Tight control of blood glucose levels has been shown to dramatically reduce the long-term complications of diabetes. Current invasive technology for monitoring glucose levels is effective but underutilized by people with diabetes because of the pain of repeated finger-sticks, the inconvenience of handling samples of blood, and the cost of reagent strips. A continuous glucose sensor coupled with an insulin delivery system could provide closed-loop glucose control without the need for discrete sampling or user intervention. We describe an optical glucose microsensor based on absorption spectroscopy in interstitial fluid that can potentially be implanted to provide continuous glucose readings. Light from a GaInAsSb LED in the 2.2-2.4 μm wavelength range is passed through a sample of interstitial fluid and a linear variable filter before being detected by an uncooled, 32-element GaInAsSb detector array. Spectral resolution is provided by the linear variable filter, which has a 10 nm band pass and a center wavelength that varies from 2.18-2.38 μm (4600-4200 cm<sup>-1</sup>) over the length of the detector array. The sensor assembly is a monolithic design requiring no coupling optics. In the present system, the LED running with 100 mA of drive current delivers 20 nW of power to each of the detector pixels, which have a noise-equivalent-power of 3 pW/Hz<sup>1/2</sup>. This is sufficient to provide a signal-to-noise ratio of 4500 Hz<sup>1/2</sup> under detector-noise limited conditions. This signal-to-noise ratio corresponds to a spectral noise level less than 10 μAU for a five minute integration, which should be sufficient for sub-millimolar glucose detection.
Optical properties of whole bovine blood are examined under conditions of different glucose loadings. Partial least-squares (PLS) is used to compute calibration models for glucose from spectra collected over the combination spectral region (5000 - 4000 cm<sup>- 1</sup>) and first overtone - short wavelength spectral regions (9000 - 5400 cm<sup>-1</sup>). These models achieve a prediction accuracy of approximately 1mM. Calibration models built for specific glucose absorption regions perform better than models generated strictly from the short wavelength region in which light scattering effects dominate. Net analyte signal (NAS) analysis is employed to investigate the spectral information that forms the basis for the models. The NAS reveals the portion of the glucose spectrum that is orthogonal to the spectral variance induced by the blood matrix. To investigate the selectivity of the spectral measurements, the glucose NAS is compared to residual absorbance spectra formed after subtraction of the non-glucose variance (estimated by application of principal component analysis to a set of blood samples with endogenous glucose concentrations). A match between the NAS and the residual spectra reveals that direct information associated with absorption of light by the glucose molecule is present in the measured data. A similar comparison is made with the regression vector associated with the PLS model. A match between the NAS and regression vector confirms that the correlations encoded in the calibration model do, in fact, arise from glucose absorption information. The results obtained through this work demonstrate that NAS analysis is a valuable tool for use in investigating the selectivity of multivariate calibration models.
Tight control of blood glucose levels has been shown to dramatically reduce the long-term complications of diabetes. Current invasive technology for monitoring glucose levels is effective but underutilized by people with diabetes because of the pain of repeated finger-sticks and the cost of reagent strips. Optical sensing of glucose could potentially allow more frequent monitoring and tighter glucose control for people with diabetes. The key to a successful optical non-invasive measurement of glucose is the collection of an optical spectrum with a very high signal-to-noise-ratio in a spectral region with significant glucose absorption. Unfortunately, the optical throughput of skin is very small due to absorption and scattering. To overcome these difficulties, we have developed a high-brightness tunable laser system for measurements in the 2.0-2.5 μm wavelength range. The system is based on a 2.3 micron wavelength, strained quantum-well laser diode incorporating GaInAsSb wells and AlGaAsSb barrier and cladding layers. Wavelength control is provided by coupling the laser diode to an external cavity that includes an acousto-optic tunable filter. Tuning ranges of greater than 110 nm have been obtained. Because the tunable filter has no moving parts, scans can be completed very quickly, typically in less than 10 ms. We describe the performance of the laser system and its potential for use in a non-invasive glucose sensor.
We have performed in vivo measurements of near-infrared rat skin absorption in the 4000-5000 cm<sup>-1</sup> spectral
range (2.0-2.5 μm wavelength) during a glucose clamp experiment. The goal of this work is to identify the presence
of glucose-specific spectral information in order to evaluate the requirements for a noninvasive transcutaneous
glucose instrument. Skin spectra are collected using an FTIR spectrometer coupled with a fiber-optic interface.
In the experiment, an animal is allowed to stabilize at a euglycemic level for three hours while blood glucose
values are monitored using samples taken from an arterial catheter. The blood glucose level is then increased
above 30 mM by venous infusion of glucose and held for two hours, after which it is allowed to return to normal.
Spectra are recorded continuously during the procedure and are analyzed to identify changes due to the glucose
variations. Because the change in absorbance due to an increase in glucose concentration is small compared
to changes due to other variations (e.g., the thickness of the skin sample), a simple subtraction of absorbance
spectra from the hyperglycemic and euglycemic phases is not instructive. Instead, a set of principal components
is determined from the euglycemic period where the glucose concentration is constant. We then examine the
change in absorbance during the hyperglycemic period that is orthogonal to these principal components. We find
that there are significant similarities between these orthogonal variations and the net analyte signal of glucose,
which suggests that glucose spectral information is present.
Optical properties of whole bovine blood are examined under conditions of different glucose loadings. A strong dependency is established between the scattering properties of the whole blood matrix and the concentration of glucose. This dependency is explained in terms of variations in the refractive index mismatch between the scattering bodies (predominately red blood cells) and the surrounding plasma, and also by variations in the size and shape of the red blood cells. Measurements in the presence of a well-known glucose transport inhibitor indicate that variations in refractive index mismatch are related to the penetration of glucose into the red blood cells. In addition, results measure the glucose dependent aggregation properties of red blood cells. In this experiment, pulsations in transmitted light intensity are explained by cycles of aggregation and disaggregation of red blood cells in response to a propagating pump wave through the blood matrix. Magnitude of these pulsations depends on the concentration of glucose in the sample. Results are also presented to characterize the time-dependent variation in light transmission in response to a step change in glucose concentration. Finally, multivariate calibration models are presented for the measurement of glucose in a whole blood matrix. These models are based on near infrared spectra and Kromoscopy data collected from eighty different samples prepared from a single whole blood matrix. The best model is generated for combination near infrared spectra, which provides a standard error of prediction is less than 1 mM over a concentration range of 3 to 30 mM.
Four-channel Kromoscopic analysis is demonstrated for <i>in vitro</i> measurement of glucose in cow blood in the spectral range of the first overtone of CH stretching vibrations and also the short wavelength near infrared region. The sample set included 48 blood samples with glucose concentration randomly distributed in the range 4.55 -37.60 mM. Solid glucose was added to each blood sample to create a wide range of glucose concentrations. During the measurement procedure blood was circulated through a 1 mm path length cell at a flow rate of 3 m:/min. Regular pulsations on the order of 1% were observed and corresponded to periodic aggregation and disaggregation of red blood cells under these flow conditions. Glucose can be quantified with a standard error of prediction of 1 mM over all blood samples. The impact of scattering and absorption proceses are also discussed.
We describe on-line optical measurements of urea concentration during the regular hemodialysis treatment of several patients. The spectral measurements were performed in the effluent dialysate stream after the dialysis membrane using an FTIR spectrometer equipped with a flow-through cell. Spectra were recorded across the 5000-4000 cm<SUP>-1</SUP> (2.0-2.5 micrometers at 1-minute intervals. Optically determined concentrations matched concentrations obtained from standard chemical assays with a root-mean-square error of 0.29 mM for urea (0.8 mg/dl urea nitrogen), 0.03 mM for creatinine, 0.11 mM for lactate, and 0.22 mM for glucose. The observed concentration ranges were 0-11 mM for urea, 0-0.35 mM for creatinine, 0-0.75 mM for lactate, and 9-12.5 mM for glucose.
The ability of Kromoscopy to measure glucose selectively is demonstrated in solutions composed glucose, urea, triacetin, bovine serum albumin (BSA), cholesterol, and hemoglobin (Hb). Kromoscopic measurements are made with a four-channel instrument designed for measuring light between 1500 and 1900 nm. The channels are configured to respond to four individual bands of near infrared light centered at 1600, 1700, 1750, and 1800 nm. An equation is proposed that describes the relative response for each channel as a function of relevant experimental parameters. This equation predicts the linear response observed for these types of measurements as a function of solute concentration. In addition, molar absorptivities are provided for glucose, urea, triacetin, BSA, Hb, and water. The non-negligible absorptivity of water demands the consideration of water displacement caused by solute dissolution. Channel responses are measured for a series of thirty-one samples. The chemical composition of these samples is designed to minimize the correlations between glucose concentration and the concentrations of all other solutes. Likewise, these samples provide negligible correlation between the concentration of glucose and the extent of water displacement. A calibration model is constructed for glucose by using a conventional P-matrix multiple linear regression analysis of the four-channel information. The resulting model demonstrates selectivity for glucose with values of 1.27 and 1.34 mM for the standard errors of calibration and prediction, respectively, over a glucose concentration range of 1.9 to 19 mM.
The selectivity of a four-channel Kromoscopic analysis is demonstrated for the measurement of glucose in separate binary and tertiary matrices. A novel virtual search procedure is used to identify three different sets of four, overlapping transmission filters. The first filter set includes filters centered at 900, 1300, 1410, and 1538 nm and is selected to differentiate glucose and urea in a series of binary mixtures. These binary mixtures were prepared with independent levels of 1-10 mM glucose and 9- 213 mM urea dissolved in an aqueous phosphate buffer solution. A second filter set contains filters centered at 1064, 1100, 1224, and 1290 nm and is used to measure glucose in a series of tertiary mixtures composed of glucose, urea and bovine serum albumin. This tertiary matrix consists of 2-13 mM glucose, 13-129 mM urea and 0.05-0.46 g/L bovine serum albumin dissolved in the same type of buffer. Multilinear regression is used to relate the relative Kromoscopic responses to the concentration of glucose in these sample solutions. In both cases, the prediction errors are on the order of 0.6-0.8 mM. The impact of solution temperature is also investigated by examining glucose responses obtained from solutions maintained at temperatures ranging from 35 to 39 degree(s)C. The filter set used in this experiment is composed of filters centered at 1100, 1150, 1254, and 1300 nm. Results from this particular filter set indicate that the directionality and magnitude of the glucose responses are independent of solution temperature. Finally, accurate glucose measurements are demonstrated when a same-temperature blank is used to generate the relative channel response.