| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Articles |
Department of Chemistry and Optical Science and Technology Center, University of Iowa, Iowa City, IA 52242.
a Author for correspondence. Fax 319-353-1115; e-mail mark-arnold{at}uiowa.edu
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResults and DiscussionConclusionsReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionConclusionsReferences |
|---|
Scattering and molecular absorption combine to attenuate near-IR light as it passes through human tissue. The principal phenomena responsible for light attenuation depend on the spectral range. Molecular absorption dominates in the combination spectral range (2.0–2.5 µm) where water, fat, and protein are the primary absorbers. In comparison, scattering is more important in the first-overtone region (1.52–1.85 µm), although molecular absorption by water and fat is still significant. The impact of protein is much less in the first-overtone region compared with the combination region (11). The thickness of the tissue within the measurement site is critical regardless of the spectral range. The tissue must be thick enough to provide sufficient glucose for reliable detection and yet thin enough to yield sufficiently large radiant powers for high spectral quality. The composition and thickness of the tissue probed by the transmitting radiation strongly impact analytical performance by affecting the SNR of the measurement.
In this study, first-overtone near-IR spectra collected noninvasively from human subjects are characterized with the goal of identifying potential measurement sites for research purposes. Results from an in vitro model (11) are used to identify the chemical and physical properties of the ideal measurement site. Noninvasive human spectra are presented for several potential measurement sites. The amounts of water and fat within the probed tissue are estimated by a regression analysis. The suitability of each putative measurement site is evaluated on the basis of the effective aqueous pathlength and the amount of fatty tissue. Results identify the tongue as an excellent measurement site for noninvasive human spectra over the first-overtone region.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionConclusionsReferences |
|---|
Noninvasive human spectra were collected with a modified Midac M series Fourier transform infrared spectrometer. Details of these modifications are provided elsewhere (12). Briefly, the original source was replaced with a 150 W tungsten-halogen lamp configured as a projector bulb with a back-reflector plate (model EFR; Apollo). This lamp was powered by a DCR 40-13B2 DC power supply from Sorensen. A 630 nm interference filter was positioned in front of each photodiode detector associated with the interferometer. These filters were necessary to prevent detector saturation by the strong incident radiation of the 150 W source. The internal temperature of the interferometer housing was controlled by a feedback air-cooling system. The internal temperature of the interferometer housing was measured and used to control the speed of the cooling fan. This arrangement permitted operation at 41.0 °C with constant control to ± 0.1 °C. The mirror velocity was set at 0.24 cm/s, which represented the most stable mirror movement of the available settings. Finally, a 1-mm diameter, thermally electric cooled, InGaAs detector (ETX-1000TE-GR1.9; Eppitax) was incorporated into the spectrometer. The detector temperature was maintained at 15 °C to lower the noise relative to room temperature and still provide high detectivity up to 1.82 µm.
reagents
Aqueous solutions were prepared in 0.1 mol/L phosphate buffer
maintained at pH 7.35 with 4.4 g/L 5-fluorouracil added to prevent
bacterial contamination. Glucose solutions were prepared by dissolving
weighed, dried amounts of reagent-grade material. Fat tissue was from
bovine samples obtained from a local supermarket. Fat filters were
prepared as described previously (11) by sandwiching a known
amount of blended fat material between parallel sapphire windows with
Teflon spacers. Bundles of quartz fibers were purchased from Dolan
Jenner Industries.
procedures
In vitro models of the human body were prepared as described
previously (11) by combining individual layers of fatty
tissue and aqueous buffer. The relative thickness of each layer was
adjusted so that the first-overtone spectra collected from the in vitro
model closely matched those obtained from a given measurement site on a
particular subject (11).
In vivo spectra were collected either by using a fiber optic probe or by focusing the incident beam directly onto the tissue section of interest. In both cases, the tissue section in question was sandwiched between two flat sapphire windows. In the fiber optic configuration, one fiber bundle was used to bring the incident light from the source optics to the measurement site. A fraction of the transmitted radiation was collected by a second fiber bundle, and this light was directed to the detection optics. This fiber optic arrangement permitted collection of spectra from hard to reach locations. Considerable losses of radiant power through the fiber-optic couplings, however, led to low SNRs and poor spectral quality. Higher SNRs were achieved by directly focusing the incident radiation onto the tissue. A 1-inch diameter, 25-mm focal length convex lens was used to focus the incident beam onto the sample tissue. The InGaAs detector module was positioned immediately behind the tissue with the window on the photodiode casing touching the sapphire window. The best performance was obtained by focusing the incident light through the sample tissue onto the surface of the photodiode.
In vivo absorbance spectra were computed from single-beam tissue spectra and attenuated single-beam air background spectra. Single-beam air background spectra were collected in either the fiber optic or direct focusing configuration described above. Neutral density filters with known percentages of transmissions were used to attenuate the air spectra and avoid detector saturation. Intensity values for each air background spectrum were multiplied by the appropriate factor to account for the extent of attenuation provided by the neutral density filter. Similarly, each background spectrum was adjusted to account for any differences in instrument gain settings between a given air background and the corresponding sample spectrum. Absolute absorbance was then estimated using the ratio of each sample spectrum to the corresponding "corrected" air background spectrum.
Effective optical pathlengths through the fat and aqueous regions of
the tested measurement sites were estimated by the regression method
described in detail elsewhere (11). Briefly, the tissue of
interest was sandwiched between two sapphire windows, and calipers were
used to measure the combined thickness of the tissue/window assembly.
The thickness of the tissue section was calculated by subtracting the
known window thickness. Absolute absorbance spectra collected
noninvasively from human tissue were fitted by regression to standard
absorbance spectra for fat and water. The regression calculation was
performed according to the following expression:
 | (1) |
where ST, Sw, and Sf correspond to absolute absorbance spectra for known thickness of tissue, water, and fat, respectively, and ßi values represent the corresponding regression coefficients. Effective optical pathlengths were obtained by multiplying the respective regression coefficients by the known thicknesses of the individual standard materials of fat and water.
|
Results and Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionConclusionsReferences |
|---|
physical and chemical requirements
The Beer-Lambert law of absorption spectroscopy indicates that
glucose absorbance values depend on the absorptivity, concentration,
and optical pathlength. Molar absorptivity is fixed at each wavelength
by the vibrational absorption properties of glucose. The concentration
range is defined by the intended clinical application and is generally
2–22 mmol/L (36–396 mg/dL) for blood glucose. Optical pathlength is
the only adjustable experimental parameter, which makes it vitally
important when designing noninvasive measurement technology. The
optical pathlength must be sufficiently long to allow enough glucose
molecules within the optical path to produce a measurable absorbance
signal. If the optical pathlength is too long, however, excessive light
attenuation will increase noise by reducing radiant power at the
detector. The ideal pathlength depends on instrumental performance and
spectral range.
Instrumental performance is best represented as the root-mean-square
noise on 100% lines (RMSN-100%). This value is obtained by collecting
two single-beam spectra for the exact same sample. One spectrum is
divided by the other, and in the ideal case of no noise or spectrometer
variation, the result is a horizontal line at 1.00 (or 100%
transmission). Both spectrometer noise and instrument variation are
clearly evident when the resulting product ratio is plotted as a
function of wavelength [see Fig. 2
in Ref. (13)]. As is
detailed elsewhere, the RMSN-100% tracks the optical throughput of the
sample with lower noise at higher throughput
(13)(14). It is convenient to report the
RMSN-100% in microabsorbance units (µAU) by converting the
percentage of transmission to absorbance units [log (1/T)]. The
actual RMSN-100% value is found by fitting the data to either a first-
or second-order polynomial function and computing the root-mean-square
value for the data relative to this fitted function. RMSN-100% values
are best provided over a series of defined and narrow spectral regions.
Generally, values computed over wide spectral ranges will be dominated
by the noisiest regions, making it difficult to characterize spectral
quality in the lower-noise regions where most of the analyte-specific
information resides.
|
Spectral range is critical because the shape and size of the glucose absorption features differ significantly in the combination (15)(16)(17), first-overtone (14), and short-wavelength (17)(18) regions of the near-IR spectrum. The focus of this report is the first-overtone region, where optical pathlengths of 5–10 mm are necessary to measure clinically relevant concentrations of glucose in aqueous solutions (14). Our previous work indicated that prediction errors <0.6 mmol/L are possible for glucose but only with 5–10 mm optical pathlengths coupled with RMSN-100% values <10 µAU over the 5975–5850 cm-1 spectral range (14). These values were established under ideal conditions with a Fourier transform spectrometer in combination with fixed optical pathlengths and controlled sample temperatures.
features of noninvasive human spectra
Absorbance spectra collected across human tissue are dominated by
the absorption properties of water and fatty tissue (11).
The high water content of body tissue and the high absorptivity of
O
H absorption bands at 7000 and 5200 cm-1
restrict the optical transmission window for first-overtone spectra to
6400–5600 cm-1. First overtones of CH
vibrations are located within this optical window. The high CH content
of body fat is evident within in vivo spectra, in which fat absorption
is observed as two strong, overlapping absorbance bands centered at
5790 and 5690 cm-1. Glucose, on the other hand,
possesses three absorption bands centered at 6200, 5920, and 5775
cm-1. Fundamental studies indicate the optimum
region for partial least-squares modeling is the 5975–5850
cm-1 spectral range, which includes the 5920
cm-1 absorption band of glucose (14).
The first-overtone spectrum in Fig. 1
illustrates the effective transmission window through human
tissue. Fig. 1
includes a single-beam spectrum collected across the
webbing tissue between the thumb and forefinger (spectrum C), a
single-beam spectrum of a matching in vitro model (spectrum B), and an
absorbance spectrum of dissolved glucose (spectrum A). Comparison of
these spectra reveals the principal complication caused by body fat
when attempting to measure glucose from noninvasive human spectra.
Fatty tissue absorbs significant amounts of the incident radiation over
an important portion of the glucose absorbance spectrum. Fat absorption
reduces the optical throughput over the critical 5920
cm-1 glucose absorption band, thereby adversely
affecting spectral quality by reducing the SNR for the measurement.
This point is clear when comparing RMSN-100% values for webbing and
buffer spectra. A typical RMSN-100% value over the 6000–5900
cm-1 spectral range is 271.1 µAU for a 6.0-mm
thick sample of webbing tissue (see Table 2
below). Over this same
spectral range, RMSN-100% values are ~5–9 µAU for water samples
with similar thickness (5.2 and 10 mm) (14). Fat within the
webbing reduces spectral quality by a factor of 30–50 right in the
most important spectral region for glucose.
|
|
The impact of body fat on the ability to measure glucose from spectra collected noninvasively from the human body stems from a reduction in optical throughput over the spectral range that contains glucose-specific information. Fortunately, the absorption features of glucose do not overlap directly with those of fatty tissue but are shifted slightly to higher frequencies.
putative measurement sites
From the above analysis, the ideal measurement site for
noninvasive first-overtone spectra provides an effective aqueous
pathlength of 5–10 mm, high radiant throughput, and minimal body fat.
Various putative measurement sites were evaluated to identify those
that most closely match these characteristics. Spectra from the tongue,
nasal septum, cheek, upper lip, lower lip, and the webbing tissue
between the thumb and forefinger were examined.
Sample absorbance spectra are presented in Fig. 2
for the examined tissues. Each spectrum in Fig. 2
includes the
measured absorbance spectrum (relative to an air background) along with
the fitted spectrum from the regression analysis based on Eq. 1
. These
spectral results correspond to 6-mm thick samples of tissue for all
sites except the webbing, which corresponds to a thickness of 5.75 mm.
The fat absorption bands are evident in the webbing spectrum.
Conversely, the magnitude of fat absorption is strikingly lower for the
other five measurement sites.
Regression analysis can be used to estimate the amounts of fat and
water in the optical path. Results of this regression analysis are
presented in Table 1
for the various tested measurement sites. These values
corroborate inspection of the absorbance spectra and indicate that the
amount of fat tissue is substantially less in all tested tissues
relative to the webbing. The tongue possesses the lowest fat content
overall, with a fat thickness one order of magnitude lower than the
webbing.
|
comparison between webbing and tongue tissues
Individual webbing spectra were collected from 19 volunteers. In
each case, the webbing tissue was held snugly between two sapphire
windows maintained 5.75 mm apart. The regression analysis indicates the
effective optical pathlength through water and fat regions of the
tissue. The effective optical pathlengths for the webbing were
4.67 ± 0.41 mm through water and 2.4 ± 0.52 mm through fat.
Similarly, tongue spectra were collected from 10 volunteers. In this
case, the distance between the sapphire windows was 6 mm, and the
regression analysis revealed effective optical pathlengths through
water and fat of 5.90 ± 0.30 mm and 0.20 ± 0.04 mm,
respectively. These values are consistent with those found in our
screening experiment and reported in Table 1
.
Differences in optical throughput across webbing and tongue tissues are
evident in Fig. 3
. The single-beam spectra presented in Fig. 3
were collected
under identical instrumental conditions for 6-mm thick samples of
webbing and tongue tissues. An absorbance spectrum of glucose is also
provided to enhance the comparison. Clearly, the tongue spectrum offers
greater radiant throughput, particularly in the region of the 5920
cm-1 glucose absorption band. The larger radiant
powers at the detector should provide high spectral quality for the
tongue spectra.
|
Spectral quality can be judged by comparing RMSN-100% values computed
for these two tissues. In this analysis, 10 back-to-back single beam
spectra were collected for each tissue type. Samples were removed and
reinserted into the sample holder between each spectrum. The RMSN-100%
values were then computed for sequentially collected spectra. The mean
values for each tissue are tabulated in Table 2
for 100 cm-1 segments over the
6400–5800 cm-1 spectral range. The magnitude of
these RMSN-100% values reflect both the optical throughput of the
measurement and the reproducibility of sampling the tissue between
sequential spectra.
The RMSN-100% values were lower for the tongue tissue over the
spectral ranges lower than 6100 cm-1. The
differences are most striking between 6000–5900 and 5900–5800
cm-1 where fat absorbs. Fig. 4
shows representative 100% lines obtained from this experiment.
Systematic variations in the 100% lines for webbing tissue correspond
to absorption bands for fatty tissue. The presence of such features in
these 100% lines illustrates the variability induced by sampling
slightly different regions of the tissue in question. The superimposed
glucose absorbance spectrum highlights the impact of this fat-induced
variance on the glucose measurement. In contrast, the representative
100% lines for the tongue spectra show very little variation in the
fat-sensitive spectral region. The lower RMSN-100% values reflect this
reduction in variance.
|
The RMSN-100% values reported in Table 2
for the spectral regions
above 6100 cm-1 reveal poorer spectral quality
for the tongue tissue relative to the webbing. These results reflect a
challenge when working with the tongue as a measurement site. The
challenge is to stabilize the tongue while the spectrum is being
recorded. Each spectrum requires 1 min for collection. With the sample
holder used in this experiment, the tongue must be fully extended
throughout this entire period. Slight movements during data collection
are unavoidable and certainly add variation to the measurement.
Conversely, the webbing tissue is easier to hold in a constant position
during data collection. Higher RMSN-100% values for the tongue spectra
reflect this difference between these two measurement sites.
|
Conclusions |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionConclusionsReferences |
|---|
|
Acknowledgments |
|---|
|
Footnotes |
|---|
2 Nonstandard abbreviations: Near-IR, near-infrared; SNR, signal-to-noise ratio; InGaAs, indium-gallium-arsenide; RMSN-100%, root-mean-square noise on 100% lines; and µAU, microabsorbance unit(s).
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResults and DiscussionConclusionsReferences |
|---|
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  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]
|
|
|
|
|
|
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]
|
|
|
|
 |
|
 G. L. Coté Noninvasive and Minimally-Invasive Optical Monitoring Technologies J. Nutr., May 1, 2001; 131(5): 1596S - 1604. [Abstract] [Full Text]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
