations moni-tored by Raman spectroscopy are sharper andhave less overlap compared with NIR combi-nation bands. Additionally, water has a low Ra-man cross section. The advent of NIR Ramanspectrometers and use of multivariate spectralanalysis for extracting diagnostic, chemical,and morphological information extended theapplication of Raman spectroscopy to variousclinical applications.130,131
KHALIL680
TABLE 5. HUMAN DATA ON METHODS BASED ON LIGHT SCATTERING MEASUREMENTS
Dermis,subcutaneousfat, and muscles0.8–10 mmdepth, abdomen
Dermis 0.4 to �2mm depth,forearm
Blood vessels
Upper dermis200–600 �mdepth, forearm
Glucose clamp orOGTT,calibrationmodel
OGTT, calibrationmodel/prediction
Clamp, calibrationmodel
OGTT, calibrationmodel
5264_e12_p660-697 9/29/04 12:48 PM Page 680
Raman spectra of blood samples excited at830 nm yielded an SEP of 3.6 mmol/L glu-cose.132 Feld’s group developed a method oflinear multivariate calibration called hybridlinear analysis (HLA),133 which involves incor-porating the spectrum of the desired species(glucose) into the calibration procedure. TheSEP for glucose in serum obtained with PLS is1.2 mmol/L, and the SEP obtained with HLAis 0.94 mmol/L. In whole blood, the PLS-SEPfor glucose was 4.4 mmol/L, and the HLA-de-termined SEP was 3.5 mmol/L.134 The calibra-tion model was stable over a 7-week period.SEP increased on going from serum to wholeblood.
Raman spectra of the AH of the eyes from 32anesthetized rabbits that were excited at 785nm and corrected to eliminate broadband flu-orescence showed fair correlation with glucoseconcentration (Rc � 0.87).135 Correlation im-proved when fluorescence was subtractedprior to linear multivariate analysis (Rc � 0.95).Adding an artificial neural network to the anal-ysis further improved the correlation (Rc �0.99). AH glucose exceeded blood glucose val-ues at normoglycemic levels. It correlated lin-early with plasma glucose above 11 mmol/L.
Raman measurements on intact human skinare expected to be more complex than the caseof solutions or of blood. Raman spectral imagesof skin showed some complex lipid struc-tures.136 Chaiken et al.137,138 studied Ramanspectra of human skin under tissue modulationconditions, which involved the use of thermaland/or mechanical stimuli to produce distrib-utions of blood-rich “replete” and blood-defi-cient “deplete” regions of skin. Raman spectraof human blood in vivo were excited at 785 nmor 830 nm, and the corrected integrated nor-malized intensities of the Raman shifted bandsbetween 400 and 1,800 cm�1 were calculated.Blood volume variations and fluorescence cor-rections were used to improve signal quality.138
Human data that were obtained at 905 nm ex-citation showed that the corrected integratedsignal could be fitted to the capillary glucosevalues for 23 individuals with Rc � 0.74. Re-peated measurements on three individualswith a total of 28 data points yielded Rc � 0.63.Fitting the Raman signal to glucose concentra-tion separately for each of five individuals
yielded Rc �0.9 for the single-person calibra-tion.
Surface-enhanced Raman scattering (SERS)is an effect that results in enhancing the inten-sity of Raman bands of molecules localized inclose proximity of silver and gold surfaces. Ra-man scattering intensity increases when a mol-ecule is spatially confined within the range ofthe electromagnetic fields generated upon ex-citations of the localized plasmon resonance ofnanostructured silver or gold surface. Severalorders of magnitude of sensitivity enhance-ment can be gained over conventional Ramanspectroscopy,139 which may shorten the dataacquisition time. Molecules confined within adecay length of the plasmon electromagneticfield of 0–4 nm will exhibit SERS spectra evenif they are not chemisorbed.140
In a recent report, glucose was partitionedinto an alkanethiol monolayer adsorbed on asilver film over nanospheres (AgFON).141 PLScalculations demonstrated the ability to estab-lish a calibration model for SERS signal versusglucose concentration. The model was vali-dated by LOOCV over a 0–25 mmol/L concen-tration range with an SECV of 1.8 mmol/L.141
Enhancement in the Raman signal is importantfor both minimally invasive and NI glucose de-termination. The partition layer was changedto alkane thiolate tri(ethylene glycol) monolay-ers in order to increase its stability.142 The par-tition rates improved, and the sensor stabilityincreased. The SECV of glucose in water was1.8 mmol/L, and 4.56 mmol/L in the presenceof albumin.142 The rate of improvement in thistechnology makes it a promising sensor at leastfor minimally invasive sensing in the ISF.
PA spectroscopy
PA spectroscopy is used to detect weak ab-sorbencies in liquids and gases.143 A PA mea-surement is an alternative detection technologyfor light interaction with tissues.144–148 Themedium is excited by a picosecond to nanosec-ond laser pulse at a wavelength that is ab-sorbed by a particular molecular species in themedium. Light absorption and subsequent ra-diationless decay cause microscopic localizedheating in the medium, which generates an ul-trasound pressure wave that is detectable by a
NON-INVASIVE GLUCOSE MEASUREMENT 681
5264_e12_p660-697 9/29/04 12:48 PM Page 681
hydrophone or a piezoelectric device. Thepulsed PA signal is related to the properties ofa clear medium by the equation:
PA � k(�vn/Cp)E0�a (6)
where PA is the signal amplitude, k is a pro-portionality constant, E0 is the incident pulseenergy, � is the thermal expansion coefficient,v is the speed of sound in the medium, n � 2,Cp is the specific heat, and �a is the light ab-sorption coefficient. This equation applies toclear solutions or crystals.
PA spectroscopy is relatively unaffected byscattering in optically thin (clear) media, but isaffected by scattering in optically dense media.Light scattering in the medium affects theshape of the PA pulse. In highly scattering me-dia �eff replaced �a. Dispersion of the PA sig-nal as a function of wavelength mimics the ab-sorption spectrum in optically thin (clear)medium. It is equivalent to the diffuse re-flectance spectrum in an optically densemedium. Fainchtein et al.145 provided detailedanalysis of generation and propagation of PAsignals in blood.
MacKenzie et al.144 studied the PA effect inglucose solution in the low scattering case. PAsignal generation is assumed to be due to ini-tial light absorption by the glucose molecules.Solutions were excited by NIR laser pulses inthe 1,000–1,800 nm range, at wavelengths thatcorresponded to NIR absorption of glucose.There was a linear relationship between PA sig-nal and glucose concentration in aqueous so-lutions.144 Human experiments showed thatthe PA signal tracks change in glucose concen-tration.144 No data analysis was presented toshow an advantage of PA spectroscopy over anNIR transmission or reflectance for NI mea-surement of glucose.
A different approach is the use of ultravioletlaser pulses at 355 nm. The generated pulsedPA time profile is used to detect the effect ofglucose on tissue scattering, which is reducedby increasing glucose concentration.146 PA timeprofiles were analyzed to yield �eff, which like�s is related to changes in the refractive indexof the medium induced by changes in glucoseconcentration. In vivo PA profiles measured inrabbit sclera before and after intravenous glu-
cose administering demonstrated that a 1mmol/L increase in glucose concentration re-sulted in a 3% decrease of �eff. In this case theabsorbing species is one of the amino acidresidues in the sclera, and the PA pulse shapeis modulated by light scattering expressed by�s, which is dependent on the concentrationof glucose through its effect on the refractiveindex. Additionally, the thermoelastic parame-ters of the medium �,v, and Cp also affect themagnitude of the signal in a way that is relatedto glucose concentration.
Absorption of laser pulses by the glucosemolecules is not the only mechanism necessaryfor observation of the PA effect. From Eq. 6,once a compression wave is generated afterlight absorption by any chromophore in themedium, the resultant PA pulse intensity canbe modified by the medium thermoelasticproperties �,v, and Cp. All three properties aredependent on glucose concentration. Using theequation P � k(�vn/Cp)E0�a to analyze thedata presented in the dissertation by Zhao148
shows that in response to a 1% change in glu-cose concentration in water solutions, the cal-culated thermal expansion coefficient �, spe-cific heat Cp, and acoustic velocity v changedby 1.2%, �0.6%, and 0.28%, respectively. By in-cluding the three parameters in the equation,the calculated change in the amplitude of thePA signal is 2.05%. The measured change in theacoustic signal, at 905 nm excitation, was 2%.Thus it is possible to account for the magnitudeof the PA signal excited at 905 nm without taking into consideration the absorption coeffi-cient of glucose. The optical absorption coeffi-cient at 905 nm is 0.007 mm�1. At this wave-length the optical absorption by glucosecontributes negligibly to the signal.148
Addition of a scattering component to themedium enhanced the PA signal exited at 905nm. The PA signal resulting from 55.56mmol/L (1 g/dL) glucose increased by 250%in 3% milk solution, by 50% in tissue, and by700% in blood as compared with its magnitudein clear water solutions.147,148 Temporal dis-persion PA curves of blood suspensions con-taining glucose and excited by a 905 nm pulsedlaser source demonstrated that glucose de-creased �s.147,148 This effect is similar to the PAsignals from the eye sclera.146 It is also similar
KHALIL682
5264_e12_p660-697 9/29/04 12:48 PM Page 682
to the scattering �s on tissue and turbid me-dia.106,119,122 The absorbing species at 905 nmis most probably the hemoglobin molecule. Theparameter measured by the PA effect in bloodsuspensions at 905 nm is most probably the ef-fective attenuation coefficient �eff.
The interplay of light absorption andmedium properties change on the PA signalwas further investigated by Shen et al.,149 whodetermined time profiles of pulsed PA signalsin graphite suspensions at excited at 900 nm.The suspensions had absorbance values rang-ing between 12 to 53 absorbance units/cm. ThePA-determined �a, using Eq. 6, was linearly re-lated to absorbance measured by an NIR spec-trophotometer. PA signals for glucose in waterwere measured at 1,450 nm with concentrationsranging between 111 and 833 mmol/L. The be-havior of the signal change was indicative ofthe water displacement effect. The PA signalreached the detector faster at higher glucoseconcentrations. The data were interpreted asmeaning that in the graphite case, the domi-nant effect was the absorption coefficient in Eq.6. In the glucose case at 1,450 nm excitation, itwas the effect of glucose on the sound velocityin the medium.149 The direction of signalchange was similar to that of water displace-ment in the NIR absorption at 1,450 nm. Theconcentration and optical densities used weremuch higher than the physiological glucoseconcentration or tissue light absorption ranges.
Measurement of sound propagation in tissueis dependent on mechanical coupling betweenthe skin and the measuring probe and the pres-sure of the probe on the skin. This effect is quitesimilar to ultrasound propagation in tissuewhere coupling jells are used to improve thespeed of sound matching and decrease soundreflections. Dependence of the signal onprobe–skin interaction is obvious in NIR opti-cal methods.
Still further work is needed to understandPA origination and propagation in tissue andits use for NI determination of glucose. The PAeffect was not shown to offer any advantagesin detection specificity over other NIR absorp-tion or diffuse reflectance methods. It is possi-ble to account for the measured magnitude ofthe signal under some experimental conditionsby factors other than the NIR absorption of glu-
cose. The enhancement of the signal in bloodthat was excited at 905 nm is probably due tolight absorption by hemoglobin and modifica-tion of the signal by glucose.147,148 In the scat-tering mode, PA does not offer any noticeableadvantage over other scattering methods. ThePA signal tracks �eff in scattering media. Thepulse profile is related to the �s and is depen-dent on the refractive index mismatch betweenscattering centers and the medium.146 The na-ture of absorptive events or the “absorbingchromophore” that generates the PA signal isnot known with certainty. There is no evidencethat glucose molecules are the primary ab-sorbing species of the NIR light pulse that gen-erates the compression wave. Water, glucose,blood, and tissue amino acid residues can ab-sorb NIR light pulses. The generated compres-sion wave is modulated by scattering and ther-moelastic properties of the medium, whichchange with change in glucose concentrations.
Glucon, Inc. presented a novel PA applica-tion in the Diabetes Technology Meeting in At-lanta in November 2002, which is posted ontheir web site.150 Additionally, several varia-tions on combining ultrasound and PA spec-troscopy were described in a patent publica-tion.151 The methods presented include:
1. Use of an ultrasound transducer to locate abolus of blood in a vessel and then illumi-nate it with a pulsed laser at a glucose ab-sorption wavelength. The same ultrasoundtransducer detects the generated PA signal.
2. A second variation is to detect ultrasoundsignal reflected from a blood vessel beforeand after PA excitation. Glucose is then de-termined from the difference in reflected ul-trasound intensity.
3. A third approach is to excite a blood bolusin a vessel via a PA effect. The change in thedimensions and speed of the excited boluscauses a Doppler shift to an ultrasoundpulse directed towards the blood vessel.Glucose is determined from the magnitudeand the delay of the Doppler-shifted ultra-sound peak.150,151
The purpose of combining of ultrasound andPA is to target a specific body compartment,mainly a blood vessel, thus addressing the
NON-INVASIVE GLUCOSE MEASUREMENT 683
5264_e12_p660-697 9/29/04 12:48 PM Page 683
compartmentalization issue. Specificity needsto be addressed. Is the PA signal due to spe-cific glucose absorption or due to blood com-ponent absorption of the pulse, and then glu-cose affects the plasma thermoelastic andrefractive properties? There are no publishedclinical data, but it is posted on the Glucon, Inc.web site.150 Human data on pulsed PA deter-mination of glucose are given in Table 6.
PA is emerging as a novel blood vesselimaging modality and may be useful to studydiabetes vascular complications. Hoelen etal.152 imaged blood vessels in highly scatter-ing samples, using 532 nm light, to depths of�1 cm.
Thermal and IR emission measurements
Thermal gradient spectroscopy (TGS), an IRemission technique, is based on measuring thefundamental absorption bands of glucose at9.1–10.5 �m, and bands of other analytes suchas water and proteins. A prototype incorporat-ing the technology has been described.153–156
In TGS technology the spectroscopic energysource is the body’s naturally emitted IR en-ergy, which is absorbed by glucose at �10 �m.
In one application of TGS the surface of theskin is cooled to approximately 10°C to sup-press its absorptive effect. NIR light penetra-tion depth in skin increased by lowering itstemperature, which can help targeting a par-ticular cutaneous layer.85–87 This is attributedto a decrease in absorption and scattering co-efficients. The cooling-induced skin trans-
parency allows monitoring of IR emission fromthe ISF and cutaneous layers.153–156 Using IRemission instead of IR absorption allows sam-pling of deeper cutaneous layers below the SC,which is the layer sampled by mid-IR absorp-tion measurements. In vivo clinical data on sev-eral individuals with type 1 diabetes have beenreported.153 A linear response between in vivodetected glucose and reference blood glucosevalues has been reported. Multiple LLS fittingto the calibration data yielded an SD of 1.4mmol/L.
In an alternate application of TGS155,156 theskin surface is continually cycled up and downin temperature �5°C around normal skin sur-face temperature at a rate of approximately 1Hz. The resulting thermal oscillation producesa modulated IR signal, which is dispersed intoselected mid-IR bands by IR-pass filters, de-tected, and synchronously demodulated. In thistechnique the relative phase of the signal at eachwavelength relates directly to the relative spec-tral absorption of the tissue at each mid-IRwavelength. An algorithm transforms thisphase difference into a mid-IR absorption spec-trum. Once the mid-IR absorption spectrum hasbeen determined conventional spectroscopictechniques are proposed to be employed toquantify glucose. TGS offers specificity andcompartmentalization advantages. LLS fittingwas used, but the fitting conditions and theproprietary transformation algorithm were notdisclosed in sufficient details.
Measurement of the amplitude of IR emis-sion from the tympanic membrane using a fil-
KHALIL684
TABLE 6. SUMMARY OF PULSED PA DETERMINATION OF GLUCOSE
ter at 10.5 �m and signal difference to a knownglucose concentration has been proposed. The data indicate tracking between the signaland glucose concentration in an MTT.157,158
Buchert157 suggested a method that is based onthe premises that the human body naturallyemits IR energy, a portion of which is charac-teristic of glucose and can be utilized for its NIdetermination. The method utilized a sensor in-serted in the ear canal to measure IR radiationemitted by the tympanic membrane. Ear ther-mometers are commonly used to measure bodytemperature from the wavelength-integratedintensity of IR radiation emitted from the tym-panic membrane. The tympanic membrane isan excellent site to measure body temperaturebecause it shares its blood supply with the hy-pothalamus, the center of core body tempera-ture regulation. When compared with the the-oretical blackbody radiation this IR radiation ofthe tympanic membrane is spectrally modifiedby blood glucose, which changes the mem-brane’s emissivity and make it possible to mea-sure the concentration of blood glucose.157,158
Malchoff et al.158 described an NI glucosemonitoring test device that is schematicallyshown in Figure 6. The instrument, which de-tects IR radiation from the tympanic mem-brane, consists of two thermopile detector/op-tical IR filter sets. One of the sensing elementsis covered by a 9.6-�m IR filter sensitive to a
glucose IR wavelength. The second sensing el-ement is covered with a filter that does not havespectral bands of glucose such as the quasi-isosbestic point at about 8.5 �m. Spectrallymodified IR radiation from the tympanic mem-brane illuminates both detectors. The differ-ence in radiation intensity between the two ra-diation paths provides a measure proportionalto glucose concentration.157
IR emission from the tympanic membranewas calibrated versus serum glucose concen-tration using 432 paired measurements from 20subjects with insulin-dependent diabetes. Thecalibration was subsequently tested in a blindfashion with 126 paired measurements from sixvolunteers with diabetes.158 Based on the cali-bration model, predicted glucose concentra-tions had an SD of 1.48 mmol/L, with Rp �0.87. Data presentation differs from used formost multivariate methods. It is not clearwhether this SD is equivalent to SECV or SEP.
The simplicity of this method and the ac-ceptance of the ear thermometers make it quiteappealing. Overlap between the effect of glu-cose on the signal and temperature variationsdue to circadian periodicity needs to be delin-eated. There is a need to separate the mea-surement technique as a temperature responseto glucose change, blood flow response to glu-cose change, or an intrinsic glucose property(IR emission from the glucose molecule). Prob-ably a “fever effect” such as shown in other cor-relations should be explored.11
Fluorescence measurements
Skin fluoresces at 370 and 455 nm when it is excited by ultraviolet light. Multiple regres-sion analysis shows correlations between skinautofluorescence and pigmentation (melanin),redness (hemoglobin), and epidermal thick-ness. Skin autofluorescence depended on pig-mentation, redness, and epidermal thickness,in a descending order.159 Snyder and Grund-fest160 patented the use of laser-induced fluo-rescence for the NI determination of glucose.When glucose solutions were excited with anexcimer laser line at 308 nm, fluorescence wasdetected at 340 and 400 nm, with fluorescencemaximum at 380 nm.160 Fluorescence intensitychanged with change in glucose concentration
NON-INVASIVE GLUCOSE MEASUREMENT 685
FIG. 6. Schematic of the IR emission detection of glu-cose. The target is the tympanic membrane; the wave-guide, inserted in the ear canal, directs the IR radiationto the light valve and then to the detector. Reproducedwith permission of the American Diabetes Association,from Malchoff et al.158
5264_e12_p660-697 9/29/04 12:48 PM Page 685
in an aqueous medium. Glucose does not haveelectronic absorption bands in this short ultra-violet spectral range. The patent does not ex-plain the molecular origin of this fluorescence.Exciting the skin at this short wavelength willlead to a strong scattering component, in addi-tion to the fluorescence. There are no availablehuman data on NI glucose determination by di-rect ultraviolet excitation of human tissue.
In vitro fluorescence assays for glucose havebeen studied as steps towards its in vivo deter-mination. Several methods are based on fluo-rescent resonance energy transfer (FRET) andon competition between glucose and dextranfor concanavalin-A (con-A) binding sites. Theassay components are con-A labeled with anenergy acceptor, or an energy donor, and dex-tran labeled with the complementary moleculefor FRET.161–163
A fluorescence biosensor incorporates rho-damine-labeled con-A [tetramethylrhodamineisothiocyanate (TRITC)-con-A)] as the energyacceptor and fluorescein isothiocyanate-dex-tran (FITC-dextran) as the energy donor. Bothmolecules are chemically conjugated into a hy-drogel network.161 In the absence of glucose,TRITC-con-A binds with FITC-dextran, and theFITC fluorescence is quenched through FRET.Competitive glucose binding to TRITC-con-Aliberates FITC-dextran, resulting in increasedfluorescence intensity proportional to the glu-cose concentration. The in vitro fluorescence re-sponse was linear over a glucose range be-tween 0 and 33 mmol/L.161 Another glucoseassay uses FRET between con-A labeled withfluorescent protein allophycocyanin as donor,and dextran-malachite green as an energy ac-ceptor. Glucose competitively displaces dex-tran-malachite green and increases allophy-cocyanin fluorescence intensity. The in vitroassay had a glucose dynamic range of 2.5–30mmol/L.162,163
Evans et al.164 developed an in vitro cell cul-ture model of skin-component cells as a modelto test the NI glucose monitoring by measuringNAD(P)H-related fluorescence changes in tis-sues. NAD(P)H solutions fluoresce at 400–500nm when excited at 340 nm. 3T3-L1 fibroblastsand adipocytes were grown in culture, and theresponse to added glucose was assessed bychanges in steady-state autofluorescence at
400–500 nm.164 Spectral properties indicatedthat the fluorescence was due to NAD(P)H pro-duction. Cells stained with the fluorescent mi-tochondrial marker rhodamine-123 showed im-mediate and marked decrease in fluorescencewhen exposed to glucose.164
Masters et al.165 demonstrated the use of mul-tiphoton excitation fluorescence microscopy(MPFM) for functional imaging of the meta-bolic states of in vivo human skin cells. MPFMat 730 and 960 nm was used to image in vivohuman skin autofluorescence from the surfaceto a depth of approximately 200 �m. Fluores-cence lifetime images were obtained at selectedlocations near the surface (0–50 �m) and atdeeper depths (100–150 �m) for both excitationwavelengths. Cell borders and cell nuclei werethe prominent structures observed. NAD(P)Hwas found to be the primary source of the skinautofluorescence for the 730 nm two-photonexcitation. A two-photon fluorescence emissionat 520 nm was attributed to flavoprotein.165
NAD(P)H is involved in glucose metabolism.Change in NAD(P)H fluorescence in adipo-cytes, keratinocytes, and other upper dermiscomponents, when measured by MPFM, maybe used to track glucose changes in humanskin. Spectroscopic methods to decrease or di-minish the effect of skin scattering can be usedto improve the quality of fluorescence signals.The effects of other compounds that utilize theNAD(P)H in their metabolic pathways need tobe investigated.
Measurement of skin fluorescence from theepidermis, or the upper dermis, will require theuse of confocal fluorescence microscopy.166
Swindle et al.167 reported the ability of in vivofluorescence point-scanning laser confocal mi-croscopy to produce real-time, high-resolutionimages of the microscopic architecture of nor-mal human epidermis using an NI imagingtechnology.
March and co-workers polymerized a fluo-rescent complex within a hydrogel to make anintraocular lens that responds well to glucoseconcentration.168,169 The patient wears the lens,which changes its fluorescence depending onglucose concentration. Fluorescence is excitedand detected by a hand-held device.
Lackowicz’s group described boronic acidfluorophores (BAFs), which undergo spectral
KHALIL686
5264_e12_p660-697 9/29/04 12:48 PM Page 686
changes in the presence of sugars, leading to achange in fluorescence intensity and wave-length upon binding to D-fructose.170 The bind-ing affinity decreases for D-galactose and D-glu-cose. These BAF probes showed wavelengthshifts and intensity changes when embedded ina commercial contact lens (a polyvinyl alcohol-type photo-cured polymer) and immersed inglucose solutions. The probes included stilbene,polyene, and chalcone derivatives. Examples ofBAF probes are 4-dimethylaminostilbene-4-boronic acid, 4-cyanostilbene-4-boronic acid,and 1-(p-boronophenyl)-4-(p-dimethylamino-phenyl)buta-1,2-diene. Stilbene polyene deriv-atives were excited at 320–340 nm, while chal-cone derivatives were excited at 430 nm.170
Argose, Inc. explored the use of fluorescence forthe NI determination of glucose in the skin.171
Fluorescence methods that depend on interac-tion between glucose and a specific bindingmolecule, such as boronic acid and con-A, orare involved in a metabolic pathway will havea specificity advantage, when successfully ap-plied in vivo.
Use of photonic crystals
Asher’s group has developed a photonicsensing material that responds to analyte con-centrations via diffraction of visible light frompolymerized crystalline colloidal arrays (PC-CAs).172,173 PCCAs are periodic crystalline col-loidal arrays (CCAs) of spherical polystyrenecolloids polymerized within thin hydrogelfilms. The CCAs are brightly colored as theydiffract visible light due to Bragg’s diffrac-tion.172–177
Bragg diffraction depends on the refractiveindex of the system (solvent, hydrogel, and col-loids), the spacing d between the diffractingplanes. The array will act as a diffraction grat-ing for white light, allowing a specific dif-fracted wavelength � to be detected at a spe-cific glancing angle between the incident lightpropagation direction and the diffractingplanes. Incorporation of charged species, orchange in electric charge in the PCCAs, causesthe arrays to expand, changing the spacing d.The diffraction pattern then changes causing awavelength shift in the light reflected off thearray.172–177
Several studies show the ability of the PCCAfilms to detect metal ions,174 creatinine,175 andglucose.176,177 Asher’s group constructed a glu-cose photonic sensor in the form of a thin acry-lamide diffracting PCCA hydrogel film thatcontains glucose oxidase or phenyl boronicacid crystals as the molecular recognition ele-ments. Attachment of glucose causes change incharge distribution. Glucose capture by glucoseoxidase, or phenyl boronic acid, results inchange in the spacing d in the Bragg equation,causing shifts in wavelengths of diffractedlight. For example, 0.1 mmol/L glucose causesthe diffracted light to shift from yellow at 550nm to red at 600 nm. The diffraction spectralshift responds to change in glucose concentra-tion in in vitro experiments. The plot shows ahigh sensitivity to glucose concentration be-tween 0 and 10 mmol/L.176,177
The polymer film sensor is conceived to beused as a contact lens, which changes color ac-cording to the glucose concentration in tears.The use of glucose oxidase or phenyl boronicacid to capture glucose from tears has speci-ficity advantages. The polymer film methodsmeasure glucose in tears; the relative concen-tration and lag time between glucose concen-tration in blood and in tears require detailedstudies.
NON-OPTICAL METHODS FOR THE NIDETERMINATION OF GLUCOSE
Impedance measurements
Measurement of tissue and cell impedance atdifferent frequencies of an oscillating alternat-ing current field yields the dielectric perme-ability of the cell membranes. The plot of thedielectric permeability �* () at the different os-cillation frequencies is the dielectric spectrumof the sample, where is the oscillation fre-quency in Hz (cycles per second). The dielec-tric spectrum is measured over a frequencyrange of 100 Hz to 100 MHz (102–108 oscilla-tions per second).178–183 The frequency distri-bution and the magnitude of �* () were mea-sured for a suspension of RBCs in glucosesolutions. Both frequency distribution andmagnitude of �* () were dependent on the con-centration of the metabolically active enan-
NON-INVASIVE GLUCOSE MEASUREMENT 687
5264_e12_p660-697 9/29/04 12:48 PM Page 687
tiomer D-glucose, but were independent of theconcentration of the metabolically inactiveenantiomer L-glucose.183
Changes in blood glucose concentration in-duce changes in cells involved in carbohydratemetabolism. Hillier et al.178 studied the effect ofhyperglycemia in decreasing the Na� concen-tration in healthy subjects under insulin defi-ciency conditions. In addition to its role in inducing hyperglycemia, insulin deficiency re-duces the permeability of most cells to glucoseand accentuates its osmotic effect. Mean serumNa� and plasma glucose concentrations for sixhealthy fasting volunteers, who were given so-matostatin to suppress insulin secretion, weredetermined in a glucose infusion experiment.Infused glucose brought up its blood concen-tration to 33.3 mmol/L within an hour, andthen glucose infusion was stopped and insulinwas infused to lower the blood glucose con-centration to 7.78 mmol/L. As shown in Fig-ure 7, serum glucose is reversibly related toNa� concentration. The decrease in Na� con-centration as a result of increased glucose wasless pronounced when glucose was infusedwithout suppressing endogenous insulin se-cretion.180
RBCs undergo a decrease in sodium ion con-centration and an increase in potassium ionconcentrations due to water movement from
RBCs to plasma, induced by change in glucoseconcentration.178 Variation in electrolyte bal-ance in blood causes change in RBC membranepotential, which can be followed by determin-ing the permittivity and conductivity of cellmembranes using time domain dielectric spec-troscopy.181,182
Suspensions of human RBCs in phosphate-buffered saline solution containing variableconcentrations of D-glucose or L-glucose of con-stant osmolality showed a spike in �* () at aD-glucose concentration of 12 mmol/L and no such effect when L-glucose was used. Thisshows an interesting specificity for the meta-bolically active enantiomer.183
Changes in the glucose concentrations weremonitored in an NI experiment by varying thefrequency in the radio band over a range thatwas optimized to measure the impact of glu-cose on the impedance pattern.179 A number ofhuman hyperglycemic excursion measure-ments were performed with healthy volunteersusing a wristwatch-size sensor, which holds anopen resonant circuit coupled to the skin anda circuit179:
1. The impedance sensor signals were corre-lated with changes in blood glucose or glu-cose in the ISF during a glucose clamp. Amicrodialysis catheter abdominally placedin the subcutaneous tissue measured bloodglucose and glucose changes in ISF. Bloodglucose was increased rapidly in eighthealthy subjects from 5.56 to 16.67 mmol/L.A good correlation between changes inblood glucose and sensor recordings was re-ported in five out of the eight experiments.The profile for the glucose changes in the ISFis superimposed as well, showing the typi-cal lag time between changes of glucose inblood and ISF.
2. An oral glucose load was administered tofour healthy subjects to raise its values inblood from 5.56 to 16.67 mmol/L. The base-line was set by infusion of somatostatin tosuppress endogenous insulin secretion.Three out of the four experiments showed agood correlation between changes in bloodglucose and the sensor signals.
3. Sensor signals over time were recordedwhile blood glucose remained unchanged.
KHALIL688
0 20
Na+ Glucose
40 60Time (min)
80 100 120 160
Na+
(meq
/L)
Glu
cose
(m
g/dl
)
−20 140115
120 100
200
300
400
500
600
700
0
125
130
135
140
145
FIG. 7. Mean values serum sodium and plasma glucoseconcentrations for six healthy adults in a glucose infusionexperiment. Subjects were fasting and were given so-matostatin to suppress insulin secretion. Glucose con-centration was increased to 33 mmol/L within an hourand then stopped. Reproduced with permission fromCaduff et al.179
5264_e12_p660-697 9/29/04 12:48 PM Page 688
Blood glucose was kept constant for 8 h infour healthy subjects in order to studychanges in the impedance pattern, whichmight occur over time. In 75% of the exper-iments glucose changes could be closelytracked. Only small changes occurred in thesensor signal over time.
The results of this human experimentshowed a proof of concept for this non-optical,NI monitoring approach.179 It is attractive as acontinuous glucose monitoring method. Be-cause of the indirect nature of this measure-ment, a number of questions remain to be clar-ified, such as the effect of body water contentor dehydration on this measurement. There isa dehydrating effect that accompanies hyper-glycemia.178 Some of the co-morbidities that af-fect cell membranes were listed in the sectiondiscussing RBC light scattering.
Tissue temperature measurements
Cho and Holzgreve184,185 proposed a differ-ent non-optical technique that is based on bodythermal effects. Measurement of body temper-ature was suggested as an NI method to de-termine glucose in the blood. The method isbased on measuring body temperature at an ex-tremity, such as the forefinger, using contactand non-contact temperature measuring tech-niques and correlating a mathematical func-tion of the measured temperature with glucoseconcentration. Body temperatures at extremi-ties depend on environmental temperature, ac-tivity, time from meal, alcohol and nicotineconsumption, and other disease states.184,185
Accurate measurement of changes in body tem-perature was claimed to yield NI glucose cali-bration plots.184,185 It is difficult to separate circadian temperature periodicity from tem-perature change because of the glucose con-centration effect. Thermal variations that char-acterize tissue metabolism have circadianperiodicity and are not related to glucose me-tabolism only, but can be associated with otherphysiological effects or disease states. Circa-dian variations in body temperature have beendiscussed.27,28 Temperature changes upon glu-cose ingestion.29 The tympanic membrane is abetter measurement site because of the absence
of the SC, calluses, and variable skin thickness,which can affect temperature measurements.
A new publication has described a methodfor combining temperature measurements withoptical measurements for NI determination ofglucose.186 The method is suggested to dependon measurement of blood flow, as gauged bytemperature, and blood oxygenation is deter-mined from the optical measurements. The var-ious physical parameters are used to calculateglucose concentration, presumably based onthe generated metabolic heat. The experimen-tal details and calibration methods and algo-rithm used are not described in this shortcommunication.186
The relationship between glucose concentra-tion and blood flow was not established. Thereare specific issues associated with blood flowin patients with diabetes. LDF studies showedan impaired decrease in cutaneous blood flowand difference in blood flow response to cool-ing or warming in patients with diabetes.47–49
Patients with diabetes exhibited differencesfrom subjects without diabetes in cutaneousblood flow,52–55 and in response to contralat-eral cooling.58 Both insulin and glucose are re-ported to have vascular activities affectingblood flow.62–64 The relationship between oxy-gen saturation and glucose concentration is notestablished in any of the previous studies. Inthe absence of corroborating studies, it is pos-sible to interpret the published Clarke errorgrid as representing an overmodeled calibra-tion relationship.
Table 7 summarizes the reported TGS,153
thermal emission,154 impedance frequency,179
temperature measurement,185 and the methodproposed by Ko et al.186 that combines temper-ature, blood flow and oxygen saturation mea-surement, and non-optical methods.
CONCLUSIONS AND AUTHOR’S VIEWS
In addition to the emergence of new detectionmethods, improvements in the measurementtechnologies and methods to reduce noise in NIglucose measurement were pursued. Advanceswere made in understanding and resolving thespecificity, compartmentalization, and calibra-tion issues of NI glucose measurements.
NON-INVASIVE GLUCOSE MEASUREMENT 689
5264_e12_p660-697 9/29/04 12:48 PM Page 689
Specificity of NI glucose measurements
The issue of specificity of measuring a glu-cose molecular property is still unresolved, atleast in a commercial product. Most NI tech-niques provide indirect evidence that the mea-sured signal is related to a molecular propertyof glucose. Multivariate techniques are used toextract glucose concentration and to establishcorrelation between signal and glucose valuesto compensate for lack of specificity; this is es-pecially true of NIR measurements, but to alesser extent for mid-IR spectroscopy. Theavailability of a large number of reported NIRstudies, and limited number of mid-IR studies,makes this conclusion rather tenuous. The in-terplay among in vitro experiments, in vivo ex-periments, and simulation studies is used toprove the specificity of some methods. OGTTand MTT results can have coincidental overlapwith physiologic events. A way to improve glu-cose correlation is to experimentally suppresscoincidental correlations and minimize mea-surement noise.
There are significant advances in methodsthat depend on measuring the effect of glucoseon properties of tissue or blood. The PA effecttook a different turn from measuring light ab-sorption to tracking scattering effects. Fluores-cence measurements appeared on the horizon.Fluorescence and photonic crystal diffractionmethods that depend on interaction betweenglucose and a specific binding molecule or a
metabolic pathway will have a specificity ad-vantage, when successfully applied in vivo.
Compartmentalization of glucose values
The precision and specificity of the NI mea-surements do not yet allow determination ofdifferences in glucose concentrations in bodycompartments. Several techniques target spe-cific body compartments. Transmission anddiffuse reflectance measurements track glucoseconcentrations in the vascular bed, while lo-calized reflectance and frequency domainmethods detect change in glucose concentra-tion in the ISF via its effect on �s. New meth-ods target the vascular compartment, the der-mis, tympanic membrane, and AH. Methodsthat depend on using a contact lens-type mem-brane will be dealing with the tears as a bodyfluid in the extracorporeal compartment.
Calibration of NI glucose devices
It is possible to track changes in glucose con-centration during a glycemic swing using cur-rent methods. Chemometric methods are nec-essary for data analysis in some technologies,and it is important to guard against overfitting.Calibration should allow a random spot testingas well as sequential continuous monitoring ofglucose, and be easily but not frequently up-dated. It should cover multiple physiologicaldisease, environmental, and activity condi-
KHALIL690
TABLE 7. MID-IR THERMAL EMISSION AND NONOPTICAL METHODS
Effect of glucoseon temperature,blood flow, andoxygensaturation
ISF fluid incutaneous layer,forearm
Tympanicmembrane
Vascular system,wrist
Finger
Finger
Calibration model
Calibration model,prediction
Calibration model
Calibration model
Calibration model
5264_e12_p660-697 9/29/04 12:48 PM Page 690
tions. Researchers should strive to achieve mul-tipatient calibration, which requires under-standing the physical and physiological factorsaffecting the signals, interperson differences,and measurement of noise. It is preferable tohave a self-calibrating detection method thatyields a universal calibration model, which isnot unique to an individual and does not re-quire multiple invasive data points and chemo-metric methods. Available clinical data do notshow this with great certainty for any method.
Most of the reported methods relied on hy-perglycemic excursion and multivariate cali-bration, thus raising two issues. The first oneis the accuracy of glucose invasive measure-ments and implanted sensors are problematicin the hypoglycemic range. This observationled Klonoff187 to suggest a different metric forthe hypoglycemic range. A second issue isthat multivariate analysis leads to one stan-dard error value across the whole glucoseconcentration range. This SEP will be smallerthan the commonly used standard deviationat high glucose concentration. But it will belarger than it is at low glucose concentration,leading to higher error in the hypoglycemicrange.
ACKNOWLEDGMENTS
Thanks to my colleagues at Abbott Labora-tories—Shu-jen Yeh, Charles Hanna, MichaelLowery, Ronald Hohs, Brenda Calfin, XiaomaoWu, Tzyy-Wen Jeng, Stan Kantor, James Babb,Gary Oosta, and Eric Shain—who helped menavigate through the minefield of NI glucosetesting.
REFERENCES
1. Khalil OS: Spectroscopic and clinical aspects of non-invasive glucose measurements. Clin Chem 1999;45:165–177.
3. Koschinsky T, Heinemann L: Sensors for glucosemonitoring: technical and clinical aspects. DiabetesMetab Res Rev 2001;17:113–123.
4. Gutman S, Bernhardt P, Pinkos A, Moxey-Mims M,Knott T, Cooper J: Regulatory aspects of noninva-sive glucose measurements. Diabetes Technol Ther2002;4:779–781.
5. Heinemann L, Koschinsky T: Continuous glucosemonitoring: an overview of today’s technologies andtheir clinical applications. Int J Clin Pract Suppl2002;129:75–79.
6. Koschinsky T, Jungheim K, Heinemann L: Glucosesensors and alternate site testing-like phenomenon:relationship between rapid blood glucose changesand glucose sensor signals. Diabetes Technol Ther2003;5:829–842.
7. Thennadil SN, Rennert JL, Wenzel BJ, Hazen KH,Ruchti TL, Block MB: Comparison of glucose con-centration in interstitial fluid, and capillary and ve-nous blood during rapid changes in blood glucoselevels. Diabetes Technol Ther 2001;3:357–365.
8. Boyne MS, Silver DM, Kaplan J, Saudek D: Timingof changes in interstitial and venous blood glucosemeasured with a continuous subcutaneous glucosesensor. Diabetes 2003;52:2790–2794.
9. Rebrin K, Steil GM, van Antwerp WP, MastrototaroJJ: Subcutaneous glucose predicts plasma glucose in-dependent of insulin: implications for continuousmonitoring. Am J Physiol 1999;277:E561–E571.
10. Maruo K, Tsurugi M, Tamura M, Ozaki Y: In vivononinvasive measurement of blood glucose by near-infrared diffuse-reflectance spectroscopy. ApplSpectrosc 2003;57:1236–1244.
11. Yeh S-j, Hanna CF, Khalil OS: Monitoring blood glu-cose changes in cutaneous tissue by temperature-modulated localized reflectance measurements. ClinChem 2003;49:924–934.
13. Rosenthal RD, Paynter LN, Mackie LH, inventors;Futrex, Inc., assignee: Non-invasive measurementsof blood glucose. US patent 5,028,787. July 2, 1991.
14. Rosenthal RD, Paynter LN, Mackie LH, inventors;Futrex, Inc., assignee: Non-invasive measurementsof blood glucose. US patent 5,086,229. February 4,1992.
15. Gabriely I, Wozniak R, Mevorach M, Kaplan J,Aharon Y, Shamoon H: Transcutaneous glucosemeasurement using near infrared spectroscopy dur-ing hypoglycemia. Diabetes Care 1999;22:2026–2032.
16. Finarov A, Kleinman Y, Fine I, inventors; OrsenseLtd., assignee: Optical device for non-invasive mea-surement of blood related signals utilizing a fingerholder. US patent 6,213,952. April 10, 2001.
17. Harjunmaa H, Peura PA, Mendelson Y, inventors;Worcester Polytechnic Institute, assignee: Methodand apparatus for measuring the concentration ofabsorbing substances. US patent 5,112,124. May 12,1992.
18. Peters RK, Elmerick D, inventors; Emerging Tech-nology Systems, assignee: Non-invasive glucosemeasuring device and method for measuring bloodglucose. US patent 5,910,109. June 8, 1999.
19. Burmeister JJ, Arnold MA, Small GW: Noninvasiveblood glucose measurements by near-infrared trans-
NON-INVASIVE GLUCOSE MEASUREMENT 691
5264_e12_p660-697 9/29/04 12:48 PM Page 691
mission spectroscopy across human tongues. Dia-betes Technol Ther 2000;2:5–16.
20. Heinemann L, Schmelzeisen-Redeker G: Non-inva-sive continuous glucose monitoring in Type I dia-betic patients with optical glucose sensors. Non-In-vasive Task Force (NITF). Diabetologia 1998;41:848–854.
21. Heise HM, Marbach R: Human oral mucosa studieswith varying blood glucose concentration by non-in-vasive ATR-FT-IR-spectroscopy. Cell Mol Biol(Noisy-le-grand) 1998;44:899–912.
22. Rabinovitch B, March WF, Adams RL: Noninvasiveglucose monitoring of the aqueous humor of the eye:I. Measurement of very small optical rotations. Dia-betes Care 1982;5:254–258.
23. Abreu MM, inventor: Contact device for applicationin direct apposition to the conjunctive of the eye. USpatent 6,312,393. November 6, 2001.
24. Martens H, Naes T: Multivariate Calibration. NewYork: John Wiley and Sons, 1989.
25. Haaland DM, Easterling RG: Application of newleast squares analysis methods for the quantitativeanalysis of multicomponent samples. Appl Spectrosc1982;36:665–673.
26. Haaland DM, Easterling RG, Vopicka DA: Multi-variate least-squares methods applied to quantita-tive spectral analysis of multicomponent samples.Appl Spectrosc 1985;39:73–84.
27. Seaman GV, Engel R, Swank RL, Hissen W: Circa-dian periodicity in some physicochemical parame-ters of circulating blood. Nature 1965;207:833–835.
29. Hillson RM, Hockaday TD: Facial and sublingualtemperature changes following intravenous glucoseinjection in diabetics. Diabet Metab 1982;8:15–19.
30. Clemens AH, Chang PH, Myers RW: The develop-ment of Biostator, a glucose controlled insulin infu-sion system (GCIIS). Horm Metab Res Suppl 1977;7:23–33.
31. DeFronzo RA, Tobin JD, Andres R: Glucose clamptechnique: a method for quantifying insulin secre-tion and resistance. Am J Physiol 1979;237:E214–E223.
32. Clemens AH, Hough DL, D’Orazio PA: Develop-ment of the Biostator glucose clamping algorithm.Clin Chem 1982;28:1899–1904.
33. Arnold MA, Burmeister JJ, Small GW: Phantom glu-cose calibration models from simulated noninvasivehuman near infrared spectra. Anal Chem 1998;70:1773–1781.
34. Samann A, Fischbacher CH, Jagemann KU, DanzerK, Schuler J, Papenkordt L, Muller UA: Non-inva-sive blood glucose monitoring by means of near in-frared spectroscopy: investigation of long-term ac-curacy and stability. Exp Clin Endocrinol Diabetes2000;108:406–413.
35. Wilson B: Measurement of tissue optical properties:methods and theories. In: Welch AJ, Van Gemert
MCC, eds.: Optical-Thermal Response of Laser-Irra-diated Tissue. New York: Plenum Press, 1995:233–261.
36. Murrer LH, Marijnissen HP, Star WM: Monte Carlosimulations for EndoBronchial Photodynamic Ther-apy: the influence of variations in optical and geo-metrical properties and of realistic and eccentriclight sources. Lasers Surg Med 1998;22:193–206.
37. de Mul FF, Steenbergen W, Greve J: Doppler MonteCarlo simulations of light scattering in tissue to sup-port laser-Doppler perfusion measurements. Tech-nol Health Care 1999;7:171–183.
38. Pavlin M, Jarm T, Miklavcic D: Monte-Carlo simu-lation of light transport for NIRS measurements intumors of elliptic geometry. Adv Exp Med Biol 2003;530:41–49.
39. Youcef-Toumi Y, Saptari VA: Noninvasive bloodglucose analysis using near infrared absorption spec-troscopy. Progress Report MIT Home Automationand Healthcare Consortium 1999. Available onlineat: http//darbeloffllab.mit.edu.home automation/report2-4/.
40. Graaff R, Aarnoudse JG, Zijp JR, Sloot PMA, de MulFFM, Greve J, Koelink MH: Reduced light-scatteringproperties for mixture of spherical particles: a sim-ple approximation derived from Mie calculations.Appl Opt 1992;31:1370–1376.
41. Vasko PD, Blackwell J, Koenig JL: Infrared and Ra-man spectroscopy of carbohydrates. I: Identificationof O-H and C-H vibrational modes for D-glucose,malose, cellobiose, and dextran by deuterium sub-stitution methods. Carbohydr Res 1971;19:297–310.
42. Vasko PD, Blackwell J, Koenig JL: Infrared and Ra-man spectroscopy of carbohydrates. II: Normal co-ordinate analysis of �-D-glucose. Carbohydr Res1972;23:407–417.
43. Lucassen GW, Caspers PJ, Puppels GJ: Infrared andRaman spectroscopy of human skin in vivo. In:Tuchin VV, ed. Handbook of Optical Biomedical Di-agnostics. Bellingham, WA: SPIE Press, 2002:787–823.
44. Lucassen GW, van Veen GNA, Jansen JAJ: Bandanalysis of hydrated human skin stratum corneumattenuated total reflectance Fourier transform in-frared spectra in vivo. J Biomed Opt 1998;3:267–280.
45. Brancaleon L, Bamberg MP, Sakamaki T, Kollias N:Attenuated total reflection-Fourier transform in-frared spectroscopy as a possible method to investi-gate biophysical parameters of stratum corneum invivo. J Invest Dermatol 2001;116:380–386.
46. Brancaleon L, Bamberg MP, Sakamaki T, Kollias N:Spectral differences between stratum corneum andsebaceous molecular components in the mid-IR.Appl Spectrosc 2000;54:1175–1182.
47. Rendell M, Bergman T, O’Donnell G, Drobny E, Bor-gos J, Bonnor R: Microvascular blood flow, volume,and velocity, measured by laser Doppler techniquesin IDDM. Diabetes 1989;38:819–824.
48. Rendell M, Bamisedun O: Diabetic cutaneous mi-croangiopathy. Am J Med 1992;93:611–618.
KHALIL692
5264_e12_p660-697 9/29/04 12:48 PM Page 692
49. Wilson SB: Detection of microvascular impairmentin type-1 diabetes by laser Doppler flowmetry. ClinPhysiol 1992;12:195–208.
50. Braverman IM, Keh-Yen A: Ultrastructural abnor-malities of the microvasculature and elastic fibers inthe skin of juvenile diabetics. J Invest Dermatol1984;82:270–274.
51. Rayman G, Williams SA, Spenser PD, Smajie LH,Wise PH, Tooke JE: Impaired microvascular hyper-emic response to minor skin trauma in type-1 dia-betes. Br Med J (Clin Res Ed) 1986;292:1295–1298.
52. Khan F, Elhadd T, Green SA, Belch J: Impaired skinmicrovascular function in children, adolescent, andyoung adults with type-1 diabetes. Diabetes Care2000;23:215–220.
53. Tooke JE, Ostergren J, Lins P-E, Fragrell B: Skin mi-crovascular blood flow control in long duration di-abetics with and without complications. DiabetesRes 1987;5:189–192.
54. Salomaa V, Riley W, Kark JD, Nardo C, Folsom AR:Non-insulin-dependent diabetes mellitus and fast-ing glucose and insulin concentrations are associatedwith arterial stiffness indexes. The ARIC Study. Ath-erosclerosis Risk in Communities Study. Circulation1995;91:1432–1443.
55. Hassan AAK, Tooke J: Effect of change in local skintemperature on postural vasoconstriction in man.Clin Sci 1988;74:201–206.
57. Stansberry KB, Shapiro SA, Hill MA, McNitt PM,Meyer MD, Vinik AI: Impaired peripheral vasomo-tion in diabetes. Diabetes Care 1006;19:715–721.
58. Haak E, Haak T, Grozinger Y, Krebs G, Usadel KH,Kusterer K: The impact of contralateral cooling ofskin capillary blood cell velocity in patients with di-abetes mellitus. J Vasc Res 1998;35:245–249.
59. Akbari CM, Saouaf R, Barnhill DF, Newman PA,LoGerfo FW, Veves A: Endothelium-dependent va-sodilatation is impaired in both microcirculation andmacrocirculation during acute hyperglycemia. JVasc Surg 1998;28:687–694.
60. Caballero AE, Arora S, Saouaf R, Lim SC, Sma-kowski P, Park JY, King GL, LoGerfo FW, HortonES, Veves A: Microvascular and macrovascular re-activity is reduced in subjects at risk for type-2 dia-betes. Diabetes 1999;48:1856–1862.
61. Wiernsperger N: Defects in microvascular heamo-dynamics during prediabetes: Contributor orepiphenomenon? Diabetologia 2000;43:1439–1449.
62. Williams SB, Goldfine AB, Timimi FK. Ting HH,Roddy M-A, Simonson DC, Creager MA: Acute hy-perglycemia attenuates endothelium-dependent va-sodilation in humans in vivo. Circulation 1998;97:1695–1701.
64. Oomen PHN, Kant GD, Dullaart RPF, Reitsma WD,Smit AJ: Acute hyperglycemia and hyperinsulinemiaenhance vasodilatation in Type-1 diabetes mellituswithout increasing capillary permeability and in-ducing endothelial dysfunction. Microvasc Res 2002;63:1–9.
65. MacGregor AR, inventor: Method and apparatus fornon-invasively determining blood analytes. USpatent 5,617,852. April 8, 1997.
66. Rendell M, Hovelson C, O’Connor K, Cheung L,Huard S, Kong TS, Catania A, Rosenthal R: Deter-mination of blood flow in the finger using near-in-frared spectroscopy. Clin Physiol 1998;18:426–434.
67. Wu X, Yeh S-j, Jeng T-W, Khalil OS: Noninvasive de-termination of hemoglobin and hematocrit using atemperature-controlled localized reflectance tissuephotometer. Anal Biochem 2000;287:284–293.
68. Rendell M, Anderson E, Schlueter W, Mailliard J,Honigs D, Rosenthal R: Determination of hemoglo-bin levels in the finger using near infrared spec-troscopy. Clin Lab Haematol 2003;25:93–97.
69. Sibbald RG, Landolt SJ, Toth D: Skin and diabetes.Endocrinol Metab Clin North Am 1996;25:463–472.
70. Monnier VM, Bautista O, Kenny D, Sell DR, FogartyJ, Dahms W, Cleary PA, Lachin J, Genuth S: Skin col-lagen glycation, glycoxidation, and crosslinking arelower in subjects with long-term intensive versus con-ventional therapy of type-1 diabetes: relevance of gly-cated collagen products versus HbA1c as markers fordiabetic complications. Diabetes 1999;48:870–880.
71. James VJ, Belbridge L, McLennan SV, Yue DK: Useof X-ray diffraction in study of human diabetic andaging collagen. Diabetes 1999;40:391–394.
72. Firsov NN, Priezzhev AV, Ryaboshapka OM, SirkoIV: Aggregation and disaggregation of erythrocytesin whole blood: study by back scattering technique.J Biomed Opt 1999;4:76–84.
77. Fusman R, Rotstein R, Elishkewich K, Zeltser D, Co-hen S, Kofler M, Avitzour D, Arber N, Berliner S,Shapirea I: Image analysis for the detection of in-creased erythrocyte, leukocyte and platelet adhe-siveness/aggregation in the peripheral blood of pa-tients with diabetes mellitus. Acta Diabetol 2001;38:129–134.
NON-INVASIVE GLUCOSE MEASUREMENT 693
5264_e12_p660-697 9/29/04 12:48 PM Page 693
78. Cicha I, Suzuki Y, Tateishi N, Maeda N: Enhance-ment of red blood cell aggregation by plasma triglyc-erides. Clin Hemorheol Microcirc 2001;24:247–255.
79. Ercan M, Konukoglu D, Erdem T, Onen S: The ef-fects of cholesterol levels on hemorheological pa-rameters in diabetic patients. Clin Hemorheol Mi-crocirc 2002;26:257–263.
80. Yedgar S, Koshkaryev A, Barshtein G: The red bloodcell in vascular occlusion. Pathophysiol HaemostThromb 2002;32:263–268.
81. Ami RB, Barshstein G, Zeltzer D, Goldberg Y,Shapira I, Roth A, Keren G, Miller H, Prochorov V,Eldor A, Berliner S, Yedgar S: Parameters of redblood cell aggregation as correlates of the inflam-matory state. Am J Physiol Heart Circ Physiol2001;280:H1982–H1988.
82. Firsov NN, Bjelle A, Korotaeva TV, Priezzhev AV,Ryaboshapka OM: Clinical applications of the mea-surements of spontaneous erythrocyte aggregationand disaggregation. A pilot study. Clin HemorheolMicrocirc 1998;18:87–97.
83. Yeh S-j, Khalil OS, Hanna CF, Kantor S, Wu X, JengT-W, Bolt RA: Temperature dependence of opticalproperties of in-vivo human skin. Proc SPIE 2001;4250:455–461.
84. Khalil OS, Yeh S-j, Lowery MG, Wu X, Hanna CF,Kantor S, Jeng T-W, Kanger J, Bolt RA, de Mul FF:Temperature modulation of optical properties of hu-man skin. J Biomed Opt 2003;8:191–205.
85. Yeh S-j, Khalil OS, Hanna CF, Kantor S: Near in-frared thermo-optical response of the localized re-flectance of intact diabetic and non-diabetic humanskin. J Biomed Opt 2003;8:534–544.
86. Khalil OS, Yeh S-j, Wu X: Response of near IR local-ized reflectance signals of intact diabetic human skinto thermal stimuli. Proc SPIE 2003;5086:142–148.
87. Yeh S-j, Hanna CF, Kantor S, Hohs R, Khalil OS: Dif-ferences in thermal optical response between intactdiabetic and no diabetic human skin. Proc SPIE2003;4958:213–224.
88. Elsner P, Berardesca E, Maibach HI, eds.: Bioengi-neering of the Skin: Water and the StratumCorneum, Vol. 1. Boca Raton, FL: CRC Press, 1994.
96. Janatsch G, Kruse-Jarres JD, Marbach R, Heise HM:Multivariate calibration for assays in clinical chem-istry using attenuated total reflection infrared spec-tra of human blood plasma. Anal Chem 1989;61:2016–2023.
97. Deissel F, Willman S, Kamphaus P, Kurte R, DammU, Heise HM: Glucose quantification in dried-downsamples using mid-infrared attenuated total reflec-tion spectroscopy. Appl Spectrosc 2004;58:422–450.
98. Petrich W, Dolenk B, Früh J, Ganz M, Greger H, Ja-cob S, Keller F, Nikulin AE, Otto, M, Quarder O, So-morjai R.L, Staib A, Warner G, Wielinger H: Diseasepattern recognition in infrared spectra of human serawith diabetes mellitus as an example. Appl Opt 2000;39:3372–3379.
99. Kim YJ, Hahn S, Yoon G: Determination of glucosein whole blood samples by mid-infrared spec-troscopy. Appl Opt 2003;42:745–749.
100. Shen YC, Davies AG, Linfield EH, Elsey TS, TadayPF, Arnone AD: The use of Fourier-transform in-frared spectroscopy for the quantitative determina-tion of glucose concentration in whole blood. PhysMed Biol 2003;48:2023–2032.
101. Haaland DM, Robinson MR, Koepp GW, ThomasEV, Robinson PL, Eaton RP: Reagentless near in-frared determination of glucose in whole blood us-ing multivariate calibration. Appl Spectrosc1992;46:1575–1578.
103. Berman HL, Roe JN, inventors; MedOptix, Inc., as-signee: Cleaning system for infrared ATR glucosemeasurement system (II). US patent 6,362,144. March26, 2002.
104. Berman HL, Roe JN, inventors; MedOptix, Inc., as-signee: Infrared ATR glucose measurement systemhaving an ATR with mirrored ends. US patent6,421,548. July 16, 2002.
105. Berman HL, Roe JN, inventors; MedOptix, Inc., as-signee: Method for preparing skin surface and de-termining glucose levels from that system. US patent6,424,848. July 23, 2002.
106. Bruulsema JT, Hayward JE, Farrell T, Patterson M,Heinemann L, Berber M: Correlation between bloodglucose concentration in diabetics and noninvasivelymeasured tissue optical scattering coefficient. OptLett 1997;22:190–192.
107. Heinemann L, Kramer U, Klotzer HM, Hein M, VolzD, Hermann M, Heise T, Rave K: Non-Invasive TaskForce: noninvasive glucose measurement by moni-
KHALIL694
5264_e12_p660-697 9/29/04 12:48 PM Page 694
toring of scattering coefficient during oral glucosetolerance tests. Diabetes Technol Ther 2000;2:211–220.
108. Maier J, Walker S, Fantini S, Franceschini M, Grat-ton E: Non-invasive glucose determination by mea-suring variations of the reduced scattering coeffi-cient of tissues in the near-infrared. Opt Lett1994;19:2062–2064.
109. Bulow J, Astrup A, Christensen NJ, Kastrup J: Bloodflow in skin, subcutaneous adipose tissue and skele-tal muscle in the forearm of normal man during anoral glucose load. Acta Physiol Scand 1987;130:657–661.
110. Shvartsman LD, Fine I: Light scattering changescaused by RBC aggregation; physical basis for newapproach to non-invasive blood count. Proc SPIE2001;4263:131–142.
111. Cohen O, Fine I, Monashkin E, Karasik A: Glucosecorrelation with light scattering patterns—a novelmethod for non-invasive glucose measurements. Di-abetes Technol Ther 2003;5:11–17.
112. Shvartsman LD, Fine I: Optical transmission ofblood: effect of erythrocyte aggregation. IEEE TransBiomed Eng 2003;50:1026–1033.
113. Elishkevitz K, Fusman R, Koffler M, Shapirea I,Zeltser D, Avitzour D, Arber N, Berliner S, RotsteinR: Rheological determinations of red blood cell ag-gregation in diabetic patients in relation to theirmetabolic control. Diabet Med 2002;19:152–156.
115. Schmitt JM: Optical coherence tomography (OCT): areview. IEEE J Selected Top Quantum Electron 1999;5:1205–1215.
116. Knuttel A, Bohlau-Gadau M, Rist M, Welker P,Gogolin U, Kugler C: In vivo evaluation of layeredscattering coefficients and refractive indices withOCT. Proc SPIE 1999;3597:324–334.
117. Knuttel A, Boehlau-Godau M: Spatially confinedand temporally resolved refractive index scatteringevaluation of human skin performed with optical co-herence tomography. J Biomed Opt 2000; 5:83–92.
118. Essenpreis M, Knuttel A, Boecker D, inventors;Boehringer Mannheim GmbH, assignee: Methodand apparatus for determining glucose concentra-tion in a biological sample. US patent 5,710,630. Jan-uary 20, 1998.
119. Larin KV, Eledrisi MS, Motamedi M, Esenaliev RO:Noninvasive blood glucose monitoring with opticalcoherence tomography: a pilot study in human sub-jects. Diabetes Care 2002;25:2263–2267.
120. Esenaliev RO, Larin KV, Larina IV, Motamedi M:Noninvasive monitoring of glucose concentrationwith optical coherence tomography. Opt Lett 2001;26:992–994.
121. Larin K, Larina I, Motamedi M, Gelikonov Y, Kura-nov R, Esenaliev RO: Potential application of opticalcoherence tomography for non-invasive monitoring
of glucose concentrations. Proc SPIE 2001;4263:83–90.
122. Larin KV, Motamedi M, Ashitkov TV, Esenaliev RO:Specificity of noninvasive blood glucose sensing us-ing optical coherence tomography technique: a pilotstudy. Phys Med Biol 2003;48:1371–1390.
123. Rabinovitch B, March WF, Adams RL: Noninvasiveglucose monitoring of the aqueous humor of the eye:Part I. Measurement of very small optical rotations.Diabetes Care 1982;5:254–258.
124. Rawer R, Stork W, Muller-Glaser KD: Polarimetricmethods for measurement of intra ocular glucoseconcentration. Biomed Tech (Berl) 2002;47(Suppl 1Pt 1):186–188.
125. Cameron BD, Gorde HW, Satheesan B, Cote GL: Theuse of polarized laser light through the eye for non-invasive glucose monitoring. Diabetes Technol Ther1999;1:135–143.
126. Cameron BD, Baba JS, Cote GL: Measurement of theglucose transport time delay between the blood andaqueous humor of the eye for the eventual develop-ment of a noninvasive glucose sensor. Diabetes Tech-nol Ther 2001;3:201–207.
128. Lane JT, Toris CB, Nakhle SN, Chacko DM, WangYL, Yablonski ME: Acute effects of insulin on aque-ous humor flow in patients with type-1 diabetes. AmJ Ophthalmol 2001;132:321–327.
129. Bockle S, Rovati L, Ansari RR: Polarimetric glucosesensing using Brewster-reflection off the eye lens:theoretical analysis. Proc SPIE 2002;4624:160–164.
130. Hanlon EB, Manoharan R, Koo TW, Shafer KE, MotzJT, Fitzmaurice M, Kramer JR, Itzkan I, Dasari RR,Feld MS: Prospects for in vivo Raman spectroscopy.Phys Med Biol 2000;45:R1–R59.
131. Steffes PG: Laser-based measurement of glucose inthe ocular aqueous humor: an efficacious portal fordetermination of serum glucose levels. DiabetesTechnol Ther 1999;1:129–133.
132. Berger AJ, Itzkan I, Feld MS: Feasibility of measur-ing blood glucose concentration by near-infrared Ra-man spectroscopy. Spectrochim Acta A Mol BiomolSpectrosc 1997;53A:287–292.
133. Berger AJ, Koo TW, Itzkan I, Feld MS: An enhancedalgorithm for linear multivariate calibration. AnalChem 1998:70;623–627.
134. Koo TW, Berger AJ, Itzkan I, Horowitz G, Feld MS:Reagentless blood analysis by near-infrared Ramanspectroscopy. Diabetes Technol Ther 1999;1:153–157.
spectroscopic study of mobile and static tissues:blood chemistry. Proc SPIE 2000;3918:135–143.
138. Chaiken J, Finney WF, Knudson PE, Peterson KP, Pe-terson CM, Yang X, Weinstock RS: Progress in non-invasive, in-vivo, tissue modulated Raman spec-troscopy of human blood. Proc SPIE 2001;4254:216–227.
139. Haynes CL, Van Duyne RP: Plasmon-sampled sur-face-enhanced Raman excitation spectroscopy. JPhys Chem B 2003;107:7426–7433.
140. Schatz GC, Van Duyne RP: Electromagnetic mechanism of surface-enhanced spectroscopy. In:Chalmers JM, Griffiths PR, eds. Handbook of Vibra-tional Spectroscopy, Vol. 1. Wiley: New York, 2002:759–774.
141. Shafer-Peltier KE, Haynes CL, Glucksberg MR, VanDuyne RP: Toward a glucose biosensor based on sur-face-enhanced Raman scattering. J Am Chem Soc2003;125:588–593.
142. Yonzon CR, Haynes CL, Zhang X, Walsh JT, VanDuyne RP: A glucose biosensor based on surface-en-hanced Raman scattering: improved partition layer,temporal stability, reversibility, and resistance toserum protein interference. Anal Chem 2004;76:78–85.
143. Tam AC, Patel CKN: Optical absorption of light andheavy water by laser photoacoustic spectroscopy.Appl Opt 1979;18:3348–3357.
144. MacKenzie HA, Ashton HS, Spiers S, Shen Y, Free-born SS, Hannigan J, Lindberg J, Rae P: Advances inphotoacoustic noninvasive glucose testing. ClinChem 1999;45:1587–1595.
145. Fainchtein R, Stoyanov BJ, Murphy JC, Wilson DA,Hanley DF: Local determination of hemoglobin con-centration and degree of oxygenation in tissue bypulsed photoacoustic spectroscopy. Proc SPIE 2000;3916:19–33.
146. Bednov AA, Karabutov AK, Savateeva EV, MarchWF, Oraevsky AA: Monitoring glucose in vivo bymeasuring laser-induced acoustic profiles. Proc SPIE2000;3916:9–18.
147. Zhao Z, Myllylä R: Photoacoustic determination ofglucose concentration in whole blood by a near-in-frared laser diode. Proc SPIE 2001;4256:77–83.
148. Zhao Z: Pulsed photoacoustic techniques and glu-cose determination in human blood and tissue[Ph.D. Dissertation]. Oulu, Finland: University ofOulu, 2002. Available online at: http://herkules.oulu.fi/isbn9514266900/index.html.
149. Shen Y, Lu Z, Spiers S, MacKenzie HA, Ashton HS,Hannigan J, Freeborn SS, Lindberg J: Measurementof the optical absorption coefficient of a liquid by useof a time-resolved photoacoustic technique. ApplOpt 2000;39:4007–4012.
150. Glucon, Inc.: Continuous non-invasive venous bloodglucose monitor. A presentation at Diabetes Tech-nology Meeting, Atlanta, November 2002. Availableonline at: www.Glucon.com.
151. Nagar R, Pesach B, Ben-Ami U, inventors: Photoa-coustic assay and imaging system. US patent appli-cation 20030167002. September 4, 2003.
152. Hoelen CGA, de Mul FFM, Pongers R, Dekker A:Three-dimensional photoacoustic imaging of bloodvessels in tissue. Opt Lett 1998;23:648–650.
160. Snyder WJ, Grundfest WS, inventors; Cedars-SinaiMedical Center, assignee: Glucose monitoring appa-ratus and method using laser-induced emissionspectroscopy. US patent 6,232,609. May 15, 2001.
161. Russell RJ, Pishko MV, Gefrides CC, McShane MJ,Cote GL: A fluorescence-based glucose biosensor us-ing concanavalin A and dextran encapsulated in apoly(ethylene glycol) hydrogel. Anal Chem 1999;71:3126–3132.
162. McCartney LJ, Pickup JC, Rolinski OJ, Birch DJS:Near-infrared fluorescence lifetime assay for serumglucose based on allophycocyanin-labeled con-canavalin-A. Anal Biochem 2001;292:216–221.
163. Rolinski OJ, Birch DJS, McCartney LJ, Pickup JC: Flu-orescence nanotomography using resonance energytransfer: demonstration with a protein-sugar com-plex. Phys Med Biol 2001;46:N221–N226.
164. Evans ND, Gnudi L, Rolinski OJ, Birch DJ, PickupJC: Non-invasive glucose monitoring by NAD(P)Hautofluorescence spectroscopy in fibroblasts andadipocytes: a model for skin glucose sensing. Dia-betes Technol Ther 2003;5:807–816.
165. Masters BR, So PT, Gratton E: Multiphoton excita-tion fluorescence microscopy and spectroscopy of invivo human skin. Biophys J 1997;72:2405–2412.
166. Dong CY, Yu B, Kaplan PD, So PT: Performances ofhigh numerical aperture water and oil immersion
KHALIL696
5264_e12_p660-697 9/29/04 12:48 PM Page 696
objective in deep-tissue, multi-photon microscopicimaging of excised human skin. Microsc Res Tech2004;63:81–86.
167. Swindle LD, Thomas SG, Freeman M, Delaney PM:View of normal human skin in vivo as observed us-ing fluorescent fiber-optic confocal microscopicimaging. J Invest Dermatol 2003;121:706–712.
169. March WF: Apparatus for measuring blood glucoseconcentrations. Patent publication WO 2002087429published under the International Patent Treaty.
170. Badugu R, Lakowicz JR, Geddes CD: Noninvasivecontinuous monitoring of physiological glucose us-ing a monosaccharide-sensing contact lens. AnalChem 2004;76:610–618.
171. Trepagnier P, Freeman J, Mansfield J, Brand D, Hop-meier MJ, Kollias N, inventors, Argose, Inc., as-signee: Non-invasive tissue glucose level monitor-ing. US patent 6,721,582. April 13, 2004.
172. Holtz JH, Asher SA: Polymerized colloidal crystalhydrogel films as intelligent chemical sensing mate-rials. Nature 1997;389:829–832.
173. Reese E, Baltusavich ME, Keim JP, Asher SA: De-velopment of an intelligent polymerized crystallinecolloidal array colorimetric reagent. Anal Chem2001;73:5038–5042.
174. Asher SA, Sharma AC, Goponenko AV, Ward MM:Photonic crystal aqueous metal cation sensing ma-terials. Anal Chem 2003;75:1676–1683.
175. Sharma AC, Jana T, Kesavamoorthy R, Shi L, VirjiMA, Finegold DN, Asher SA: A general photoniccrystal sensing motif: creatinine in bodily fluids. JAm Chem Soc 2004;126:2971–2977.
180. Jenkins PG, Larmors C: Hyperglycemia-induced hy-ponatremia. N Engl J Med 1974;290:573.
181. Ermolina I, Polevaya Y, Feldman Y: Analysis of di-electric spectra of eukaryotic cells by computer mod-eling. Eur Biophys J 2000;29:141–145.
182. Polevaya Y, Ermolina I, Schlesinger M, Ginzburg BZ,Feldman Y: Time domain dielectric spectroscopystudy of human cells. II. Normal and malignantwhite blood cells. Biochim Biophys Acta 1999;1419:257–271.
183. Hayashi Y, Livshits L, Caduff A, Feldman Y: Di-electric spectroscopy study of specific glucose influ-ence on human erythrocyte membrane. J Phys D2003;36:369–374.
184. Cho O-K, Holzgreve B, inventors; O-K Cho, as-signee: Process and device for non-invasive deter-mination of glucose in parts of the human body. USpatent 5,795,305. August 18, 1998.
185. Cho OK, Holzgreve B, inventors; OK Kyung Cho, as-signee: Process and device for determining the ex-change of heat between the human body and the in-vented device and its correlation to the glucoseconcentration in human blood. US patent 5,924,996.July 20, 1999.
186. Ko JB, Cho OK, Kim YO, Yasuda K: Body metabo-lism provides a foundation for noninvasive bloodglucose monitoring. Diabetes Care 2004;27:1211–1212.
187. Klonoff DC: The need for separate performancegoals for glucose sensors in the hypoglycemic, nor-moglycemic, and hyperglycemic ranges. DiabetesCare 2004;27:834–836.
Address reprint requests to:Omar S. Khalil, Ph.D.
Diagnostics DivisionAbbott Laboratories
D9MS, AP 20100 Abbott Park RoadAbbott Park, IL 60064
![Glucose Sensing for Diabetes Monitoring: Recent Developments · 2017-08-15 · glucose sensing approaches studied over the past decade [9]. Spectroscopic methods for non-invasive](https://static.documents.pub/doc/80x56/5ec91add4f083434fa4c95f6/glucose-sensing-for-diabetes-monitoring-recent-developments-2017-08-15-glucose.jpg)
