Progress Report No. 2-3, March 31, 1999 The Home Automation and Healthcare Consortium Noninvasive Blood Glucose Analysis using Near Infrared Absorption Spectroscopy Prof. Kamal Youcef-Toumi Principal Investigator Vidi A. Saptari Graduate Research Assistant Abstract A modular Fourier-transform spectrometer was designed and built for the noninvasive blood glucose measurement research using the near-infrared absorption technique. A series of experiments were performed to evaluate the spectrometer s efficiency and its stability. The efficiency, which represents the photon loss through the spectrometer, is proportional to the achievable signal-to-noise ratio. The stability of the instrument is important for measurement repeatability, which would affect the sensitivity of the quantitative assessment of glucose concentration. A preliminary experiment on aqueous glucose solutions was performed, with the purpose of identifying the relevant spectral bands for glucose concentration prediction. It is found that there are two relatively strong bands where glucose molecules affect the water absorption spectrum. One is around 5950 cm around 7250 cm, where there is a glucose absorption band, and the other is, where the magnitude of water absorption is reduced by the presence of glucose molecules. Simple algorithm utilizing spectral information from these two bands may be sufficient for glucose concentration prediction. 1
I. Introduction This work is motivated by the profound need for a noninvasive way to measure glucose concentration in the blood. A noninvasive glucose monitoring device would provide a safer and a more convenient method to treat and control diabetes. The goal of diabetes therapy, within and outside hospital, is to approximate the 24-hour blood glucose profile of a normal individual, which necessitates continuous monitoring. Several optical techniques have received considerable attention over several years, which include Raman, absorption and polarimetry spectroscopy [1-4]. Among them, the absorption technique in the near-infrared region was selected for our research. As discussed in the previous reports, one of the advantages of the absorption technique is its relatively stronger signal-to-noise ratio. Furthermore, it requires simpler and less expensive instrumentation than the other optical methods. We have completed the development of a modular Fourier-transform spectrometer, which would be used for the subsequent experiments. The work was motivated by the need for a versatile and dedicated instrument for research in the field of noninvasive blood glucose sensing. Great emphasis is placed to ensure that each element is designed or chosen to give maximum signal-to-noise ratio in the near-infrared region. Overview of the system design is presented in the previous report [5]. In this report, we present the performance evaluation of the spectrometer on the basis of measurements chosen because of their impact on the instrument s capability for blood glucose concentration prediction. Identification of relevant wavelength ranges for the glucose analysis, with the experimental results is also presented in this paper. Finally, we discuss challenges and obstacles we anticipate. II. (i) System Description Theory of Operation Figure 1 shows a schematic of the FT spectrometer setup. A white light from a tungsten halogen source is collected and collimated by a concave mirror and a pair of lenses. An aperture is used to limit the amount of divergence of the light entering the interferometer. The interferometer is of Michelson type, comprising a beamsplitter and two perpendicular plane mirrors. One of the mirrors moves linearly in the axis as indicated in the diagram. This produces a variable path difference between the two beams reflected off of the mirrors. The recombined beam is then focused onto the sample and onto the detector. The detector measures the interferogram, which is stored in the computer. This signal is then Fourier-transformed to obtain the spectrum. A second Michelson interferometer with an HeNe reference laser, as indicated in the diagram, is used for controlling the velocity and position of the linear motion. In addition, it is also responsible for clocking the data acquisition at precise and repeatable locations of mirror retardation. This would enable repetitive scanning to improve the signal-to-noise ratio. 2
Reference laser Laser detector Beam-splitter (Interferometer 2) Plane Mirror Concave mirror Light source Objective lens Aperture Collimating lens +1mm Moving plane mirror Beam-splitter (Interferometer 1) Plane mirror Liquid sample Focusing lens Detector Figure 1: Fourier transform spectrometer layout (ii) System Implementation and Components Design Consideration From the literature and the previous preliminary experiments, it has been realized that signals due to the blood glucose variations in the physiological range are weak. Furthermore, in the noninvasive application, there is further loss of signal due to scattering. As a result, realization of accurate blood glucose quantification would likely be limited by the amount of signal-to-noise ratio achievable. Hence, it is important that each element of the spectrometer is designed to maximize the signal-to-noise ratio. The spectrometer uses a tungsten-halogen light source that provides a continuous spectrum from the visible through the near-infrared. An extended, thermoelectricallycooled InGaAs detector is selected for maximum signal-to-noise ratio in the 1000nm - 2200nm wavelength region. For minimum loss due to absorption, Calcium-fluoride lenses are used for light collection and collimation. The photograph below shows the setup of the system. The absorption spectrum in the near-infrared region is usually broad and overlapping. Therefore, high spectral resolution is not required. Since throughput is inversely related to the resolution, it is advantageous to match the spectrometer resolution to the required resolution [5]. Light Source 3
White Light InGaAs Photodetector Beamsplitter HeNe Laser Sample Photograph 1: Fourier-transform spectrometer setup III. (i) Spectrometer Characterization and Performance Evaluation Efficiency Efficiency here is defined as the power of radiation received by the detector as compared with the power emitted by the source. This represents the photon loss through the spectrometer, which may be due to the absorption by the lenses and the beamsplitter, the loss to the environment, as well as the loss due to misalignment of the interferometer. It is assumed to be a constant Sample value throughout the relevant wavelength range. The equation below shows its relation with the achievable signal-to-noise ratio [6]. where, U SNR = v ( T). Θ. v. t NEP 1/ 2 Voice-coil. ζactuator Voice-coil Actuator U v (T) is spectral density at wavenumber v from a black body source at a temperature T ( W / sr. cm 2. cm 1 ) 2 Θ is the throughput of the system ( cm. sr) v is the resolution of the spectrum ( cm ) t is the integrating time in seconds 4
ζ is the efficiency accounting for losses due to the optical components NEP stands for noise equivalent power, which is a sensitivity figure-of-merit of / the detector ( W. Hz 2 ) In the equation above, the source is assumed to be a black-body radiation. In the case of noninvasive measurement, the source term U v (T ) would be replaced by the radiance of the diffuse-reflected light from the sample, such as the skin. To assess the spectrometer efficiency, light from the tungsten-halogen source was directed onto the detector through a narrow-band filter at 1600nm. The radiation power at the detector, Ρ 1 at the output of the preamplifier, was recorded. The procedure was done at full throughput, limited by the area of the detector and the speed of the focusing optics. Using the same light source and band-pass filter, an interferogram was recorded using the spectrometer. The peak of the interferogram, Ρ 2, which corresponded to the radiation power when all the wavelengths interfere constructively, was recorded. Efficiency was then computed as Ρ 2 Ρ1, which was equal to 0.085. This value is very good considering the complexity of the system, and is in good agreement with the rule-of-thumb estimations of Fourier-transform spectrometer efficiencies of 10% [7]. (ii) Stability Another criterion of importance is the spectrometer stability, which would translate into repeatability of the measurements. The weak glucose signal intrinsic to the absorption technique requires excellent stability of the spectrometer. The repeatability of the spectrum is especially important in the regions where glucose absorption bands are to be analyzed. Irreproducibility of measurements could arise from many sources, such as change in alignment, interferometer thermal expansion, change in sample temperature, scattering and variation due to the light source itself. To assess stability, water spectra were acquired every 3 minutes over a period of one hour. This duration and periodicity was chosen to simulate an actual experiment. Each spectrum was a result of a Fourier-transformation of 30 averaged, double-sided interferograms. The first spectrum was subtracted from each subsequent spectrum to get the difference spectra, which are plotted in figure 2. To examine the impact of these variations, the plot should be examined together with the difference spectra of glucose solutions, which is presented in the next section. 2 1.5 1 0.5 0 5
Relative Magnitude Figure 2: Differential spectra of repetitive water measurements IV. Measurement of Glucose in Water A preliminary aqueous glucose measurement experiment was done with the purpose of identifying the relevant bands for glucose analysis, as well as examining the capability of the instrument for the task. Glucose solutions with high concentration were used in this study. The lowest concentration was 50 mmol / L, which corresponded to about 4 times as high as the upper physiological range. Figure 3 shows the transmission spectra of pure water and glucose solution. Figure 4 shows the differential spectra of glucose solutions with different concentrations with respect to the water transmission spectrum. It can be seen from the two plots that the presence of glucose affects the spectrum of water in two regions. One is between 5800 cm and 6100 cm where there is a glucose absorption band. The other is a narrower band at around 7250 cm, where there is, in fact, more transmission due to the presence of glucose. This might result from the displacement of water molecules by glucose molecules. In these regions, effects of glucose on water spectra are more pronounced than in the lower wavelength regions, which have been considered by several other investigators [2, 8]. However, water absorption here is much higher. Therefore, in order to probe these bands, a long-wave-pass filter with a cutoff around 1400nm was required. Otherwise, they would not be visible at all, as the shorter-wave radiation would dominate. The water-glucose differential spectra (figure 4) were obtained using the same throughput and detector amplifier gain setting as the experiment to obtain the stability differential spectra (figure 2). Therefore, the magnitudes of the spectra between the two figures can be directly compared. At 50 mmol / L, glucose altered the water spectrum by 6 units at around 6000 cm and by 3 units at 7250 cm. On the other hand, the extreme values of the variations of water spectra are 1.5 units at 6000 cm and 2.5 units at 7250 cm. This result suggests that for accurate lower glucose concentration prediction, stability needs to be improved. 6
100 90 80 Relative Magnitude 70 60 50 40 30 20 10 0 4000 4500 5000 5500 6000 6500 7000 7500 8000 8500 cm Figure3: Water-glucose solution spectrum (blue: pure water ; magenta: 100 mmol / L glucose solution) 30 20 10 0-10 -20-30 4000 4500 5000 5500 6000 6500 7000 7500 8000 8500 Figure4: Water-glucose differential spectra (red: 50 mmol / L ; magenta: 100 mmol / L ; yellow: 200 mmol / L ; black: 500 mmol / L ) cm V. Research Plan 7
The experimental result shows that simple algorithm utilizing the two relevant bands might be sufficient for glucose concentration prediction. However, repeatability of the measurements needs to be improved before low concentration prediction could be achieved. Our short-term plan includes improvement of measurement repeatability. This involves identification of the source of error, which may be due to instrumental factor or the sample element. When this is accomplished, careful analysis of achievable sensitivity for waterglucose and blood-glucose measurement will be attempted. This should be investigated through both experimentation and theoretical studies. Additional problems with respect to the noninvasive application are signal loss due to light scattering in tissue and physiological variations such as skin temperature changes. Both degrade the signal-to-noise ratio. In this research, we aim to understand the origins of the signals and the sources of errors. This would enable appropriate modifications to the instrumentation design and/or procedural scheme to improve the accuracy of noninvasive blood glucose measurement. VI. References 1. Marbach R. et. al., Noninvasive Blood Glucose Assay by Near Infrared Diffuse Reflectance Spectroscopy of the Human Inner Lip, Applied Spectroscopy 47 (7): 875-881, 1993 2. Robinson M.R. et. al., Noninvasive Glucose Monitoring in Diabetic Patients: A Preliminary Evaluation, Clin. Chem. 38 (9): 1618-1622, 1992 3. Cote G.L. et. Al., Noninvasive Glucose Sensing Utilizing a Digital Close-Loop Polarimetric Approach IEEE Trans. On Biomed. Eng. 44(12): 1221-1227, 1997 4. Feld M.S. et. Al., Rapid, Noninvasive Concentration Measurements of Aqueous Biological Analytes by Near-Infrared Raman Spectroscopy, Applied Optics 35(1): 209-212, 1996 5. The Home Automation and Healthcare Consortium Progress Report 2-2 6. Griffiths P.R., de Haseth J.A., Fourier Transform Infrared Spectrometry, John Wiley & Sons, 1986 7. Mattson D.R., Sensitivity of A Fourier Transform Infrared Spectrometer, Applied Spectroscopy 32(4): 335-338, 1978 8. Muller U.A. et. al., Non-invasive Blood Glucose Monitoring by Means of Near Infrared Spectroscopy: Methods for Improving the Reliability of the Calibration Models, Artif. Pancreas and Related Tech. In Diabetes and Endocrinology 20 (5): 285-290, 1997 8