speciﬁcity of noninvasive blood glucose sensing using ...bol.egr.uh.edu › sites › bol ›...

INSTITUTE OF PHYSICS PUBLISHING PHYSICS IN MEDICINE AND BIOLOGY

Phys. Med. Biol. 48 (2003) 1371–1390 PII: S0031-9155(03)58795-4

Specificity of noninvasive blood glucose sensing usingoptical coherence tomography technique: a pilot study

Kirill V Larin1,2, Massoud Motamedi2, Taras V Ashitkov1,2

and Rinat O Esenaliev1,2,3,4

1 Laboratory for Optical Sensing and Monitoring, The University of Texas Medical Branch,Galveston, TX 77555-0456, USA2 Center for Biomedical Engineering, The University of Texas Medical Branch, Galveston,TX 77555-0456, USA3 Department of Physiology and Biophysics, The University of Texas Medical Branch,Galveston, TX 77555-0456, USA4 Department of Anesthesiology, The University of Texas Medical Branch, Galveston,TX 77555-0456, USA

E-mail: [email protected]

Received 27 January 2003Published 7 May 2003Online at stacks.iop.org/PMB/48/1371

AbstractNoninvasive monitoring of blood glucose concentration in diabetic patientswould significantly reduce complications and mortality associated with thisdisease. In this paper, we experimentally and theoretically studied specificity ofnoninvasive blood glucose monitoring with the optical coherence tomography(OCT) technique. OCT images and signals were obtained from skin ofYucatan micropigs and New Zealand rabbits. Obtained results demonstrate that:(1) several body osmolytes may change the refractive index mismatch betweenthe interstitial fluid (ISF) and scattering centres in tissue, however the effectof the glucose is approximately one to two orders of magnitude higher; (2) anincrease of the ISF glucose concentration in the physiological range (3–30 mM)may decrease the scattering coefficient by 0.22% mM−1 due to cell volumechange; (3) stability of the OCT signal slope is dependent on tissueheterogeneity and motion artefacts; and (4) moderate skin temperaturefluctuations (±1 ◦C) do not decrease accuracy and specificity of the OCT-based glucose sensor, however substantial skin heating or cooling (several ◦C)significantly change the OCT signal slope. These results suggest that the OCTtechnique may provide blood glucose concentration monitoring with sufficientspecificity under normal physiological conditions.

(Some figures in this article are in colour only in the electronic version)

0031-9155/03/101371+20$30.00 © 2003 IOP Publishing Ltd Printed in the UK 1371

http://stacks.iop.org/pb/48/1371

1372 K V Larin et al

1. Introduction

Diabetes mellitus is a chronic systemic metabolic disease that affects approximately17 million people in the USA (or 6.2% of the population) and more than 140 millionpeople worldwide (this number is expected to rise to almost 300 million by the year2025). Along with its complications, including cardiac diseases, stroke, high bloodpressure and blindness, this disease results in about 200,000 deaths in the USA everyyear (National Institute of Diabetes and Digestive and Kidney Diseases 1994, AmericanDiabetes Association 1998).

Diabetes develops when the pancreas produces insufficient amounts of insulin to meet thebody’s needs (insulin-dependent diabetes mellitus (IDDM) or type I diabetes), or when thepancreas produces insulin, but the cells are unable to efficiently use it in order to metabolizeglucose (non-insulin-dependent diabetes mellitus (NIDDM) or type II diabetes). As a result,a high and potentially dangerous concentration of glucose can be accumulated in blood(hyperglycemia). In addition, recent studies demonstrate that intensive insulin therapy ofpatients with type I diabetes could significantly increase the risk of severe hypoglycemia(Diabetes Control and Complications Trial Research Group 1997), a dangerous conditionsometimes resulting in coma or even death.

Glucose is a monosaccharide sugar with chemical formula C6H12O6. The normal rangeof glucose concentration in human blood is 70 mg dl−1 to 160 mg dl−1 (3.9–8.9 mM, 1 mM =18.0 mg dl−1) depending on time of last meal, extent of physical tolerance and other factors.Monitoring of the glycemic status in patients with diabetes is considered a cornerstone oftheir care. Results from monitoring are used to assess the efficacy of therapy and to guideadjustments in medical nutrition therapy, exercise and medications to achieve the best possibleglucose control to prevent hyper- or hypoglycemia (National Institute of Diabetes and Digestiveand Kidney Diseases 1994, Goldstein et al 1995).

At present, a widely used method of self-monitoring of blood glucose (SMBG) involvesdetermination of blood glucose concentration with devices utilizing chemical analysis ofblood samples taken by puncturing a finger. Even though SMBG has revolutionized themanagement of diabetes, discomfort and inconvenience are barriers for effective complianceand, therefore, optimum management (Goldstein et al 1995, National Steering Committeefor Quality Assurance in Capillary Blood Glucose Monitoring 1993). These drawbacks maylimit the number of blood glucose measurements performed by patients with diabetes and,therefore, result in improper management of the disease. Therefore, there is a great demandfor a noninvasive method for frequent or continuous blood glucose monitoring, which wouldconsiderably improve the quality of life for diabetic patients, improve compliance for glucosemonitoring, and reduce complications and mortality associated with the disease.

Significant efforts have been made by several scientific groups and companies in thepast two decades in order to develop a biosensor for noninvasive blood glucose analysisbased on different optical approaches (reviewed in Cote (1997)). These approaches includeNIR absorption and scattering (Pan et al 1996, Gabriely et al 1999, Robinson et al 1992,Maier et al 1994), polarimetry (Rabinovitch et al 1982, Cote et al 1992), Raman spectroscopy(Goetz et al 1995) and photoacoustics (MacKenzie et al 1999). Despite significantefforts, the described techniques for noninvasive monitoring of glucose concentration havefaced limitations associated with low sensitivity, accuracy and insufficient specificity ofglucose concentration measurement at physiological range (4–30 mM or 72–540 mg dl−1).The invasive devices for home use have accuracy of 20% (Nettles 1993). A noninvasiveglucose sensor should have at least the same or better accuracy and have an acceptablespecificity.

Specificity of blood glucose sensing using OCT 1373

1.1. Mechanism of glucose-induced changes in tissue scattering

An important ability of glucose to change the tissue scattering coefficient, µs, is based onits osmotic properties (Tuchin et al 1997). Changes in tissue scattering are more specificallyattributed to glucose as an osmolyte than changes in tissue absorption spectra due to thepresence of glucose as a chromophore in the NIR spectral range. The scattering coefficient oftissues is dependent on the refractive index, n, mismatch between the extracellular fluid (ECF)and the cellular components. In the NIR spectral range, the index of refraction of ECF is1.35–1.36, whereas the index of refraction of the cellular components and protein aggregatesis in the range of 1.35–1.41 (Duck 1990, Drezek et al 1999).

In a simple model of scattering dielectric spheres, the reduced scattering coefficient canbe approximately calculated as (Graaff et al 1992)

µ′s = 3.28πr2ρs

(2πr

λ

)0.37 (ns

nmedium− 1

)2.09

(1)

where r is the sphere radius, ρs is the volume density of the spheres, λ is the wavelength ofthe incident light, and ns and nmedium are the refractive indices of scatterers and surroundingmedium, respectively. This equation is valid for non-interacting Mie scatterers and g > 0.9;5 < 2πr/λ < 50; 1 < ns/nmedium < 1.1. If the refractive index of scatterers remains the sameand is higher than the refractive index of the medium, the increase of glucose concentrationin the medium reduces the refractive index mismatch, �n = ns − nmedium, and, hence, thescattering coefficient is also reduced:

µ′s = 3.28πr2ρs

(2πr

λ

)0.37 (ns

nmedium + δnglucose− 1

)2.09

(2)

where δnglucose is the glucose-induced increase of nmedium.The same is true for the tissues: the increase of glucose concentration in the ECF reduces

the refractive index mismatch, δn = ns − nECF, and, hence, the tissue scattering coefficient isalso reduced:

ECF + GLUCOSE → �n↓ → µs↓. (3)

Therefore, the increase of blood glucose concentration will raise the refractive index ofthe ECF that will decrease the scattering coefficient of the tissue as a whole. Thiseffect was studied by Kohl et al in tissue-simulating phantoms (Kohl et al 1995), Tuchinet al in vitro (Tuchin et al 1997), and Maier et al and Bruulsema et al in vivo in humanvolunteers (Maier et al 1994, Bruulsema et al 1997). They demonstrated (0.05–0.1)% mM−1

decrease of the scattering coefficient in human muscle and adipose tissue (Kohl et al 1994),0.6% mM−1 in muscle tissue of human thigh with the relative error of �µs

µs

∼= ±0.4%

(Maier et al 1994), and (0.11–0.34)% mM−1 in human adipose tissue with the relative errorof �µs

µs

∼= ±0.3% (Bruulsema et al 1997). Results of these studies showed a potential for theoptical methods to detect small changes in tissue scattering as a function of blood glucoseconcentration. However, the detection of diffusively scattered photons results in low sensitivityand accuracy of glucose concentration measurements due to integration of the signal over theentire optical path in tissues. A blood glucose-monitoring system needs to be more sensitiveand accurate than the systems tested in these studies.

Recently, we proposed to use the optical coherence tomography (OCT) technique fornoninvasive, accurate, and sensitive monitoring of the blood glucose concentration in skin(Esenaliev et al 2001, Larin et al 2002). The OCT technique has several advantages overother optical methods proposed for noninvasive measurement and monitoring of tissueoptical properties such as high resolution and capability of probing specific tissue layers.


The skin consists of three major layers: dead keratinized layer of squames (stratum corneumof epidermis); prickle cells layer (epidermis) and connective tissue of dermis. The OCTtechnique allows measurement of glucose-induced changes in the skin scattering coefficientdirectly from the dermis, the only layer with the developed blood microvessel network.

1.2. Optical coherence tomography technique

OCT is a new imaging technique that provides images of tissues with resolution of about10 µm or less (8–25 times greater than any existing ultrasound modality) at a depth of up to1 mm depending on optical properties of tissue. This technique was introduced in 1991 toperform tomographic imaging of the human eye (Huang et al 1991). Since then OCT is beingactively developed by several research groups for many diagnostic applications.

In its most basic configuration, OCT system consists of a simple Michelson interferometerwith a low temporal coherence laser source in one arm, in-depth scanning system in the second‘reference’ arm, an object under study in the third ‘sample’ arm and registering photodiode inthe fourth arm. Usually, the sample arm has additional scanning system that allows formationof cross-sectional two- and three-dimensional images of tissues. The intereferometric signalsin OCT system can be formed only when the optical path length in the sample arm matchesthat in the reference arm within the coherence length of the source. Therefore, the coherencelength of the source and the group refractive index of tissues will determine the in-depthresolution of the OCT system.

Attenuation of light intensity for ballistic photons, I, in a medium with scattering andabsorption is described by the Beer–Lambert law:

I = I0 e−µt z (4)

where µt = µs + µa is the attenuation coefficient of ballistic photons, µs and µa are thescattering and absorption coefficients, respectively, and I0 is the incident light intensity. Sinceabsorption in tissues is substantially less than scattering (µa � µs) in the NIR spectral range,the exponential attenuation of ballistic photons in tissue is dependent mainly on the scatteringcoefficient:

I = I0 e−µsz. (5)

Since the tissue scattering coefficient (µs) changes with glucose concentration, theexponential profile of light attenuation in tissue is dependent on glucose concentration.Therefore, one can monitor glucose concentration by measuring the exponential slope oflight attenuation in tissue with the OCT technique (Esenaliev et al 2001, Larin et al 2002).

In this paper, we present results of our experimental and theoretical investigations of thespecificity of the OCT-based glucose sensor. Skin optical properties can vary due to changesin different physiological or environmental conditions (such as an osmolyte concentrationfluctuations and temperature) and by chemical or physical manipulations (such as byintroducing index-matching osmotically active chemical agents or compression (Tuchin et al1997, Vargas et al 1999, Chan et al 1996)). For example, skin squeezing decreases theconcentration of ECF and the introduction of osmotically active chemical agents (such asglucose) reduces water concentration in the ECF. All of these processes decrease the scatteringcoefficient. Changes in scattering coefficient can be produced by: (1) altering the refractiveindex mismatch between the interstitial fluid (ISF) and scattering centres and (2) structuralmodifications in tissue (e.g., cell volume change). Besides that, tissue heterogeneity, motionartefacts, changes in temperature, blood pressure, heart rate, and hematocrit, and surgical andanaesthesia procedures may influence the signals. This study is devoted to the experimentaland theoretical analysis of the influence of these parameters on the OCT signal slope.


Figure 1. Experimental set-up used in the studies with scattering media and animals (Yucatanmicropigs and New Zealand white rabbits).

2. Materials and methods

2.1. Experimental set-up

The experimental set-up used in our studies is depicted in figure 1. The experiments wereperformed with a portable OCT system with the wavelength of 1300 nm, superluminescentdiode (SLD) with a power of 500 µW, and coherence length of approximately 14 µm.The radiation emitted from the SLD was coupled into a single-mode fibre-optic Michelsoninterferometer along with an aiming (visible) beam of a CW light-emitting diode with thewavelength of 640 nm. Light in the sample arm of the interferometer was directed intothe objects to be imaged using a single-mode optical fibre and focused with lenses. Theoutput power in the sample arm was ∼200 µW. Light scattered from the sample and lightreflected from the reference arm mirror formed an interferogram, which was detected by aphotodiode. The signal from the photodiode was preamplified, electronically filtered aroundthe Doppler-shifted frequency ( fD = 700 kHz), amplified by a logarithmic amplifier, digitizedby an analogue-to-digital converter (ADC), and recorded by a laptop computer (PC) forfurther processing. The logarithmic amplifier had the dynamic range of 0–100 dB with theamplification coefficient of 23 mV dB−1, and the noise level of 5–10 dB without averagingand ∼0.2 dB with averaging of 512 signals (Kholodnykh et al 2003).

In-depth scanning was produced electronically by piezo-electric modulation of the fibrelength. Specially designed optical probes were used to scan the incident beam over the samplesurface in the lateral direction. In-depth (Z-axial) scanning by the interferometer and thelateral scanning by the probe allowed for reconstruction of two-dimensional images (300 by400 pixels). Each scan in the Z direction was averaged 5–7 times. The lateral scanninglength was up to 3 mm. Each lateral scan was accomplished every 20–30 s. The OCT system


operation was automated and controlled by the PC. The lateral resolution of the OCT systemwas 12 µm.

2.2. Theoretical calculations

The Mie theory of scattering (Van de Hulst 1981) was used to calculate the scatteringcoefficient as a function of glucose concentration in the phantoms. A subroutine for calculatingthe scattering and absorption cross sections and anisotropy parameter (BHMIE given by Bohrenand Huffman (Bohren and Huffman 1983)) was applied to calculate the scattering coefficientfor polystyrene microspheres with a diameter of 760 nm in water.

The wavelength dependences of refractive indices of water, nwater, and polystyrene, npol,used in our calculations are

nwater(λ) = 1.3199 +6.878 × 103

λ2− 1.132 × 109

λ4+

1.11 × 1014

λ6(6)

npol(λ) = 1.5626 +1.169 × 104

λ2− 1.125 × 109

λ4+

1.72 × 1014

λ6

where λ is in nanometers (Maier et al 1994). The glucose-induced change in refractive indexis �n = 2.73 × 10−5 mM−1 (Tuchin et al 1997).

2.3. Animals studies

Yucatan micropigs (age: 4–5 months) and New Zealand white rabbits (weight: 3–3.5 kg) wereused in animal studies. The hairless Yucatan micropig is recognized as the best model of humanskin (Fujii et al 1997, Kurihara-Berhstrom et al 1986). All procedures with animals wereperformed in accordance to an animal protocol approved by the Institutional Animal Use andCare Committee of the University of Texas Medical Branch. Animals were pre-anaesthetizedwith a standard telazol/xylazine/ketamine mixture given intramuscularly. Yucatan micropigswere placed in a specially designed holder in order to minimize motion of the animals duringexperiments.

Two glucose intravenous (i.v.) administration protocols were used in the animal studies:(1) bolus glucose injections resulting in sharp and rapid increase of blood glucoseconcentration and (2) glucose clamping. The clamping technique allowed slow, controlledi.v. administrations of glucose at desired rates to simulate normal physiological changesin blood glucose concentration. Therefore, it minimizes a possible tissue physiologicalresponse to a sharp increase in the glucose concentration such as a change in cell volume, ECFfractional volume,blood microvessel diameter that can be induced by bolus injections. Glucoseadministration was performed through the left femoral vein and blood samples were taken


(a)

(c) (d)

(b)

Depth (µm) Depth (µm)

Depth (µm)

OC

T S

igna

l (ar

b. u

n.)

OC

T S

igna

l (ar

b. u

n.)

Figure 2. (a) Typical histology of the Yucatan micropig skin obtained from dorsal area.(b) OCT image obtained from the Yucatan micropig (dorsal area, dimensions are in mm) and(c) corresponding OCT signal. (d) Representative OCT signals obtained from Yucatan micropigskin during glucose clamping experiment at low and high blood glucose concentration (top) anda part of the OCT signal in the dermis area with the linear fit of the OCT signals in this layer(bottom).

(stratum corneum of epidermis), prickle cell layer (epidermis) and connective tissue ofdermis. These major skin layers are also visible in the OCT image (figure 2(b)) and signal


Figure 3. Dependence of refractive index on osmolyte concentration.

(figure 2(c)). The stratum corneum (average thickness ∼15 µm), a highly scattering skinlayer, produced the first peak in the OCT signal. The prickle cell layer has less scattering(thickness of ∼110 µm) and produced the minimum in the OCT signal. The second peak andthe rest of the OCT signal are produced by the dermal skin layer.

OCT images obtained from skin of animals were processed using an original programdeveloped in Microsoft Visual C++ 6.0 to obtain OCT signals and to calculate the OCT signalslopes. The two-dimensional images (figure 2(b)) were averaged in the lateral (x-axial)direction into a single curve to obtain an OCT signal that represents one-dimensionaldistribution of light in depth (figure 2(c)). The OCT signals were plotted in a logarithmic scaleand the slope of the signals was calculated at specific depths by the linear least-squares method.As an example, figure 2(d) shows two representative signals obtained from a Yucatan micropigskin during glucose clamping experiment at low (∼100 mg dl−1) and high (∼300 mg dl−1)blood glucose concentrations. The difference between these two OCT signals was observedat a depth of 260 to 400 µm that corresponds to the dermis skin layer. The linear fit of theOCT signals at this layer revealed an OCT signal slope decrease of 25% (from −0.002 04 to−0.001 53) due to the increase of the blood glucose concentration from 100 to 300 mg dl−1

that yields 2.3% change in the OCT signal slope per 1 mM.

3. Results and discussion

3.1. Refractive index matching

Variation of glucose concentration in the extracellular space produces changes in the refractiveindex mismatch between the extracellular fluid and scatterers and, therefore, affects the tissuescattering properties. However, the variation of the concentration of other tissue componentsmay affect the tissue scattering coefficient if these components alter the refractive indexmismatch.

Figure 3 shows the dependence of the refractive index on the concentration of differentosmolytes of the body. The graph shows data for glucose, potassium chloride (KCl), sodium


Table 1. Changes in the refractive index and scattering coefficient (calculated using the Mie theoryof scattering) for major osmolytes of ECF.

�n × 10−5 %C (relative to �µs (relativeSubstance (mM−1) C (mM) 302 mOsm l−1) �µs (%) to glucose)

Glucose 2.55 3–30 8.9 1.2 1NaCl 0.98 135–145 3.3 0.16 0.13KCl 0.96 3.5–5 0.5 0.027 0.023Urea 0.85 2.9–8.9 1.98 0.087 0.073

chloride (NaCl) and urea that may produce the highest changes in refractive index comparedwith other tissue osmolytes (Handbook of Chemistry and Physics 1974). One can see fromfigure 3 that all of the listed osmolytes increase the refractive index of a background solutionwhen concentration increases, however glucose has the most pronounced effect. Accordingto these data, the absolute change in the refractive index is equal to 2.55 × 10−5 mM−1 forglucose, 0.96 × 10−5 mM−1 for KCl, 0.98 × 10−5 mM−1 for NaCl and 0.85 × 10−5 mM−1

for urea.A summary of the changes in the refractive index and scattering coefficient (calculated

using the Mie theory of scattering) for the substances shown in figure 3 is presentedin table 1. Column 2 shows absolute (�n) changes in the refractive index. Normalreference laboratory values of physiological changes in concentrations of these substances(Kratz and Lewandrowski 1998) and their maximum influence on the total osmolality changein the extracellular fluid (relative to total osmolarity of ISF 302 mOsm l−1 (Guyton 1996)) arelisted in the third and fourth columns, respectively. The Mie theory of scattering was used tocalculate corresponding changes in the tissue scattering coefficient (the fifth column) producedby a variation in the concentration of the substances listed in the third column. The lastcolumn shows changes in the scattering coefficient of the KCl, NaCl and urea relative to thatof glucose. One can see from this column, the maximal possible influence of these substanceson the scattering coefficient (assuming that their concentration changes in the physiologicalrange) is much less (from 0.023 to 0.13) than that produced by glucose.

Therefore, the data presented in figure 3 and table 1 show that although a concentrationchange of KCl, NaCl and urea in the physiological ranges can potentially change the tissuescattering coefficient by altering the refractive index mismatch, the effect of glucose is at least8 to 40 times greater.

3.2. Glucose as an osmotic agent

The volume change upon the exposure of cells to osmotic pressure induced by differentosmolytes has been extensively studied in a great variety of cells. The physiological effectsand mechanism of the cell volume change under osmotic conditions have been reviewed in(Haussinger and Lang 1991). The response of cells exposed to osmotic pressure consists oftwo major parts: an initial change in their volume and return of their volume to the nearly initialvalues using regulatory mechanisms. A hypo-osmotic stress (e.g.,made by lowering osmolytesconcentration in extracellular fluid) results in water transport from extracellular to intracellularspace and increases cell volume (cell swelling). On the other hand, a hyper-osmotic stress(e.g., made by increasing osmolytes concentration in extracellular fluid) results in watertransport from intracellular to extracellular space and decreases cell volume (cell shrinkage).These changes in cell volume can alter light scattering in tissues (Chance et al 1995).


Figure 4. Scattering coefficient as a function of the scatterer diameter calculated using the Mietheory of scattering.

Therefore, it is important to estimate the possible influence of glucose-induced changes incell volume on scattering properties of tissues.

In order to illustrate the effect of glucose-induced changes on cell volume, one canconsider the response of a cell to an osmotic stress without volume regulation mechanism.For example, data obtained for isolated rabbit ventricular myocytes subjected to osmoticstress (Suleymanian and Baumgarten 1996) show that doubling of an osmolyte concentrationresults in approximately a 35% decrease in cell volume. Total osmolarity of the ISF, πT, is302 mOsm l−1 (Guyton 1996). The blood glucose concentration can be in the range from3 mM to 30 mM and, thus, may alter πT by a maximum of 9%. Therefore, glucose-inducedchanges in the relative volume of myocytes and diameter (assuming that the cells are spherical)may be a maximum of 3.3% and 1.5%, respectively.

Figure 4 shows the dependence of the scattering coefficient (calculated by using Mietheory) of a medium on diameter of scattering centres. As one can see from the figure,the theory predicts 6% change in the scattering coefficient upon 1.5% change in thediameter. Therefore, maximal possible glucose-induced decrease of the scattering coefficientin rabbit ventricular myocytes due to variation of cell volume is approximately 6% if glucoseconcentration increases from 3 to 30 mM (or 0.22% mM−1). These changes are much lowerthan the changes in the OCT signal slope obtained from skin (Esenaliev et al 2001, Larin et al2002).

Addition of osmotically active agents to tissues may produce structural modificationsif concentrations of osmolytes are high. For instance, Vargas demonstrated that addition ofhigh concentration of glucose (4.2 M) to the rat skin results in 20% water loss (Vargas 2001)that may significantly decrease overall tissue scattering. However, at physiologically relevantconcentrations of glucose (3–30 mM), the water loss would be only 0.005% mM−1 (assuminga linear dependence between glucose concentration and water loss). Therefore, most likelythe glucose-induced structural changes do not contribute to the change in the OCT signal slopeat physiologically relevant concentrations.


(a)

(b)

Time (min)

BG

C (m

g/dL)

OC

T S

igna

l Slo

pe (

arb.

un.

)

BGC (mg/dL)

BGC (mM)

OC

T S

igna

l Slo

pe (

arb.

un.

)

Figure 5. Slope of OCT signals (OCT SS) obtained from rabbit ear during bolus glucose injectionexperiment versus: (a) time and (b) blood glucose concentration (BGC). R: correlation coefficient,P: p-value.

3.3. Bolus glucose injection versus glucose clamping

Figure 5 shows results obtained from the skin of the rabbit ear during bolus glucose injectionexperiments. OCT images were obtained from the left ear, blood samples were taken from theleft femoral vein, and the glucose solution was administrated through a right ear vein. Circlesand squares in figure 5(a) and all of the following similar graphs present the actual valuesof the OCT signal slope and blood glucose concentration, respectively. Corresponding solidcurves (in figure 5(a) and all of the following similar graphs) were obtained by smoothingof the actual values using adjacent averaging algorithm: the smoothed value at index i is theaverage of the data points in the interval [i − (n − 1)/2, i + (n − 1)/2], inclusive, wheren is the number that controls the degree of smoothing. The number n was chosen to smooth


(a)

(b)

Time (min)

BG

C (m

g/dL)

OC

T S

igna

l Slo

pe (

arb.

un.

)

BGC (mg/dL)

BGC (mM)

OC

T S

igna

l Slo

pe (

arb.

un.

)

Figure 6. Slope of OCT signals obtained from rabbit ear during glucose clamping experimentversus: (a) time and (b) blood glucose concentration.

over 3–4 min of the experiments and typically was equal to 10. Single bolus glucose injectionwas performed 4 min after the onset of the experiment and resulted in the increase of bloodglucose concentration from a baseline of approximately 380 mg dl−1 to its maximum of535 mg dl−1 in 5 min followed by slow physiological decrease. The increase and decreaseof the blood glucose concentration are followed by the decrease and increase of the measuredOCT signal slopes, respectively. Approximately, 200 µm thick layer at the depth of 700 µmwas used to calculate the OCT signal slope. The slope change was approximately 8% mM−1

in this experiment.The slopes of the OCT signal obtained from the rabbit skin (ear) and corresponding blood

glucose concentrations measured at different time during the glucose clamping experimentare shown in figure 6(a). Figure 6(b) demonstrates their linear dependence on blood glucose


Figure 7. Difference between two OCT signals recorded at two different times from the same placeof a homogeneous object: aqueous suspension of polystyrene microspheres (thick line). Same foran inhomogeneous object: rabbit skin (thin line).

concentration. Glucose solution (50% w/w) was injected for approximately 45 min (from20th to 65th min of the experiment) using a digital pump at the rate of 0.18 ml min−1. Thisresulted in the increase of blood glucose concentration from baseline level of 110 mg dl−1

(∼6 mM) to 400 mg dl−1 (∼21.6 mM). The slope of the OCT signals at the depth from 300 µmto 400 µm changed approximately 54% in the range from 110 to 400 mg dl−1 (3.3% mM−1).Good correlation between changes in actual blood glucose concentration and the slope wasobserved with the correlation coefficient of 0.88 and p-value less than 0.01%. The lagtime between increase of blood glucose concentration and decrease of the OCT signal slopewas approximately 8 min. The lag time of 0–15 min observed in our experiments is inagreement with the lag time between glucose concentrations in the ISF and blood measuredby other researchers (Pickup et al 1989, Matthews et al 1988, Jensen et al 1995, Stout et al2001).

Comparison of results obtained from experiments involving these two differentprotocols of glucose administration showed that the changes in the OCT signal slope were3–4% mM−1 higher during bolus glucose injection than that during the glucose clamping. Theonly difference between these two methods is the rate of i.v. glucose administration. Therefore,the tissue physiological response to the sharp increase of ECF osmolyte concentration, besidesrefractive index matching, could contribute to the OCT signal slope changes. However, as itwas mentioned earlier, tissue response during slow changes in blood glucose concentration inthe physiological range should be insignificant.

3.4. Tissue heterogeneity and motion artefacts

Skin and other tissues are highly inhomogeneous and have non-uniform density and refractiveindex distribution in depth and lateral directions. Heterogeneity of skin can change the OCTsignal due to tissue motion. Figure 7 shows the difference between two OCT signals recorded


(a)

(b)

Time (min)

Time (min)

BG

C (m

g/dL)

BG

C (m

g/dL)

OC

T S

igna

l Slo

pe (

arb.

un.

)O

CT

Sig

nal S

lope

(ar

b. u

n.)

Figure 8. Slope of OCT signals obtained from a micropig lip during a glucose clamping experimentand corresponding blood glucose concentrations versus time: (a) raw data and (b) data correctedfor the motion artefact. The arrow shows the time of the motion impact.

at two different times from the same place of a homogeneous object (aqueous suspension ofpolystyrene microspheres) and of an inhomogeneous object (rabbit skin). The difference isseveral times higher for the skin than that for the homogeneous object that produces fluctuationsof data points shown in the previous figures.

High optical inhomogeneity of skin and motion artefacts may change the OCT signal slopeand, thus, may decrease accuracy and specificity of the glucose-monitoring system. Figure 8shows results obtained from a glucose clamping experiment performed with a micropig.OCT signals were measured from the lip. The animal suddenly moved at 100th min of theexperiment and the system started to probe another area of the lip. This was clearly reflectedin the value of the OCT signal slope (figure 8(a)). Correction of the slope for this displacement(by scaling of the OCT signal slope axis) yielded better correlation with the blood glucoseconcentration values (figure 8(b)). Therefore, it is important to minimize or control the opticalprobe displacement relative to the skin surface in order to provide accurate measurements ofthe glucose concentration with the OCT technique.


(a)

(b)

Time (min)

Time (min)

Skin Tem

perature (C°)

BG

C (m

g/dL)

OC

T S

igna

l Slo

pe (

arb.

un.

)O

CT

Sig

nal S

lope

(ar

b. u

n.)

Figure 9. Slope of OCT signals obtained from rabbit skin and corresponding skin temperatureand blood glucose concentration versus time during: (a) skin heating and (b) glucose clampingexperiments.

3.5. Temperature

Several experiments were performed to elicit the effect of skin temperature on the OCT signalslope. Experiments performed in skin tissue in vivo using the OCT technique showed adependence of the OCT signal slope on the skin temperature.

Figure 9(a) shows a typical result obtained from rabbit skin during skin heating with hotair produced by a heat gun. Skin temperature was monitored using a thermocouple placedon the skin surface near the OCT probe. The results show approximately 50% decrease ofthe OCT signal slope with the increase of skin temperature from 35 ◦C to 42 ◦C and increaseback to the original values due to passive tissue cooling. Therefore, tissue heating may changeoptical properties. Most likely, the mechanism of the temperature-induced changes in tissueoptical properties is the increase of perfusion in tissue and change in cell volume. However,


Figure 10. OCT signal slope measured from micropig skin during glucose clamping experimentand corresponding blood glucose concentration (squares), mean blood pressure (rhombi) and heartrate (circles).

normal temperature fluctuations of the skin (±1 ◦C) did not change the OCT signal slope in acontrol experiment without heating (figure 9(b)).

3.6. Blood pressure and heart rate

Heart rate and average blood pressure were recorded in most animal experiments. Figure 10shows the OCT signal slope (measured in micropig skin), corresponding blood glucoseconcentration, mean blood pressure and heart rate recorded during a glucose clampingexperiment. No significant effect of the blood pressure and heart rate on the OCTsignal slope was observed in these experiments. Therefore, fluctuations of blood pressureand heart rate in our experiments do not decrease accuracy and specificity of the OCT-based glucose sensor. However, further investigation required in order to understand theinfluence of the heart rate and blood pressure on the tissue optical properties under differentconditions.

3.7. Hematocrit

Influence of hematocrit on the OCT signal slope was studied in several experiments in vivo.The hematocrit was varied by injecting saline (without glucose) in blood. Figure 11(a) showsthe OCT signal slope and corresponding hematocrit measured during an experiment withsaline injection. The OCT signals were recorded in vivo from rabbit blood circulating in a thinglass tube. The ends of the glass tube were inserted into two latex tubes that were attachedto an ear artery and vein to make a closed loop for blood circulation. Results presented infigure 11(a) demonstrate that the variation in hematocrit produce changes in the OCT signalslope obtained from the blood. An 18% decrease in hematocrit lowers the OCT signal slopeby 16%. This is in agreement with our previous results: the OCT signal slope is linearlyproportional to the concentration of scattering particles in a medium.


(a)

(b)

Time (min)

Time (min)

Hem

atocrit (%)

Hem

atocrit (%)

OC

T S

igna

l Slo

pe (

arb.

un.

)O

CT

Sig

nal S

lope

(ar

b. u

n.)

Figure 11. Slope of OCT signals recorded from rabbit: (a) blood and (b) skin at differenthematocrit.

However, the OCT signal slope obtained from skin was not dependent on the concentrationof red blood cells. As demonstrated in figure 11(b), variation of hematocrit from 41% to 37%did not significantly change OCT signal slope obtained from rabbit skin. Therefore, theseresults suggest that there is no influence of the hematrocrit on OCT signal slope measuredfrom skin.

3.8. Surgery and anaesthesia

Figure 12 shows data obtained in a control experiment when no glucose was administered.The same surgery and anaesthesia procedures were used during this experiment. The OCTsignal slope was calculated at different depths. No significant changes in blood glucose orthe OCT signal slopes at any depth were observed. As an example, the data obtained at thedepth from 300 µm to 400 µm (dermis area) are shown in figure 12. Therefore, these results


Figure 12. Slope of OCT signal obtained from the rabbit skin and blood glucose concentrationmeasured during control experiment without glucose administration.

indicate that the surgical procedure and anaesthesia did not decrease accuracy and specificityof glucose monitoring in our experiments.

4. Conclusions

Our theoretical and experimental studies of specificity of the glucose-induced changes in theOCT signal slope demonstrated that: (1) several body osmolytes may change the refractiveindex mismatch between the ISF and scattering centres, however the effect of glucose isapproximately one to two orders of magnitude greater; (2) an increase of the ISF glucoseconcentration in the physiological range (3–30 mM) may decrease the scattering coefficientby 0.22% mM−1 due to cell volume change; (3) bolus glucose injections produce greaterchanges in the OCT signal slope compared with that measured during glucose clampingexperiments (most likely due to the tissue physiological response to a rapid increase of theosmolyte concentration such as changes in cell volume and concentrations of scattering centres,and, possibly, changes in tissue structure); (4) stability of the OCT signal slope is dependent ontissue heterogeneity and motion artefacts; (5) moderate skin temperature fluctuations (±1 ◦C)do not decrease accuracy and specificity of the OCT-based glucose sensor, however substantialskin heating or cooling (several ◦C) significantly change the OCT signal slope; and (6)fluctuations of blood pressure, heart rate, hematocrit as well as surgical procedures andanaesthesia did not alter the skin optical properties in our experiments. These results suggestthat the observed changes in the OCT signal slope during our animal and clinical experimentsin healthy subjects (Esenaliev et al 2001, Larin et al 2002) are most likely due to variation ofthe blood glucose concentration. However, further investigation of specificity of the OCT-based glucose sensor is required, in particular, in diabetic patients. For example, it is knownthat patients with diabetes sometimes develop hyperkalemia (an elevated serum potassiumlevel) (Stevens and Dunlay 2000) that may influence specificity of the OCT-based glucosesensor.

The specificity and accuracy of the OCT-based glucose sensor can be improvedby using different approaches including, for instance, application of multi-wavelength


measurements when the OCT signal is recorded simultaneously at two or more differentwavelengths.

Acknowledgments

The authors would like to thank Dr Donald J Deyo and Mr Brent Bell for technicalassistance during experiments. This work was supported by the Texas Higher EducationCoordinating Board Advanced Technology Program (grant no 004952-0133-1999) and theNational Institutes of Health (National Institute of Diabetes and Digestive and Kidney Diseases,grant no R21 DK58380).

References

American Diabetes Association 1998 Economic consequences of diabetes mellitus in the U.S. in 1997 Diabetes Care21 296–309

Bohren C F and Huffman D R 1983 Absorption and Scattering of Light by Small Particles (New York: Wiley)Bruulsema J T et al 1997 Correlation between blood glucose concentration in diabetics and noninvasively measured

tissue optical scattering coefficient Opt. Lett. 22 190–3Chan E et al 1996 Effects of compression on soft tissue optical properties IEEE J. Sel. Topics Quantum Electron. 2

943–50Chance B et al 1995 Effects of solutes on optical properties of biological materials: models, cells, and tissues Anal.

Biochem. 227 351–62Cote G L 1997 Noninvasive optical glucose sensing—an overview J. Clin. Eng. 22 253–9Cote G L, Fox M D and Northrop R B 1992 Noninvasive optical polarimetric glucose sensing using a true phase

measurement technique IEEE Trans. Biomed. Eng. 39 752–6Diabetes Control and Complications Trial Research Group 1997 Hypoglycemia in the diabetes control and

complications trial Diabetes 46 271–86Drezek R, Dunn A and Richards-Kortum R 1999 Light scattering from cell: finite-difference time-domain simulations

and goniometric measurements Appl. Opt. 38 3651–61Duck F A 1990 Physical Properties of Tissue (London: Academic)Esenaliev R O et al 2001 Noninvasive monitoring of glucose concentration with optical coherent tomography Opt.

Lett. 26 992–4Fujii M et al 1997 Evaluation of Yucatan micropig skin for use as an in vitro model for skin permeation study Biol.

Pharm. Bull. 20 249–54Gabriely I et al 1999 Transcutaneous glucose measurement using near-infrared spectroscopy during hypoglycemia

Diabetes Care 22 2026–32Goetz Jr M J et al 1995 Application of a multivariate technique to Raman spectra for quantification of body chemicals

IEEE Trans. Biomed. Eng. 42 728–31Goldstein D E et al 1995 Tests of glycemia in diabetes mellitus Diabetes Care 18 896–909Graaff R et al 1992 Reduced light-scattering properties for mixtures of spherical particles: a simple approximation

derived from Mie calculations Appl. Opt. 31 1370–6Guyton A G 1996 Textbook of Medical Physiology 9th edn (Philadelphia, PA: Saunders)Handbook of Chemistry and Physics 1974 54th edn (Boca Raton, FL: CRC Press)Haussinger D and Lang F 1991 Cell volume in the regulation of hepatic function: a mechanism for metabolism

control Biochim. Biophys. Acta 1071 331–50Huang D et al 1991 Optical coherence tomography Science 254 1178–81Jensen B M et al 1995 Glucose content in human skin: relationship with blood glucose levels Scand. J. Clin. Lab.

Invest. 55 427–32Kholodnykh A I et al 2003 Precision of measurement of tissue optical properties with OCT Appl. Opt. at pressKohl M, Cope M, Essenpreis M and Bocker D 1994 Influence of glucose concentration on light scattering in tissue-

simulating phantoms Opt. Lett. 19 2170–2Kohl M, Essenpreis M and Cope M 1995 The influence of glucose concentration upon the transport of light in

tissue-simulating phantoms Phys. Med. Biol. 40 1267–87Kratz A and Lewandrowski K B 1998 Case records of the Massachusetts General Hospital. Weekly clinicopathological

exercises. Normal reference laboratory values N. Engl. J. Med. 339 1063–72


Kurihara-Berhstrom T et al 1986 Characterization of the Yucatan miniature pig skin and small intestine forpharmaceutical applications Lab. Anim. Sci. 36 396–9

Larin K V, Eledrisi M S, Motamedi M and Esenaliev R O 2002 Noninvasive blood glucose monitoring with opticalcoherence tomography: a pilot study in human subjects Diabetes Care 25 2263–7

MacKenzie H A et al 1999 Advances in photoacoustic noninvasive glucose testing Clin. Chem. 45 1587–95Maier J S et al 1994 Possible correlation between blood glucose concentration and the reduced scattering coefficient

of tissue in the near infrared Opt. Lett. 19 2062–4Matthews D R et al 1988 An amperometric needle-type glucose sensor tested in rats and man Diabet. Med. 5 248–52National Institute of Diabetes and Digestive and Kidney Diseases 1994 Diabetes Overview (Bethesda, MD: NIH)

NIH publ no 94-3235National Steering Committee for Quality Assurance in Capillary Blood Glucose Monitoring 1993 Proposed strategies

for reducing user error in capillary blood glucose monitoring Diabetes Care 16 493–8Nettles A 1993 User error in blood glucose monitoring. The National Steering Committee for Quality Assurance

Report Diabetes Care 16 946–8Pan S et al 1996 Near-infrared spectroscopic measurement of physiological glucose levels in variable matrices of

protein and triglycerides Anal. Chem. 68 1124–35Pickup J C, Shaw G W and Claremont D J 1989 Invivo molecular sensing in diabetes mellitus—an implantable

glucose sensor with direct electron-transfer Diabetologia 32 213–7Rabinovitch B, March W F and Adams R L 1982 Noninvasive glucose monitoring of the aqueous humor of the eye:

Part I. Measurement of very small optical rotations Diabetes Care 5 254–8Robinson M R et al 1992 Noninvasive glucose monitoring in diabetic patients: a preliminary evaluation Clin. Chem.

38 1618–22Stevens M S and Dunlay R W 2000 Hyperkalemia in hospitalized patients Int. Urol. Nephrol. 32 177–80Stout P J et al 2001 Comparison of glucose levels in dermal interstitial fluid and finger capillary blood Diabetes

Technol. Ther. 3 81–90Suleymanian M A and Baumgarten C M 1996 Osmotic gradient-induced water permeation across the sarcolemma of

rabbit ventricular myocytes J. Gen. Physiol. 107 503–14Tuchin V V et al 1997 Light propagation in tissues with controlled optical properties J. Biomed. Opt. 2 401–17Van de Hulst H C 1981 Light Scattering by Small Particles (New York: Dover)Vargas G 2001 Reduction of light scattering in biological tissue: implications for optical diagnostics and therapeutics

PhD Dissertation University of Texas, AustinVargas G et al 1999 Use of an agent to reduce scattering in skin Lasers Surg. Med. 24 133– 41

speciﬁcity of noninvasive blood glucose sensing using ...bol.egr.uh.edu › sites › bol ›...

Documents