| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Reviews |
a Abbott Laboratories, Diagnostics Division, D9NL, AP 20, 100 Abbott Park Road, Abbott Park, IL 60064. Fax 847-938-7072; e-mail Omar.khalil{at}add.ssw.abbott.com.
 |
Abstract |
|---|
 Top Abstract IntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
Self-monitoring of blood glucose concentrations has advanced over the past few years. Glucose values determined by home meters correlate well with laboratory results. Because of the importance of precision and accuracy of self-monitoring blood glucose devices, guidelines for the performance of these devices were recommended in 1987 by the American Diabetes Association (8). These goals and performance characteristics of current meters are important as a yardstick for establishing the potential performance of NI technologies. The limits of detection and quantification, the standard deviation of the measurement, the accuracy, and the total error of an NI measurement need to correlate with self-monitoring devices and with measurements in the laboratory. American Diabetes Association guidelines for home glucose monitors state that values should fall within 15% of the readings obtained by a reference method. The goal set for future self-monitoring blood glucose meters is to achieve a variability (system plus user) of less than ± 10% (8). Evaluation of commercial devices shows that not all meters fulfill this consensus goal (9). The major error component is user error, which includes the volume of the blood droplet, the accuracy with which the blood drop is placed on the pad, hematocrit effects on serum volume, timing, and the effect of temperature on the signal-generating reaction. An NI measurement can potentially eliminate some of these effects but may have a different set of error sources.
An NI body glucose monitoring device is defined in this review as a device that comes in contact with, or remotely senses, a human body part, without protrusion through membranes or sampling a body fluid for analysis external to the body part. Thus implantable devices (10) and methods dependent on glucose diffusion through the skin (11) will not be discussed. For the purpose of this review, only optical methods in the visible and infrared spectral regions will be discussed.
An NI glucose monitor processes optical signals transmitted through or reflected by the stratum corneum, dermis and epidermis layers, subcutaneous tissue, interstitial fluid, and blood vessels (both arterial and venous blood), which represent independent compartments. Each of these compartments may have different optical properties, concentrations of interferents, and concentrations of glucose. An NI-determined glucose value may represent an average value of different glucose concentrations, and correlation between an NI glucose measurement and a blood glucose measurement may vary by body site, depending on the differences in tissue and vascular properties of each part. Once an NI detection technology is proven to produce clinically significant data, new goals for NI systems need to be established based on how they will affect clinical decisions and the added value to patient care. A trade-off between accuracy and ease of use, or between precision and potential for frequent testing may need to be made.
NI glucose monitors can be configured as bedside monitors for point-of-care testing, portable personal monitors, and closed-loop insulin pump/glucose monitors. Portable personal glucose monitors will have the most impact on patient care. Their use will increase the frequency of testing and lead to tighter diabetes control. They will require a rugged detection system that is not affected by environmental factors, robust algorithms, rigorous but easy-to-perform calibration routines, and miniaturized electronics and optics.
|
Research Methodologies for NI Glucose Measurements |
|---|
|
TopAbstractIntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
A second approach is the physical model approach. This methods involves the following steps: (a) measurement of a glucose optical signal in a simple matrix; (b) progressive increases in the complexity of the matrix to mimic human tissues; (c) demonstration of accuracy and precision at each step; and (d) correlation of the data with a mathematical model for light propagation in tissue. Finally, the detection system and the measurement method are applied to body parts. The in vivo signals are again correlated with the invasive data by use of chemometric techniques. This stepwise approach allows for identification of noise components so that strategies may be derived to minimize their contribution to the signal before the use of chemometric techniques.
|
Calibration of NI Glucose Measurements |
|---|
|
TopAbstractIntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
Several tissue-simulating phantoms have been proposed as calibration systems. Some of these phantoms are suspensions of lipids or polystyrene particles in solutions having different concentrations of glucose. A phantom containing fat and glucose solution has been used to mimic tissue glucose absorption in the 2000–2500 nm range (14). Glucose concentrations used in these studies have generally been higher than the physiological range. Because of the nonspecificity of the signal measured by several technologies (as will be discussed later), developing tissue-simulating phantoms as analytical standards for NI glucose determination in tissue is a challenging goal. An NI device that is analytically calibrated can be considered as universally calibrated; i.e., can be used with any patient.
clinical calibration
In vitro and in vivo measurements are performed on a fasting
subject at time intervals during an OGTT, meal-tolerance test, or a
glucose clamp procedure. These methods offer a concentration range over
which the glucose signal can be monitored. Data from an OGTT can be
used to establish an NI instrument response as glucose concentration
for an individual is changing. Data that are generated during the test
period are used to predict glucose concentrations in subsequent NI
measurements. Because the response of an NI instrument may embody
non-glucose-related physiological effects, relying on clinical
calibration based on the correlation of OGTT data with NI instrument
response leads to a calibration curve that is unique to the individual
tested. This calibration curve may need to be periodically updated, by
use of an invasive test.
Use of an OGTT or a meal-tolerance test for calibration leads to a series of measurements that are sequential in time. Spurious drift and time-dependent artifacts can influence the results from multivariate calibrations when randomized sampling cannot be performed (15). Thus, the temporal distribution of signal and noise may lead to erroneous glucose correlation. Furthermore, Arnold et al. (16) have challenged the validity of several in vivo blood glucose measurements based on spectra collected in a time-dependent manner. Chance temporal correlation between assigned glucose concentrations and some uncontrolled experimental parameter have been suggested as responsible for the apparent functionality of some of these models.
data presentation
In addition to the use of statistical methods to present the
performance of NI glucose measurements, the Clarke error grid has also
been used for data presentation. The Clark error grid analysis offers a
quick estimation of the accuracy of the measurement compared with a
reference method. Data from the test device are plotted against the
results of a reference method. A scatter diagram is established and
divided into A, B, C, D, and E zones (17)(18).
Data points falling in zone A represent acceptable performance, whereas
those in zones C, D, and E represent unacceptable performance. A
personal glucose monitor is considered to have acceptable performance
in the hands of a user if >95% of the data points on the scatter
graph fall in A zone and 0% fall in the C, D, and E zones
(18). This will be approximately equivalent to a total error
of 20% at blood glucose concentrations >3.5 mmol/L. A constant error
(i.e., not a percentage) is used below this concentration. Data points
in the C, D, and E zones represent progressively increasing deviation
between the home glucose monitor and the reference method. This
increased inaccuracy may lead to the wrong type of intervention.
|
Principles of Glucose Tissue Measurements |
|---|
|
TopAbstractIntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
The fundamental IR absorption bands of glucose have been reported in
solid pellets and in solution (26)(27)(28)(29)(30). The strongest bands
that can generate intense combinations and overtones are the broad OH
stretch at 3550 cm-1 and the C
H stretch
vibrations at 2961 and 2947 cm-1. Possible
combination bands are a second OH overtone band at 939 nm (3
OH) and
a second harmonic CH overtone band at 1126 nm (3CH). A first OH
overtone band can be assigned at 1408 nm (2OH). The 1536 nm band can
be assigned as an OH and CH combination band (OH + CH). The 1688
nm band is assigned as a CH overtone band (2CH). Other bands at
>2000 nm are possibly combinations of a CH stretch and a CCH, OCH
deformation at 2261 nm (CH + CCH, OCH) and 2326 nm (CH +
CCH, OCH). The presence of the CCH, OCH ring deformation component
confers some glucose specificity on these bands. The calculated
near-IR overtone and combination spectra of glucose overlap with
several (more intense) combinations and overtone bands of water and fat
and hemoglobin electronic absorption bands, as shown in Table 1
(31). These are the major potential interferences
with the NI determination of glucose. Near-IR (2000–2500 nm)
spectrophotometric determination of glucose has been achieved in
aqueous media (32)(33)(34)(35)(36)(37)(38).
|
|
Optical Properties of Tissues |
|---|
|
TopAbstractIntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
. The near-IR spectral region is commonly proposed or used in
all reported technologies. The 600-1100 nm region of the spectrum
represents a window between the hemoglobin and visible absorption bands
and water IR absorption (39)(40)(41)(42). Light can penetrate deep
enough into the tissue to allow a spectral measurement or a therapeutic
procedure. This spectral region is used for oxygen saturation, pulse
oximetry, and laser-Doppler flow measurements (41)(42)(43)(44)(45)(46).
|
near-ir tissue optical properties
The transport equation, and the diffusion theory approximation to
this equation (47)(48)(49)(50)(51), describes the path of photons
through human tissue. It expresses light propagation in tissues by a
set of spectroscopic properties; the absorption coefficient
µa, the scattering coefficient
µs, the refractive index of the cells and the
interstitial fluid; and the anisotropy factor g (the average cosine of
the angle at which a photon is scattered). Another set of properties
are the transport properties, such as the reduced scattering
coefficient µs', where
µs' = µs[1 -
g]. The absorption coefficient µa equals the
absorbance per unit path length, 2.303 
C
cm-1, where is the molar absorptivity and C
is the molar concentration. The scattering coefficient
µs = 
where is the scattering
cross-section and is the number density of the particle. It has the
same unit as µa (cm-1)
and is equivalent to the product of an absorptivity caused by
scattering and the concentration of the scattering centers.
Attenuation of light in tissue is described, according to light
transport theory, by the effective attenuation coefficient
µeff, i.e.:
 |
Where:
 |
An exact solution of the light transport equation in turbid media can be modeled by following the path of each individual photon and calculating the probability of scattering or absorption in a series of steps, using the Monte Carlo simulation (49)(50). This modeling is used to study the path of photons in tissues and is widely used for optimization of photodynamic therapy, improvement in pulse oximetry, laser-Doppler flowmetry, and optical mammography; all have very important clinical utility. Several recent volumes of the proceedings of the Society of Photo-optical Instrumentation Engineers have covered these topics. Methods that are used for measuring the optical properties of tissues (µs, µa, and g) include transmission, diffuse and localized reflectance, and frequency domain measurements (52)(53)(54)(55)(56)(57)(58)(59)(60)(61)(62)(63)(64)(65).
effect of glucose on absorption properties of tissues
Glucose can affect the measured transmitted or reflected signal by
absorption of light at the overtone and combination band wavelengths.
Light absorption can be expressed as I =I0e-µa, where
l is the effective path length in the medium, and
µa (2.303C) is the absorption coefficient.
Changes in glucose concentration can influence the measured
µa of tissue through changes in absorption
corresponding to water displacement (absorption decreases as glucose
concentration increases) or changes in its intrinsic absorption
(absorption increases as glucose concentration increases). Changes in
µa because of water displacement are
nonspecific, and analytes with higher molecular weights will displace
more water than does glucose. Changes in the temperature and hydration
status of the body may affect water absorption bands and act as noise
sources for an NI glucose measurement. The glucose
µa in the near-IR is low and is much smaller
than that of water. It is higher at longer wavelengths. However, its
magnitude is too small to allow for direct absorption measurements at
wavelengths <1400 nm. Attenuation of light (<1400 nm) in a small body
part such as an average-sized finger varies in the range 3–4
absorbance units, and the expected change in absorbance because of a 5
mmol/L change in glucose concentration is
~10-5 absorbance units.
effect of glucose on tissue scattering
Changes in the glucose concentration affect the intensity of light
scattered by tissue. The reduced scattering coefficient of a tissue can
be expressed in a function form as: µs' =
f (, a,
ncells/nmedium), where
is the number density of scattering cells in the observation volume,
a is the diameter of the cells, ncells
is their refractive index, and nmedium is the
refractive index of interstitial fluid (51).
Changes in the nmedium are not specific for a
particular analyte and are affected by any change in the total
concentration of solutes in blood and interstitial fluid. The
calculated 
nwater as a function of the change
in the concentration of several metabolites, as calculated from the
slope of tabulated n values at different solute concentrations, is
given in Table 3
(66). During hyperglycemic episodes, the glucose
concentration changes rapidly, whereas other analyte concentrations
presumably change at a slower rate. It may be possible to relate
µs' to changes in glucose concentration over
a short time span. The measured nwater decreases
as temperature increases (66). This affects
ncells/nmedium in tissue
and presents a source of error in scattering measurements. Values of
µs' are reported to decrease with increasing
concentrations of glucose and other sugars in tissue-simulating
phantoms because of their effect on nmedium
(53)(54)(55)(56)(57)(58)(59)(60)(61)(62)(63)(64)(65).
|
A recent Monte Carlo modeling of the effect of physiologic
concentrations of glucose (5–30 mmol/L) on the diffuse reflectance or
transmittance of tissue-simulating phantoms predicts very small changes
in signal. The modeling is performed at 800 and 960 nm with water as
the only absorber. The estimated µs' is <1 x
10-3/mmol of glucose, which is much higher than the
changes µa. Other physiological factors, such
as changes in the water content, temperature, and protein
concentrations are considered. The value of
µs' caused by an increase of 1 mmol/L
in glucose concentration is equal to a 2 x
10-3 increase of water content, a 1 x
10-3 increase of protein concentration, or a
0.1 °C decrease in temperature (67). In addition, the
concentrations of pigments in the epidermis strongly influence light
penetration in the near-IR (68). The temperatures and body
water content of an individual are tightly regulated, but differ
between individuals. Although body temperature is regulated within a
fraction of a degree, temperature at the extremities may vary by
>1 °C. A temporary change in water content of 240 mL for a 75-kg
individual can lead to an error equivalent to 2 mmol/L glucose. The
reference range for albumin spans 20% of the mean concentration. This
spread will affect the signals from different subjects with normal
albumin concentrations. Hemoglobin has not been considered as an
absorber in this simulation (67). However, changes in its
concentration will affect optical signals in the 600–900 nm range.
Positioning errors and body interface effects seem to be the largest
contributors to measurement errors.
|
Optical Methods for Noninvasive Determination of Glucose |
|---|
|
TopAbstractIntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
OH and 3CH glucose overtone bands. Hiese and
co-workers (72)(73)(74)(75)(76) and Marbach et al.
(77) have reported a series of studies on the
determination of glucose in oral mucosa in the 1111–1835 nm spectral
range. This range encompasses bands corresponding to the 3OH,
2OH, OH + CH, and 2CH glucose vibrational combinations.
Jagemann et al. (78), Muller et al. (79), and
Fischbacher et al. (80) have measured reflected light
in the wavelength range 900-1200 nm through a fiber bundle touching the
skin. Khalil and Malin (81) have proposed the use of
reflected signals from four distinct near-IR spectral ranges:
1320–1340, 1440–1460, 1940–1960, and 1670–1690 and 2120–2280
nm. The 1320–1340 nm range is used for correcting for optical
coupling and sample positioning errors. The 1440–1460 nm or 1940–1960
nm ranges are where the highly absorbing water bands lie. The
1670–1690 nm range (2CH glucose overtone band) and the 2120–2280
nm range are where glucose combination bands CH + OH, CH are
located. The correction method involves normalizing the signal to the
shortest wavelength range, subtracting the normalized signal, and
correlating the subtracted signals with glucose concentration.
Brumeister et al. (82) have used the 1400–2000 nm
range for NI determination of glucose with the tongue as the body site.
Rosenthal et al. (83)(84)(85)(86)(87)(88) at Futrex Inc. have proposed using
the spectral band in 600-1100 nm near-IR transmitted or reflected
through a body part and analyzing the light by a set of filters and
detectors. This spectral range encompasses the water spectrum and the
3OH and 3CH glucose overtone absorption bands.
Purdy and co-workers (89)(90) and Barnes et al.
(91) at Biocontrol Inc. have used light at
wavelengths >1100 nm falling on the forearm of a patient from
light-emitting diodes. The reflected light is directed into a
spectrometer and analyzed. The spectral range encompasses
3CH, 2OH, CH + OH, and the
water bands.
mechanical manipulation of tissues
In a method developed by VivaScan Inc., measurements are made by
altering the blood volume by changing pressure on a body part in a
controlled way, while performing a near-IR transmission measurement in
the 1300–1600 nm range. The spectral range encompasses
2OH, CH + OH, and water absorption
bands. Difference spectra are obtained by comparing spectra collected
with different compressions within the exposed tissue samples
(92)(93)(94)(95). Wavelength pairs in the 1300–1600 nm range are
selected by use of an acousto-optic modulator, and measurements are
taken at different compressions of a thin body part such as a web or an
earlobe. This is considered as a hardware compensation for tissue
contributions to the signal.
near-ir kromoscopic© measurements
Kromoscopy, developed by Optix Corp., is reported
extensively in the patent and commercial literature
(96)(97)(98)(99)(100)(101). It is claimed to have superior sensitivity
compared with spectroscopic methods. Kromoscopy presumably has the
potential for NI determination of glucose by using broad band light
illumination, broad band overlapping filters, and multiple detectors;
however, there is limited theoretical basis for the sensitivity
improvement over photometric methods. The method reportedly is based on
the ability of the eye to determine slight changes in color
(96)(97)(98)(99)(100)(101); however, the effect of light scattering on
Kromoscopic measurements, has not been discussed.
spatially-resolved diffuse reflectance R(r)MEASUREMENTS
In this technique, a narrow beam of light illuminates a restricted
area on the surface of a sample or a body part, and the diffuse
reflectance is measured at several distances from the illumination
point (52)(53)(54)(55)(56)(57). This method is denoted as R(r) measurement.
The value of µeff can be calculated from the
data, and both µa and
µs' values for tissue can be deduced
(55)(56). Changes in the values of
µa and µs' can then be
used to calculate the change in concentration of an analyte affecting
the tissue optical properties (102)(103)(104).
near-ir frequency domain reflectance measurements
Frequency-domain reflectance measurements use an optical system
similar to that used for spatially resolved diffuse reflectance R(r),
except that the light source and the detector are modulated at a high
frequency. The difference in phase angle and modulation between
injected and reflected beam is used to calculate
µs' and µa of the
tissue (58)(59)(60)(61)(62)(63)(64)(65)(105)(106).
optical activity and polarimetry
Polarimetry has been used for quantitative analysis of solutions
of optically active (chiral) compounds such as glucose. When a plane
polarized light beam is transmitted through a solution, its plane of
polarization is rotated by an angle, 
, which is related to the
concentration of the optically active solute
(107)(108)(109)(110)(111)(112)(113)(114)(115).
The optical rotation of several chiral compounds at their physiological
concentrations is given in Table 4
. A 5.5 mmol/L change in glucose concentration yields the
highest calculated short-term effect on . It may be possible to
detect hyperglycemic swings, assuming that changes in the
concentrations of other optically active compounds occur over a longer
time frame than that for a change in glucose concentration. Scattering
is bound to depolarize the light and decrease the measured value of
.
|
raman scattering
Raman spectroscopy has been used mainly for vibration band
assignment and qualitative analysis. Instrument complexity and
difficulties with calibration delayed its acceptance as a quantitative
analytical tool (116)(117). The advent of
holographic optical elements and charge-coupled device cameras has
allowed the design of fast Raman spectrometers with which the spectrum
is acquired over a wide range of wave numbers and averaged in 
30
s. Chemometric methods have allowed ease of generating training
sets for subsequent prediction of concentration. Long wavelength solid
state lasers have allowed shifting the excitation wavelength to the
near-IR, thus reducing the fluorescence background from biological
samples. These advances in hardware and data analysis methods allow the
measurement of the concentration of a weak Raman scatterer with a
lowest limit of detection (LOD) comparable to other spectroscopic
techniques (118)(119)(120)(121)(122)(123)(124)(125)(126)(127)(128)(129)(130).
Raman spectroscopy offers several advantages for measurement of
glucose. Raman bands are specific to glucose molecular structure, the
fundamental vibrations are monitored in Raman spectra and thus are
sharper and have less overlap compared with near-IR combination bands,
and water has a low Raman cross-section as opposed to its high IR
absorption. It is possible to detect glucose by monitoring the 2900
cm-1 CH stretch band or the CO and CC
stretch Raman bands at 900-1200 cm-1, which
represents a finger-print for glucose (118).
near-ir photoacoustic spectroscopy
Photoacoustic (PA) measurement is an alternative detection
technology for near-IR light interaction with tissues. PA is used to
detect weak absorbance in liquids and gases (131). The
tissue is excited at a wavelength that is absorbed by glucose molecules
by pulses of a 1000–1800 nm near-IR laser light. Subsequent optical
absorption causes microscopic localized heating. The increase in
temperature causes rapid thermal expansion, which generates an
ultrasound pressure wave detectable by a hydrophone or a piezoelectric
device in the cuvette or located at the skin surface
(131)(132)(133)(134)(135)(136)(137). The magnitude of a pulsed PA signal,
P, is related to the absorption coefficient of solution by:
 |
where µa is the optical absorption
coefficient, ß is the thermal expansion coefficient, is the sound
velocity, Cp is the specific heat of the
solution, and K is a proportionality constant that is
related to the bulk modulus of the medium. The Cp
of a solution decreases, whereas the acoustic velocity increases with
increasing glucose concentration. At a glucose absorption wavelength,
change in the PA signal are the result of changes in
µa, , and Cp. This
multiplicative effect increases the PA signal as a function of
concentration. The speed of sound and the Cp
values change as the total solute concentration changes. When excited
in the near-IR, PA detects the absorption caused by the overtones of
OH and CH bond vibrations of glucose and other analytes; this
absorption is subsequently converted into an acoustic pulse. PA
measurements have some sensitivity advantages over other near-IR
detection methods because a PA detector collects all generated signals
in a volume of the tissue and detects signals generated at wavelengths
longer than the range of silicon or gallium arsenide detectors.
|
In Vitro and ex Vivo Results |
|---|
|
TopAbstractIntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
near-ir reflectance and transmittance measurements
Near-IR spectrophotometric determination of glucose has been
achieved in aqueous solutions and serum-like matrices; by the use of IR
absorption bands in the 2000–2500 nm range [standard error of
prediction (SEP), 0.35 mmol/L]. Digital filtering is used as a
preprocessing step before the PLS calculation to remove spectral
features not associated with glucose (32)(33)(34)(35)(36)(37)(38). The 2270 nm
wavelength provides the lowest standard error of calibration (SEC), at
0.24 mmol/L.
near-ir frequency-domain reflectance measurements
The change in phase and attenuation of a modulated laser beam has
been measured at different glucose concentrations in tissue-simulating
phantoms (61)(62). µa remains
unchanged, whereas µs' decreases with
increasing glucose concentration. The decrease in
µs' is attributed to changes in
nmedium caused by dissolution of glucose. From
these measurements, the fractional change in
µs' of tissue is estimated as ranging between
-1 x 10-3 and -5 x
10-4 per mmol/L of glucose. This
estimation assumes that only changes in glucose concentration affect
nmedium and do not affect cell size. A 1 °C
change in temperature is estimated to decrease the
µs' of the lipid emulsion by a value that is
four- to eightfold higher than a change caused by an increase of 1
mmol/L of glucose (61)(62). Similar results are reported
using glucose, sucrose, or mannitol (60)(61)(62)(63). Because
µs 
µa at
these wavelengths, it easier to measure µs
in scattering solution and relate it to the concentration of weakly
absorbing solutes.
optical activity and polarimetry
Ex vivo measured glucose concentrations show reasonable
correlation with blood glucose concentrations
(107)(108). The improved sensitivity of
polarimeters allows measurements in small path lengths and at lower
glucose concentrations (110)(111)(112)(113)(114). Glucose has been
determined in vitro in the 5.6–33 mmol/L range in cell growth media
(SEC = 0.3 mmol/L; SEP = 0.47 mmol/L) and ex vivo in bovine
ocular fluid (SEC = 1.25; SEP = 1.13 mmol/L)
(112)(113).
raman scattering
Glucose concentration can be estimated in vitro by use of Raman
bands at 900-1150 and 2850–3000 cm-1 down to
0.7 mmol/L in water and to 7.2 mmol/L in bioreactor material by use of
laser excitation and PLS analysis (119). The glucose
concentration also has been determined in fluorescent serum and plasma
samples down to 2.5 mmol/L by monitoring the anti-Stokes Raman band at
1130 cm-1 (129). The use of this
Raman band at higher energy than that of the excitation laser beam
eliminates the red-shifted fluorescence background. The reported in
vitro LOD values are summarized in Table 5
. Some LOD values are close to the glucose physiological range.
|
near-ir pa spectroscopy
The concentration of glucose in vitro has been measured in a flow
cell in the 1.7–33 mmol/L range. Pulsed laser wavelengths centered at
1440, 1550, and 1680 nm have been used (131). The use of
pulsed solid-state lasers at 780, 830, 1300, and 1550 nm has also been
reported. Linear response curves between the pulsed PA signal and
glucose concentration have shown the same slope in the presence of
fixed concentration of cholesterol, sodium chloride, and human serum
albumin. Analytical curves in aqueous solutions indicate that the PA
signal that is excited at 1680 nm varies linearly with glucose
concentration with a slope of 
3.6 x
10-2 per mmol/L of glucose and that the
intercept varies with the concentration of the interfering compound
added (137).
|
In Vivo Studies and Results |
|---|
|
TopAbstractIntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
. Robinson et al. (69), Haaland et al.
(70), and Ward et al. (71) have reported data on
three subjects with 1.1 mmol/L < SEC < 2.1 mmol/L,
depending on the spectrometer configuration and chemometric model used.
The data represent a single data set; a calibration set for one patient
on 1 day, and are not used to predict glucose values on other days. All
of these authors have noted that the baseline shifts during
measurements and use derivatives to reduce this shift. None of these
authors, however, indicates which wavelengths are necessary or whether
additional wavelengths can improve SEC.
|
Hiese and co-workers (72)(73)(74)(75)(76) and Marbach et al. (77) have reported a correlation between the measured reflected signal of oral mucosa in the 1111–1835 nm range and blood glucose concentration . The lowest SEP reported was 2.1 mmol/L. The medium through which light was transmitted and reflected differed from the skin and tissue used in other studies. This method has a potential lag time between the measurements of glucose concentrations in blood and saliva, and saliva components and residual food in the mouth present sources of interference.
Jagemann et al. (87) have measured transmission in the 900-1200 nm range. The data for 11 test persons were analyzed using PLS or an ANN. A Clark error grid presentation shows that, depending on the type of calculation used, 59–65% of the data points fall in the A zone, 25–30% fall in the B zone, and ~10% fall in the D zone.
Brumeister et al. (82) recently studied near-IR transmission through the tongue of five type I diabetics over 39 days, performing five measurements per day. They used this procedure to reduce contamination of the data set with temporal effects. PLS calibration models were generated for the different individuals, and models were used to predict blind sample concentrations (every fifth measurement) or the later data (SEP >3 mmol/L). Concentration correlation plots were superimposed on the Clark error grid for a single subject calibration model. Approximately 50% of the 39 predicted glucose values fell in the A zone, 37% fell in the B zone, 5% fell in the C zone, and 7% fell in the D zone.
No clinical performance data have been reported on the methods of Rosenthal and et al. (83)(84)(85)(86)(87)(88). The selected number of wavelengths was limited compared with the full spectrum used in other studies (69)(70)(71). Clinical data are needed to compare the result of using this wavelength range and measurement method with those data published by other groups (69)(70)(71)(72)(73)(74)(75)(76)(77)(78)(79)(80).
The method proposed by Purdy and co-workers (89)(90) and Barnes et al. (91) is the only in vivo glucose monitoring method that has clinical data submitted for regulatory approval. However, except for news in trade publication and company press releases, no published clinical data are available to allow the merits of this technology to be evaluated. Nor have any clinical data been published on the proposed method of mechanical manipulation of tissue (92)(93)(94)(95).
Published data for near-IR Kromoscopic measurements show an uncontrolled meal-tolerance test on one subject of unknown diabetic status (97). Data are presented as a plot of the signal difference from two pairs of detectors. The signal difference changed immediately after a meal and returned to near its original value 2 hours after the meal.
spatially-resolved diffuse reflectance R(r)MEASUREMENTS
A glucose clamp experiment has been conducted in conjunction with
an R(r) measurement (102). The optical probe was affixed to
the patient's abdomen, and µs' was estimated
at 650 nm. The blood glucose was held at 5 mmol/L, and then step
changes in its concentration were induced between 5 and 14 mmol/L. The
drift in µs' independent of the glucose
concentration prevented statistical analysis.
µs' was used to track the blood glucose
concentration in 30 out of 41 diabetic patients studied
(102). µs' was reported for three
subjects as -2 x 10-3, -3.4 x
10-3, and -1.1 x
10-3/mmol/L, respectively. Changes in
µs' cannot be considered exclusively as
resulting from changes in nmedium caused by
increased blood glucose concentrations. The possibility that other
physiological processes contribute to µs'
still exists (102)(103).
near-ir frequency domain reflectance measurements
µs' and µa in
tissues have been determined in vivo by use of a frequency-modulated
805 nm laser. The optical probe was placed on the thigh of a
nondiabetic subject during an OGTT test
(60)(105)(106). The results were
similar to those reported by R(r) measurement
(102)(103).
polarimetry and raman spectroscopy
No human clinical data have been published on the use of
polarimetry or Raman spectroscopy to determine glucose concentrations
in vivo.
near-ir pa spectroscopy
Initial clinical data show a change in signal with glucose
concentration in a meal-tolerance test. However, no statistical data
analysis has been presented to show its advantage over a near-IR
transmission or reflectance measurement (134)(135)(136)(137).
other reported in vivo methods
An IR emission technique based on measurement of the fundamental
absorption bands of glucose at 9.1–10.5 µm as they affect the
intensity of the tissue black body radiation has been proposed
(138). The body emits IR energy, which can be used as the
light source that is absorbed by glucose at 10 µm. In the proposed
method, the surface of the skin is cooled to eliminate its absorptive
effect, and the emission from the subcutaneous layer is detected. In
vivo clinical calibration data on several type I diabetics have been
reported (139). A linear response between in vivo detected
glucose and blood glucose values has also been reported. Multiple
linear least squares fit to the calibration data yield an SD of 1.4
mmol/L. No prediction data have been published. Measurement of the IR
emission from the tympanic membrane by use of a filter at 10.5 µm and
the measurement of the difference in the signal to that of a known
glucose concentration has been proposed. The data in the patent example
indicate that the signal and the glucose concentration in a
meal-tolerance test have been tracked (140). A totally
different technique, which is based on body thermal effects, has been
proposed by Cho and Hoizgreve (141). They claim that
accurate measurements of changes in body temperature yield good NI
glucose calibration plots. These IR emission methods are quite
interesting because they require simpler instruments.
|
Conclusion, Predictions, and Future Directions |
|---|
|
TopAbstractIntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
3 mmol/L. A Clark error grid
presentation shows performance that is not acceptable for home glucose
meters (17)(18). The accuracy is especially poor
in the 3–6 mmol/L glucose concentration range. The performance of the
reported methods also falls short of the American Diabetes Association
recommendations for home glucose monitors. The qualitative tracking between the signal and the glucose concentration can be seen in the scattering coefficient values. The magnitude of the change in signal as a function of glucose concentration is very small, and drifts in instrument or physiological conditions are quite large. The need for more sensitive and stable scattering instrumentation is quite obvious. The source of signal tracking in different test subjects is not clear, and no statistical analysis equivalent to that utilized on transmission methods has been performed on the scattering data.
Polarimetry and Raman spectroscopy show adequate detection sensitivity in vitro, but in vivo applications have not been shown. Polarimetry requires a body part with low scattering, such as the cornea; appropriate calibration; and understanding of the lag time between the glucose concentrations in blood and in aqueous humor (108). Corneal rotation, corneal birefringence, and eye motion artifacts are potential sources of error in polarimetric ocular measurements (6)(107)(108). Some issues need to be resolved before Raman spectral measurements can be used in vivo. These include the scattering effect of tissues, potential broadening of the Raman bands, fluorescence background, the dermatological effects of laser excitation, and ways to calibrate the signal vs blood glucose concentration. The eye has been suggested as the measurement site to avoid variable tissue background fluorescence (120)(121). Low laser power is required to prevent eye injury, but this may lead to undetectable Raman signals. PA measurements track glucose concentration in vivo and have good sensitivity in vitro.
It is possible to measure changes in the glucose concentration during a hyperglycemic swing, using current instrumentation. Chemometric or ANN methods are necessary for data analysis, and it is important to guard against overfitting. Calibration has, thus far, relied on individual clinical calibration with an invasive reference method. Statistical analysis of calibration performance thus far has been insufficient to prove that NI calibration models derived from the reviewed spectroscopic techniques are based on glucose-specific information. Unless the physical effects producing the input data are understood, the numerous factors needed to optimize the prediction of results raise the concern that these results may come from an overdetermined nonfunctional calibration model.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
|
References |
|---|
|
TopAbstractIntroductionResearch Methodologies for NI...Calibration of NI Glucose...Principles of Glucose Tissue...Optical Properties of TissuesOptical Methods for Noninvasive...In Vitro and ex...In Vivo Studies and...Conclusion, Predictions, and...References |
|---|
H and CH vibrational modes for D-glucose, maltose, cellobiose, and dextran by deuterium substitution methods. Carbohydr Res 1971;19:297-310.
-D-glucose. Carbohydr Res 1972;23:407-417.
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  V. V. Sapozhnikova, D. Prough, R. V. Kuranov, I. Cicenaite, and R. O. Esenaliev Influence of Osmolytes on In Vivo Glucose Monitoring Using Optical Coherence Tomography Experimental Biology and Medicine, September 1, 2006; 231(8): 1323 - 1332. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 O. S. Khalil Noninvasive Photonic-Crystal Material for Sensing Glucose in Tears Clin. Chem., December 1, 2004; 50(12): 2236 - 2237. [Full Text] [PDF]
|
|
|
|
|
|
V. L. Alexeev, S. Das, D. N. Finegold, and S. A. Asher Photonic Crystal Glucose-Sensing Material for Noninvasive Monitoring of Glucose in Tear Fluid Clin. Chem., December 1, 2004; 50(12): 2353 - 2360. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
O. K. Cho, Y. O. Kim, H. Mitsumaki, and K. Kuwa Noninvasive Measurement of Glucose by Metabolic Heat Conformation Method Clin. Chem., October 1, 2004; 50(10): 1894 - 1898. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. B. Ko, O. K. Cho, Y. O. Kim, and K. Yasuda Body Metabolism Provides a Foundation for Noninvasive Blood Glucose Monitoring Diabetes Care, May 1, 2004; 27(5): 1211 - 1212. [Full Text] [PDF]
|
|
|
|
|
|
S.-j. Yeh, C. F. Hanna, and O. S. Khalil Monitoring Blood Glucose Changes in Cutaneous Tissue by Temperature-modulated Localized Reflectance Measurements Clin. Chem., June 1, 2003; 49(6): 924 - 934. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. B. Sacks, D. E. Bruns, D. E. Goldstein, N. K. Maclaren, J. M. McDonald, and M. Parrott Guidelines and Recommendations for Laboratory Analysis in the Diagnosis and Management of Diabetes Mellitus Clin. Chem., March 1, 2002; 48(3): 436 - 472. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
C. V. Eddy and M. A. Arnold Near-Infrared Spectroscopy for Measuring Urea in Hemodialysis Fluids Clin. Chem., July 1, 2001; 47(7): 1279 - 1286. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
R. A. Shaw, S. Low-Ying, M. Leroux, and H. H. Mantsch Toward Reagent-free Clinical Analysis: Quantitation of Urine Urea, Creatinine, and Total Protein from the Mid-Infrared Spectra of Dried Urine Films Clin. Chem., September 1, 2000; 46(9): 1493 - 1495. [Full Text] [PDF]
|
|
|
|
|
|
R. Weitgasser, B. Gappmayer, and M. Pichler Newer Portable Glucose Meters--Analytical Improvement Compared with Previous Generation Devices? Clin. Chem., October 1, 1999; 45(10): 1821 - 1825. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
H. A. MacKenzie, H. S. Ashton, S. Spiers, Y. Shen, S. S. Freeborn, J. Hannigan, J. Lindberg, and P. Rae Advances in Photoacoustic Noninvasive Glucose Testing Clin. Chem., September 1, 1999; 45(9): 1587 - 1595. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. J. Burmeister and M. A. Arnold Evaluation of Measurement Sites for Noninvasive Blood Glucose Sensing with Near-Infrared Transmission Spectroscopy Clin. Chem., September 1, 1999; 45(9): 1621 - 1627. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. F. Malin, T. L. Ruchti, T. B. Blank, S. N. Thennadil, and S. L. Monfre Noninvasive Prediction of Glucose by Near-Infrared Diffuse Reflectance Spectroscopy Clin. Chem., September 1, 1999; 45(9): 1651 - 1658. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. E. Collison, P. J. Stout, T. S. Glushko, K. N. Pokela, D. J. Mullins-Hirte, J. R. Racchini, M. A. Walter, S. P. Mecca, J. Rundquist, J. J. Allen, et al. Analytical Characterization of Electrochemical Biosensor Test Strips for Measurement of Glucose in Low-Volume Interstitial Fluid Samples Clin. Chem., September 1, 1999; 45(9): 1665 - 1673. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
