Method and system for non-invasive optical blood glucose detection utilizing spectral data analysis

Abstract

Systems and methods are disclosed for non-invasively measuring blood glucose levels in a biological sample based on spectral data. This includes utilizing at least one light source configured to strike a target area of a sample, utilizing at least one light filter positioned to receive light transmitted through the target area of the sample from the at least one light source, utilizing at least one light detector positioned to receive light from the at least one light source and filtered by the at least one light filter, and to generate an output signal, having a time dependent current, which is indicative of the power of light detected, receiving the output signal from the at least one light detector with a processor, calculating the attenuance attributable to blood with a ratio factor based on the received output signal, and determining a blood glucose level based on the calculated attenuance.

Claims

1. A system for detecting glucose in a biological sample, comprising: a light detector that receives light and generates an output photocurrent signal indicative of the power of light received; a processor programmed to calculate a change in a light absorption caused by blood in the biological sample and configured to receive the output photocurrent signal from the at least one photocurrent signal generating light detector and based on the received output photocurrent signal, calculate the attenuance attributable to blood in a sample present in a target area with a ratio factor, and based on the calculated attenuance, determine a blood glucose level associated with a sample present in the target area, wherein the processor is configured to calculate the ratio factor Y.sub.ij(C,T) at a plurality of wavelengths, the i.sup.th wavelength being represented by λ.sub.i the j.sup.th wavelength being represented by λ.sub.j, C is a blood glucose concentration, T is temperature of the biological sample, I.sub.D (λ.sub.i,t) is the time dependent output current, σ[log I.sub.D (λ.sub.i,t)] is standard deviation of the logarithm of the time dependent output current, and t is time, according to the equation: $Y_{\mathrm{ij}}\left(C,T\right)=\frac{\sigma \left[\log I_D\left({\lambda }_i,t\right)\right]}{\sigma \left[\log I_D\left({\lambda }_j,t\right)\right]}.$

2. The system for detecting glucose in a biological sample according to claim 1, wherein the processor is configured to calculate the blood glucose level C.sub.optical based on the ratio factor, Y.sub.ij(C,T) and f.sub.i(T), which is a temperature dependent regression coefficient corresponding to wavelength λ.sub.i according to the equation: $C_{\mathrm{optical}}=\stackrel{N}{\underset{i\ne j}{.Math.}}{f}_i\left(T\right)Y_{ij}\left(C,T\right).$

3. The system for detecting glucose in a biological sample according to claim 2, wherein the processor is configured to extract the values of the temperature dependent regression coefficient corresponding to wavelength λ.sub.i, which is f.sub.i(T), using partial least squares regression.

4. The system for detecting glucose in a biological sample according to claim 1, wherein the processor is configured to calculate the blood glucose level C.sub.optical based on a temperature independent regression coefficient h.sub.i, a regression coefficient β for the temperature of the biological sample T, the ratio factor Y.sub.ij(C,T) according to the equation: $C_{\mathrm{optical}}=\stackrel{N}{\underset{i\ne j}{.Math.}}{h}_i{Y}_{ij}\left(C,T\right)+\beta T.$

5. The system for detecting glucose in a biological sample according to claim 4, wherein the processor is configured to extract the values of the temperature independent regression coefficient h, and the regression coefficient β for the temperature of the biological sample T using partial least squares regression.

6. A method for detecting glucose in a biological sample, comprising: receiving light using at least one photocurrent signal generating light detector and generating an output photocurrent signal by the at least one photocurrent signal generating light detector as an indicium of the power of light received; receiving the output photocurrent signal from the at least one photocurrent signal generating light detector with a processor programmed to calculate a change in a light absorption caused by blood in the biological sample; calculating the attenuance attributable to blood in the biological sample present in the target area with a ratio factor based on the received output photocurrent signal with the processor; calculating the ratio factor Y.sub.ij(C,T) at a plurality of wavelengths with the processor, where the i.sup.th wavelength being represented by λ.sub.i, the j.sup.th wavelength being represented by λ.sub.j, C is a blood glucose concentration, Tis temperature of the biological sample, I.sub.D (λ.sub.i,t) is the time dependent output current, σ[log I.sub.D (λ.sub.i, t)] is standard deviation of the logarithm of the time dependent output current, and t is time, according to the equation: $Y_{\mathrm{ij}}\left(C,T\right)=\frac{\sigma \left[\log I_D\left({\lambda }_i,t\right)\right]}{\sigma \left[\log I_D\left({\lambda }_j,t\right)\right]};$ and determining a blood glucose level associated with the biological sample present in the target area based on the calculated attenuance with the processor.

7. The method for detecting glucose in a biological sample according to claim 6, further comprising calculating the blood glucose level C.sub.optical with the processor based on the ratio factor, Y.sub.ij(C, T) and f.sub.i(T), which is the temperature dependent regression coefficient corresponding to wavelength λ.sub.i, according to the equation: $C_{\mathrm{optical}}=\stackrel{N}{\underset{i\ne j}{.Math.}}{f}_i\left(T\right)Y_{ij}\left(C,T\right).$

8. The method for detecting glucose in a biological sample according to claim 7, further comprising extracting the values f.sub.i(T) of the temperature dependent regression coefficient corresponding to wavelength λ.sub.i, utilizing partial least squares regression with the processor.

9. The method for detecting glucose in a biological sample according to claim 6, further comprising calculating blood glucose level C.sub.optical with the processor based on a temperature-independent regression coefficient h.sub.i, a regression coefficient β for the temperature of the biological sample T, and the ratio factor Y.sub.ij(C,T) according to the equation: $C_{\mathrm{optical}}=\stackrel{N}{\underset{i\ne j}{.Math.}}{h}_i{Y}_{ij}\left(C,T\right)+\beta T.$

10. The method for detecting glucose in a biological sample according to claim 6, further comprising extracting values of the temperature-independent regression coefficient h.sub.i and the regression coefficient β for the temperature of the biological sample T utilizing partial least squares regression with the processor.

11. The system for detecting glucose in a biological sample according to claim 1, further comprising at least one light source.

12. The system for detecting glucose in a biological sample according to claim 1, further comprising at least one light filter positioned to receive the light transmitted through the target area of the biological sample from an at least one light source.

13. The system for detecting glucose in a biological sample according to claim 1, wherein the light detector positioned to receive light from an at least one light source and filtered by an at least one light filter, and to generate the output photocurrent signal, having a time dependent current, which is indicative of the power of light detected.

14. The method for detecting glucose in a biological sample according to claim 6, further comprising utilizing one or more light beams to strike a target area of the biological sample.

15. The method for detecting glucose in a biological sample according to claim 6, further comprising utilizing at least one light filter positioned to receive light transmitted through the target area of the biological sample from an at least one light source.

16. The method for detecting glucose in a biological sample according to claim 6, further comprising utilizing the at least one photocurrent signal generating light detector positioned to receive the light from an at least one light source and filtered by an at least one light filter, and to generate the output photocurrent signal, having a time dependent current, which is indicative of the power of light detected.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) For a better understanding of the present invention, reference may be made to accompanying drawings, in which:

(2) FIG. 1 illustrates a plot of a pulse wave corresponding to light absorption of arterial blood, according to exemplary embodiments;

(3) FIG. 2 illustrates an exemplary system for obtaining spectral data;

(4) FIG. 3 illustrates a plot of A(t), calculated according to Equation (9) using data in FIG. 1; and

(5) FIG. 4 is a basic illustrative schematic of a preamplifier circuit that converts photocurrent into voltage prior to digitization.

DETAILED DESCRIPTION OF THE INVENTION

(6) In the following detailed description, numerous exemplary specific details are set forth in order to provide a thorough understanding of the invention. However, it will be understood by those skilled in the art that the present invention may be practiced without these specific details, or with various modifications of the details. In other instances, well known methods, procedures, and components have not been described in detail so as not to obscure the present invention.

(7) Optical spectroscopy can be used to determine the amount of light absorbed and scattered, i.e., attenuated, by a biological sample such as a human finger. By measuring the amount of light absorbed by the sample, it is possible to determine glucose, cholesterol, and hemoglobin levels of a subject non-invasively. Fingertip measurements are usually preferred because of the large concentration of capillaries in the fingertip and because of the conversion of arterial blood into venous blood that occurs in the fingertip. However, the techniques of the present invention are not limited to use with a fingertip. For example, the biological sample could be a human earlobe.

(8) When light is transmitted through a biological sample, such as a human finger, the light is attenuated by various components of the finger including skin, muscle, bone, fat, interstitial fluid and blood. It has been observed, however, that light attenuation by a human finger exhibits a small cyclic pattern that corresponds to a heartbeat. It is believed that this cyclic pattern will be present in measurements of many other human body parts, the earlobe being one of many examples.

(9) FIG. 1 depicts a plot 102 of a detector photocurrent, I.sub.D(t), that corresponds to the power of light received by a detector after the light has passed through a subject's finger. As can be seen, the detector photocurrent exhibits a cyclic pattern. This cyclic pattern is due to the subject's heartbeat, which cyclically increases and decreases the quantity of blood in the subject's capillaries (or other structures). Although the magnitude of the cyclic pattern is small in comparison to the total photocurrent generated by the detector, considerable information can be extracted from the cyclic pattern of the plot 102. For example, assuming that the person's heart rate is sixty beats per minute, the time between the start of any pulse beat and the end of that pulse beat is one second. During this one-second period, the photocurrent will have a maximum or peak reading 104 and minimum or valley reading 106. The peak reading 104 of the plot corresponds to when there is a minimum amount of blood in the capillaries, and the valley reading 106 corresponds to when there is a maximum amount of blood in the capillaries. By using information provided by the peak and valley of the cyclic plot, the optical absorption and scattering by major finger constituents that are not in the capillaries, such as skin, fat, bones, muscle and interstitial fluids, are excluded. These major constituents that are not in the capillaries are excluded because they are not likely to change during the time interval of one heartbeat. In other words, the light that is absorbed and scattered, i.e., attenuated, by the blood can be detected based on the peaks and valleys of the plot 102.

(10) Assuming that the peak of the cyclic photocurrent generated by the light-sensing device is I.sub.P, the adjacent valley of the cyclic photocurrent is I.sub.V, and the photocurrent generated by the light-sensing device without a human finger is I.sub.0, the transmittances corresponding to the peak and valley photocurrents can be defined as:

(11) $\begin{array}{cc}{T}_V=\frac{I_V}{I_0}\mathrm{and}& \left(1\right)\\ T_P=\frac{I_P}{I_0}& \left(2\right)\end{array}$

(12) The corresponding peak and valley absorbance are:
A.sub.V=−log(T.sub.V)  (3)
and
A.sub.P=−log(T.sub.P)  (4)

(13) The difference between A.sub.V and A.sub.P represents the light absorption and scattering by the blood in the finger, excluding non-blood constituents:

(14) $\begin{array}{cc}\Delta A=A_V-A_P=\log \left(\frac{I_P}{I_V}\right)& \left(5\right)\end{array}$

(15) As can be seen in the algorithm shown in Equation (5), ΔA does not depend on I.sub.0. Thus, calculating ΔA does not require a determination of the current generated by the light-sensing device without a sample. Monitoring the photocurrent corresponding to light power transmitted through a sample is sufficient to calculate ΔA.

(16) FIG. 2 depicts a simplified block diagram of an exemplary apparatus for use in an exemplary embodiment. Optical measurement system, which is generally indicated by numeral 200, uses the “pulsatile” concept for determining an amount of light absorbed and scattered solely by the blood in a sample (a human finger in this exemplary embodiment). A power source 201, such as a battery, provides power to a light source 202 that generates a plurality of light beams 204, 206, 208, 210 that are directed toward the top of the finger of a subject. In an exemplary embodiment, each of the light beams 204, 206, 208, 210 have the same wavelength or a different wavelength range, typically within 800 nm to 1600 nm. Although the optical measurement system 200 is described herein as generating four (4) light beams, it is contemplated that the light source 202 can be altered to generate fewer light beams or additional light beams in other embodiments.

(17) A first aperture 212 ensures that the light beams 204, 206, 208, 210 strike a target area of the finger. A second aperture 214 ensures that the portion of the light beams that are transmitted through the finger strike a lens 216. Light beams 204, 206, 208, 210 are attenuated by the finger and components of the optical measurement system 200, and, thus, attenuated light beams 218, 220, 222, 224 are emitted from the finger. The attenuated light beams 218, 220, 222, 224 strike the lens 216, and the lens 216 collects the attenuated light beams 218, 220, 222, 224 so that they impinge more efficiently on a detector block 226.

(18) The detector block 226 is positioned directly under the lens 216 and comprises a plurality of light-sensing devices (LSD) 228, 230, 232, 234 such as an array of photodiodes. According to one aspect of the optical measurement system 200, each of the light-sensing devices 228, 230, 232, 234 detects a specific wavelength of light as defined by corresponding interference filters (IF) 236, 238, 240, 242, respectively. The interference filter transmits one or more spectral bands or lines of light, and blocks others.

(19) Each of the light-sensing devices 228, 230, 232, 234 generates a corresponding photocurrent signal that is proportional to the power of the light received by the particular light sensing device. The photocurrent signal generated by the photodiode can be converted to another form of signal, such as an analog voltage signal or a digital signal. A processor 243 is coupled to the detector block 226 and is configured to calculate the change of photocurrent signals 244, 246, 248, 250.

(20) According to one aspect, the processor 243 executes an algorithm such as shown in the Equation (5) to calculate the change in the light absorption (ΔA) solely caused by the blood in the finger. Thereafter, this quantitative calculation of light absorption of the blood can be used to determine a characteristic of the blood. For example, by comparing the calculated light absorption value to predetermined values corresponding to different glucose levels stored in a memory (not shown), a blood-glucose level of the subject can be determined.

(21) A difficulty associated with the finger based pulsatile detection methodology is low signal-to-noise (S/N) ratio, because the amplitude of cyclic pattern (i.e., the difference between peak and valley) is typically 1%-2% of the total photocurrent generated by the light power transmitted through the finger. To obtain a S/N ratio of 100:1 in the determination of ΔA, the baseline noise of the device being used to measure the light absorption by the finger should not be larger than 3.0×10.sup.−5 in absorbance (peak to peak), within a 10 Hz bandwidth.

(22) However, a 3.0×10.sup.−5 absorbance (peak to peak) baseline noise level within a 10 Hz bandwidth is difficult to obtain with the low light power levels that are used by some battery-powered hand held non-invasive blood chemical measurement devices. One solution involves data averaging. To increase the S/N ratio, the averaged value of ΔA, as defined by the Equation below, is used in further calculation to extract blood glucose concentration:

(23) $\begin{array}{cc}\stackrel{\_}{\Delta A}=\underset{j=1}{\overset{M}{.Math.}}\Delta A_j& \left(6\right)\end{array}$

(24) In Equation (6), M is the number of heartbeats during the time interval of the pulsatile measurement. However, this approach requires long data acquisition time, due to the fact that the rate of heartbeat is in the order of one per second. For example, 25 seconds would be needed for increasing the S/N ratio by a factor of five, and 100 seconds would be needed for increasing the S/N ratio by a factor of ten. In comparison, current commercial blood drawing glucose meters can determine blood glucose level within 5 seconds. Furthermore, long detection time will significantly increase measurement errors due to finger movement, light power drift, device temperature change, etc. Thus, there is a need for new techniques to measure blood glucose levels quickly and accurately.

(25) Improving S/N Ratio by Standard Deviation

(26) The time dependent detector photocurrent output, I.sub.D(t), shown in FIG. 1 can be expressed as the sum of a small time dependent cyclic photocurrent ΔI(t), corresponding to the heartbeat, a noise current n(t), and a constant baseline photocurrent I.sub.B:
I.sub.D(t)=I.sub.B±ΔI(t)+n(t)  (7)

(27) The above Equation can be re-arranged as:

(28) $\begin{array}{cc}\frac{I_D\left(t\right)}{I_B}=1+\frac{\Delta I\left(t\right)+n\left(t\right)}{I_B}& \left(8\right)\end{array}$

(29) Applying common logarithm to both side of the Equation (8), one obtains:

(30) $\begin{array}{cc}A\left(t\right)=\log \left[\frac{I_D\left(t\right)}{I_B}\right]=\log \left(1+\frac{\Delta I\left(t\right)+n\left(t\right)}{I_B}\right)& \left(9\right)\end{array}$

(31) FIG. 3, which is generally indicated by numeral 300, shows a typical A(t) plot 302, calculated according to Equation (9) using data in FIG. 1. For a pulse function A(t) shown in FIG. 3, the following key relationship exists during the time interval of one heartbeat:
σ[A(t)]=kΔA  (10)
in which σ[A(t)] is the Standard Deviation of A(t), and k is a proportional constant.

(32) Considering the fact that I.sub.B is a constant and σ.sup.2(log I.sub.B)=0, one obtains:
σ[A(t)]=σ[ log I.sub.D(t)]  (12)

(33) Therefore, the peak-to-valley height of the A(t) plot during the time interval of one heartbeat can be obtained directly from the standard deviation of the logarithm of I.sub.D(t):

(34) $\begin{array}{cc}\Delta A=\frac{\sigma \left[A\left(t\right)\right]}{k}=\frac{\sigma \left[\log I_D\left(t\right)\right]}{k}& \left(13\right)\end{array}$
A major advantage of Equation (13) is that high S/N ratio can be achieved within short data acquisition time (approximately one second), as explained below.

(35) In a finger based pulsatile measurement depicted by FIG. 2, the value of the sum, ΔI(t)+n(t) is typically less than 2% of the large constant baseline photocurrent I.sub.B. Therefore, Equation (9) can be approximated as:

(36) $\begin{array}{cc}A\left(t\right)=\log \left[\frac{I_D\left(t\right)}{I_B}\right]\approx \frac{1}{\ln 10}\frac{\Delta I\left(t\right)+n\left(t\right)}{I_B}& \left(14\right)\end{array}$

(37) Similarly, the standard deviation of A(t) can be approximated as:

(38) $\begin{array}{cc}\sigma \left[A\left(t\right)\right]\approx \frac{1}{\ln 10}\frac{\sqrt{{\sigma }^2\left[\Delta I\left(t\right)\right]+{\sigma }^2\left[n\left(t\right)\right]}}{I_B}& \left(15\right)\end{array}$

(39) Equation (15) demonstrates great noise reduction power of Equation (13). For example, for a relatively high baseline noise with the ratio

(40) $\rho =\frac{\sigma \left[n\left(t\right)\right]}{\sigma \left[\log \Delta I\left(t\right)\right]}=0.1\left(\mathrm{or}10\%\right),$
the contribution to σ[A(t)] from the baseline noise n(t) is estimated to be less than 0.005 (or 0.5%), corresponding to an increase in S/N ratio by a factor of 20 without increasing detection time. As such, dramatic noise reduction can be obtained without increasing the data acquisition time, and a finger based pulsatile measurement can be completed within the time interval of one heartbeat (which is approximately one second), and the requirement for the S/N ratio of 100 to 1 in determination of ΔA can be satisfied using an optical system with a baseline noise of about 6.0×10.sup.−4 absorbance (peak to peak) within a 10 Hz bandwidth. It should be pointed out that when the baseline noise of an optical system is dominated by shot noise due to low light illumination power, a noise reduction by a factor of 20 equals an increasing in light illumination power by a factor of 20.sup.2=400.

(41) This ability of obtaining higher S/N ratio within the very short data acquisition time, e.g., less than one second, will significantly reduce detection error caused by factors such as finger movement, temperature change, and light power drift during the measurement, and therefore dramatically improve the accuracy and reproducibility of the pulsatile detection methodology.

(42) Furthermore, the value of k does not change with wavelength, because transmitted lights at all wavelengths have identical pulse shape due to the heartbeat. As a result, the constant k will be cancelled in data normalization discussed in next section, and σ[ log I.sub.D (t)] will be used in further regression analysis to establish correlation between the optical measurement and blood glucose level. This will greatly simplify the data analysis process since σ[ log I.sub.D(t)] involves only two standard math functions available in most popular spreadsheet programs such as Microsoft EXCEL®. EXCEL® is a federally registered trademark of Microsoft Corporation, having a place of business at One Microsoft Way, Redmond, Wash. 98052-6399.

(43) Normalization

(44) At each wavelength λ.sub.i, the absorption ΔA(λ.sub.i) is linked to the increase of amount of blood (ΔB) in the optical sensing area of the fingertip due to the heartbeat by the following Equation:
ΔA(λ.sub.i)=ε(C,λ.sub.i,T)ΔB  (16)
in which ε(C, λ.sub.i, T) is the absorption/scattering coefficient of blood at wavelength λ.sub.i, finger temperature T, and blood glucose concentration C. It is well understood that the variable ΔB differs from person to person, and may even change from day to day for the same person.

(45) The uncertainty from the variable ΔB can be cancelled by introducing the normalization factor Q.sub.i(C,T) at each wavelength λ.sub.i, as defined by the Equation below:

(46) 0$\begin{array}{cc}{Q}_i\left(C,T\right)=\frac{\Delta A\left({\lambda }_i\right)}{\underset{i=1}{\overset{N}{.Math.}}\Delta A\left({\lambda }_i\right)}=\frac{\epsilon \left(C,{\lambda }_i,T\right)}{\underset{i=1}{\overset{N}{.Math.}}\epsilon \left(C,{\lambda }_i,T\right)},& \left(17\right)\end{array}$
in which N is total number of wavelength employed. Preferably, N typically ranges from twenty to thirty.

(47) Based on Equations (13) and (17), Q.sub.i(C,T) is linked to the detector photocurrent at each wavelength λ.sub.i, I.sub.D(λ.sub.i,t), by the following Equation:

(48) $\begin{array}{cc}{Q}_i\left(C,T\right)=\frac{\Delta A\left({\lambda }_i\right)}{\underset{i=1}{\overset{N}{.Math.}}\Delta A\left({\lambda }_i\right)}=\frac{\sigma \left[\log I_D\left({\lambda }_i,t\right)\right]/k}{\underset{i=1}{\overset{N}{.Math.}}\sigma \left[\log I_D\left({\lambda }_i,t\right)\right]/k}=\frac{\sigma \left[\log I_D\left({\lambda }_i,t\right)\right]}{\underset{i=1}{\overset{N}{.Math.}}\sigma \left[\log I_D\left({\lambda }_i,t\right)\right]},& \left(18\right)\end{array}$

(49) As shown by Equation (18), the constant k is cancelled and σ[ log I.sub.D(t)] will be used in further regression analysis to establish correlation between the optical measurement and blood glucose level. This is possible because data are taken simultaneously from all detection channels.

(50) A correlation between optical measurement and blood glucose concentration can be established according to the following Equation:

(51) $\begin{array}{cc}{C}_{\mathrm{optical}}=\underset{i=1}{\overset{N}{.Math.}}{a}_i\left(T\right)Q_i\left(C,T\right)& \left(19\right)\end{array}$
in which C.sub.optical is the blood glucose concentration predicted by the optical measurement, Q.sub.i(C,T) is defined by Equations (17) and (18), and a.sub.i(T) is the temperature dependent regression coefficient corresponding to wavelength λ.sub.i. The values of a.sub.i(T) can be extracted using proper statistics methods such as Partial Least Squares (PLS) regression.

(52) Equation (19) represents ideal cases when large number of calibrations can be made at different finger temperatures. In reality, frequently only a limited number of calibrations can be made (e.g., 15 to 20), and each may be taken at a different finger temperature. Under this condition, the finger temperature can be treated as an independent variable, and the above Equation can be approximated as:

(53) $\begin{array}{cc}{C}_{\mathrm{optical}}=\underset{i=1}{\overset{N}{.Math.}}{b}_i{Q}_i\left(C,T\right)+\eta T& \left(20\right)\end{array}$
in which b.sub.1 is the temperature independent regression coefficient corresponding to wavelength λ.sub.i, and η is the regression coefficient for the finger temperature. The values of b.sub.i and that of η can be extracted using proper statistics methods such as Partial Least Squares (PLS) regression.
Ratio Methodology

(54) Alternatively, the uncertainty from the variable ΔB can be cancelled by introducing a ratio factor Y.sub.ij at wavelength λ.sub.i:

(55) $\begin{array}{cc}{Y}_{\mathrm{ij}}\left(C,T\right)=\frac{\Delta A\left({\lambda }_i\right)}{\Delta A\left({\lambda }_j\right)}=\frac{\epsilon \left(C,{\lambda }_i,T\right)}{\epsilon \left(C,{\lambda }_j,T\right)}=\frac{\sigma \left[\log I_D\left({\lambda }_i,t\right)\right]}{\sigma \left[\log I_D\left({\lambda }_j,t\right)\right]},& \left(21\right)\end{array}$
in which j can be any number from 1 to N, assuming that the device collects signal at all N wavelengths.

(56) Similar to the normalization algorithm discussed before, a correlation between optical measurement and blood glucose level can be established according to the following Equation:

(57) $\begin{array}{cc}{C}_{\mathrm{optical}}=\underset{i\ne j}{\overset{N}{.Math.}}{f}_i\left(T\right)Y_{\mathrm{ij}}\left(C,T\right)& \left(22\right)\end{array}$
in which C.sub.optical is the blood glucose concentration predicted by the optical measurement, Y.sub.ij(C,T) is defined by Equation (21), and f.sub.i(T) is the temperature dependent regression coefficient corresponding to wavelength λ.sub.i. The value of f.sub.i(T) can be obtained using statistics methods such as Partial Least Squares (PLS) regression.

(58) Equation (22) represents ideal cases when large number of calibration can be made at different finger temperatures. In reality, frequently only limited number of calibration can be made (e.g., 15 to 20), and each may be taken at a different finger temperature. Under this condition, the finger temperature can be treated as an independent variable, and the above Equation can be approximated as:

(59) $\begin{array}{cc}{C}_{\mathrm{optical}}=\underset{i\ne j}{\overset{N}{.Math.}}{h}_i{Y}_{\mathrm{ij}}\left(C,T\right)+\beta T& \left(23\right)\end{array}$
in which h.sub.i is the temperature independent regression coefficient corresponding to wavelength λ.sub.i, and β is the regression coefficient for the finger temperature. The values of h.sub.i and that of β can be extracted using proper statistics methods such as Partial Least Squares (PLS) regression.
Elimination of the Effect of Temperature Dependent Device Response

(60) It is well understood that the detector sensitivity of a silicon photodiode detector is a function of wavelength and temperature. For the device configuration shown in FIG. 2, which is generally indicated by numeral 200, the light power received by ith silicon diode detector, corresponding to wavelength λ.sub.i, is converted into a photocurrent according to the following Equation:
I.sub.D(λ.sub.i,t)=P(λ.sub.i,t)S.sub.0(λ.sub.i)[1+γ(λ.sub.i)(T.sub.Di(t)−25° C.)]  (24)

(61) In the above Equation (24), P(λ.sub.i,t) is the light power received by the detector, S.sub.0(λ.sub.i) is the photosensitivity of the detector at wavelength λ.sub.i and 25° C., γ(λ.sub.i) is the temperature coefficient of the photosensitivity at wavelength λ.sub.i, and T.sub.Di(t) is the temperature of ith photodiode detector. The temperature coefficient γ(λ.sub.i) varies with the wavelength. For example, for Hamamatsu S1337 series photodiode detectors, γ(λ.sub.i) ranges from near zero at 900 nm to over 1.0%/° C. at 1100 nm. This imposes a potential problem for the device configuration show in FIG. 2, because it is very difficult to keep temperature of each individual diode detector constant in a handheld device used by a person with diabetes under a normal household/office environment.

(62) This uncertainty due to the detector temperature T.sub.Di(t) can be eliminated using the algorithm shown by Equations (12) and (13). Applying common logarithm on both sides of the Equation (24), one obtains:
log I.sub.D(λ.sub.i,t)=log P(λ.sub.i,t)+log S.sub.0(λ.sub.i)+log [1+γ(λ.sub.i)(T.sub.Di(t)−25° C.)]  (25)

(63) Considering the fact that S.sub.0(λ.sub.i) is a constant and that detector temperature T.sub.Di(t) remains almost constant during the very short data acquisition time interval of approximately one second, one obtains:
σ[ log I.sub.D(λ.sub.i,t)]=σ[ log P(λ.sub.i,t)]  (26)
As such, the uncertainty caused by detector temperature T.sub.Di(t) is eliminated by the use of this standard deviation methodology.
Voltage Detection Mode

(64) In the device configuration shown in FIG. 2, the photocurrent of ith photodiode detector I.sub.D(λ.sub.i,t) is typically converted into a voltage using a preamplifier before digitization. FIG. 4 shows the schematic circuit diagram of a typical preamplifier, which is generally indicated by numeral 400.

(65) The output voltage 412 of ith preamplifier 400, in coupling with ith photodiode detector 408, can be expressed as:
V.sub.i(t)=R.sub.iI.sub.D(λ.sub.i,t)=R.sub.0i[1+χ.sub.i(T.sub.Ri(t)−25° C.)]I.sub.D(λ.sub.i,t)  (27)

(66) In the above Equation (27), R.sub.0i is the resistance value of feedback resistor 402 for ith preamplifier at 25° C., χ.sub.i is the temperature coefficient of the resistor, and T.sub.Ri(t) is the temperature of the resistor. Applying common logarithm to both side of the Equation (27), one obtains:
log V.sub.i(t)=log R.sub.i0+log [1+χ.sub.i(T.sub.Ri(t)−25° C.)]+log I.sub.D(λ.sub.i,t)  (28)

(67) Considering the fact that R.sub.0i is a constant and that the resistor temperature T.sub.Ri(t) does not change during the very short data acquisition time interval of approximately one second, one obtains:
σ[ log V.sub.i(t)]=σ[ log I.sub.D(λ.sub.i,t)]  (29)

(68) Substituting Equation (26) into Equation (29), one obtains:
σ[ log V.sub.i(t)]=σ[ log P(λ.sub.i,t)]  (30)
As such, the uncertainty caused by resistor temperature T.sub.R (t) is eliminated.

(69) Under the voltage detection mode, the normalization factor in Equation (18) can be expressed as:

(70) $\begin{array}{cc}{Q}_i\left(C,T\right)=\frac{\sigma \left[\log V_i\left(t\right)\right]}{\underset{i=1}{\overset{N}{.Math.}}\sigma \left[\log V_i\left(t\right)\right]}& \left(31\right)\end{array}$

(71) The mathematic correlation between optical measurement and blood glucose concentration can then be established according to Equation (19) or Equation (20), under corresponding calibration conditions.

(72) Similarly, the ratio factor defined by Equation (21) can be expressed as:

(73) $\begin{array}{cc}{Y}_{\mathrm{ij}}\left(C,T\right)=\frac{\sigma \left[\log V_i\left(t\right)\right]}{\sigma \left[\log V_j\left(t\right)\right]}& \left(32\right)\end{array}$

(74) The mathematic correlation between optical measurement and blood glucose concentration can then be established according to Equation (22) or Equation (23), under corresponding calibration conditions. The schematic circuit diagram of a typical preamplifier 400 also includes a feedback capacitor 404, an operational amplifier 406, and a ground connection 410.

(75) Digitization

(76) The voltage output 412 from the preamplifier 400 is usually digitized using an analog-to-digital convertor (ADC). The digitized signal is then sent to a computer for data analysis. The output of ith ADC, in communication with ith preamplifier that is in coupling with ith photodiode 408 collecting light power at wavelength λ.sub.i, can be expressed by the following Equation:
(ADC).sub.i=(ADC).sub.0i+G.sub.i{└I.sub.D(λ.sub.i,t)+I.sub.Dark,i┘R.sub.i+A.sub.0i}  (33)

(77) In the above Equation (33), (ADC).sub.0i is the offset of ith ADC, G.sub.i is the nominal ADC Gain used during the detection, I.sub.D(λ.sub.i,t) is the photocurrent of ith photodiode detector, I.sub.Dark,i is the dark current of ith photodiode detector, R.sub.i=R.sub.0i[1+χ.sub.i(T.sub.Ri(t)−25° C.)]] is the resistance of feedback resistor of ith preamplifier, and A.sub.0i is the offset of ith preamplifier.

(78) The contribution of the three factors, (ADC).sub.0i, I.sub.Dark,i and A.sub.0i can be removed by carrying out a dark measurement with the light source turned off right before or after the corresponding finger measurement. When the light source is turned off, the above Equation (33) becomes:
(ADC).sub.Dark,i=(ADC).sub.0i+G.sub.i(I.sub.Dark,iR.sub.i+A.sub.01)  (34)

(79) The difference between the two above Equations (33) and (34) reflects ADC output corresponding to the photocurrent:
Δ(ADC).sub.i=(ADC).sub.i−(ADC).sub.Dark,i=G.sub.iI.sub.D(λ.sub.i,t)R.sub.i  (35)

(80) Applying common logarithm to both side of the Equation (35), one obtains:
log Δ(ADC).sub.i=log G.sub.i+log I.sub.D(λ.sub.i,t)+log R.sub.i  (36)

(81) G.sub.i and R.sub.i can be considered as constants as long as the time interval between the finger measurement and the dark measurement is short. As such, one obtains:
σ[ log Δ(ADC).sub.i]=σ[ log I.sub.D(λ.sub.i,t)]  (37)
Substituting Equation (26) into Equation (37), one further obtains:
σ[ log Δ(ADC).sub.i]=σ[ log P(λ.sub.i,t)]  (38)

(82) Based on Equation (37), the normalization factor defined by Equation (18) can be expressed as:

(83) $\begin{array}{cc}{Q}_i\left(C,T\right)=\frac{\sigma \left[\log {\Delta \left(\mathrm{ADC}\right)}_i\right]}{\underset{i=1}{\overset{N}{.Math.}}\sigma \left[\log {\Delta \left(\mathrm{ADC}\right)}_i\right]}& \left(39\right)\end{array}$

(84) The mathematic correlation between optical measurement and blood glucose concentration can then be established according to Equation (19) or (20), under corresponding calibration conditions.

(85) Similar to normalization, the ratio factor defined by Equation (21) can be expressed as:

(86) 0$\begin{array}{cc}{Y}_{\mathrm{ij}}\left(C,T\right)=\frac{\sigma \left[\log {\Delta \left(\mathrm{ADC}\right)}_i\right]}{\sigma \left[\log {\Delta \left(\mathrm{ADC}\right)}_j\right]}& \left(40\right)\end{array}$

(87) The correlation between optical measurement and blood glucose concentration can then be established according to Equations (22) or (23), under corresponding calibration conditions.

(88) Thus, there has been shown and described several embodiments of a novel invention. As is evident from the foregoing description, certain aspects of the present invention are not limited by the particular details of the examples illustrated herein, and it is therefore contemplated that other modifications and applications, or equivalents thereof, will occur to those skilled in the art. The terms “have,” “having,” “includes,” “including,” and similar terms as used in the foregoing specification are used in the sense of “optional” or “may include” and not as “required.” Many changes, modifications, variations and other uses and applications of the present construction will, however, become apparent to those skilled in the art after considering the specification and the accompanying drawings. All such changes, modifications, variations and other uses and applications, which do not depart from the spirit and scope of the invention, are deemed to be covered by the invention, which is limited only by the claims that follow. It should be understood that the embodiments disclosed herein include any and all combinations of features described in any of the dependent claims.