CN112587134A - Noninvasive blood glucose detection method - Google Patents

Abstract

The invention relates to a noninvasive blood sugar detection method, in particular to a noninvasive blood sugar detection method, which breaks through the noninvasive blood sugar detection technology by using a spectrum sensor, adopts the improved near infrared spectrum transmission measurement technology, has an original calibration mechanism, no material consumption, accurate clinical verification data, small influence by the personal state and environmental change of a human body, less error than +/-15 percent and capability of being comparable with the detection equipment with wound; the method comprises the following steps: the method comprises the following steps: a spectrum sensor is arranged on one side of the fingertip position, and an LED light source is designed on the other side relative to the fingertip position; step two: a tunable filter of a Fabry-Perot interferometer is adapted in the spectrum sensor, and the optical receiving range of the tunable filter is adjusted to reach nm level; step three: light rays emitted by the LED with the wavelength of 1530-1850 nm penetrate through human fingertip tissues and are collected by the spectral sensor with the wavelength of 1350-1650 nm.

Description

Noninvasive blood glucose detection method

Technical Field

The invention relates to a non-invasive blood sugar detection method, in particular to a non-invasive blood sugar detection method.

Background

Diabetes mellitus is a common metabolic endocrine disease, is mainly characterized by hyperglycemia, has a remarkably rising trend in recent years, and is currently suffered by about 10 percent of adults all over the world. By 2019, about 4.63 hundred million diabetics are in the population of 20-79 years old all over the world, wherein the number of Chinese diabetics is the first rank, and the total number of Chinese diabetics is about 1.164 hundred million. The current treatment mode of the diabetes patient is mainly to regulate and control the glucose metabolism in the body of the patient, and the dosage is determined according to the glucose content in the blood of the patient in the clinical treatment, so the blood glucose monitoring is very important for tracking and evaluating the control and curative effect of the diabetes.

The most direct and accurate method for detecting the blood sugar value at present is a method for detecting collected venous blood, namely a biochemical analyzer is used for analysis, which is a recognized 'gold standard' for measuring the blood sugar concentration at present, but the method has strong operation expertise and high cost, and brings pain and inconvenience to patients. The second type is a invasive blood glucose meter, which uses a blood sampling pen to prick a finger to collect fingertip blood, drops the blood into test paper, and then puts the test paper into a detection instrument to obtain blood glucose data. The pricking can cause the pain of the testee, and the test paper needs to be replaced every time of measurement; the measurement cost is high, the steps are complex, and the operation is inconvenient in multiple links such as needle insertion, blood sampling and measurement. The third type is a noninvasive glucometer which is always considered as an eosin for blood sugar detection, a large number of noninvasive blood sugar measurement method researches have been carried out for nearly thirty years, but no method is approved by national drug administration for production and application at present, the main reason is that the numerical accuracy and repeatability of detection are not good, the noninvasive glucometer adopts a spectral detection or other photoelectric sensing modes for detection, and because the type selection of a sensor, the interference signals are complex and various, and the difference of various human bodies is too large, no method reaches the practical level so far.

The fundamental reason is still that the voltage and current value detected by the noninvasive glucometer has large fluctuation,

the factors of the fluctuation are the following reasons:

(1) the strength of the collected signal is weak

More than 90% of human blood is water, the proportion of the blood is only 7-8%, and the content of blood sugar in the blood is very low; moreover, water absorbs near infrared light seriously, which causes serious interference to noninvasive detection; in addition, the core of the near infrared spectral band is the frequency doubling and combined frequency absorption of molecules, the absorption peaks are wide and are overlapped seriously, and the absorbance magnitude is different from the fundamental frequency of the mid infrared by orders of magnitude. In view of the above, if the change information of the blood glucose component with a weak content is to be detected accurately, a higher requirement is put forward on the performance of the spectrum acquisition system.

(2) Many interference factors

Human tissues such as skin, muscle, skeleton and the like belong to strong near-infrared absorbers, human spectrums carry a large amount of interference information related to the tissues, and effective information which can be used for analysis is easily submerged in a strong background. Therefore, the tissue background interference problem is one of the important reasons for influencing the accuracy of non-invasive blood glucose detection.

How to extract effective information from the strong background spectrum is a key problem to be solved for noninvasive detection of blood glucose by near infrared spectrum.

(3) Individual difference

The tissue characteristics of blood, skin, muscle and the like are greatly different among different individuals, even the tissue background components of different parts of the same individual are different, which causes the acquired spectrum background noise to be complicated, and further increases the difficulty of extracting the blood component information from the human spectrum.

(4) Volume change of blood flow

The human body belongs to a complex living body, the physiological phenomena of heart pulsation, blood circulation and the like can cause the periodic fluctuation of blood flow volume, and the time-varying characteristic of the blood flow volume can cause the change of absorbance in the near-infrared spectrum of the human body and obviously influence the measurement result, which is mainly represented as the instability of a spectrum time domain.

(5) Too wide wavelength range of photoelectric sensor

The wavelength range of the photoelectric sensor which can be used for detecting the blood sugar wavelength is too wide, the spectral information of the LED with the specific wavelength cannot be accurately received, the wavelength range needs to be cut off by an optical filter, so that the wavelength range is reduced, and the accuracy is affected due to insufficient light passing rate after the optical filter is used.

(6) Too low power and insufficient light transmission of LED

The power of LEDs with specific wavelengths which can be used for detecting blood sugar in the market is too small and is concentrated in 1-3mw, so that the passing rate of light irradiating the fingers is insufficient.

In summary, it can be seen that how to implement non-invasive blood glucose detection by using spectroscopic techniques is a technical problem that has yet to be solved.

Disclosure of Invention

The technical problem to be solved by the invention is to overcome the existing defects and provide a noninvasive blood glucose detection method; the technology for detecting the blood sugar by the spectrum sensor in a non-invasive way adopts the improved near infrared spectrum transmission measurement technology, an original calibration mechanism is adopted, no material consumption is caused, the clinical verification data is accurate, the influence of the personal state and the environmental change of a human body is small, the error is less than +/-15 percent, and the technology can be compared with the detection equipment with wounds.

In order to solve the technical problems, the invention provides the following technical scheme: a non-invasive blood glucose assay method comprising the steps of: the method comprises the following steps: a spectrum sensor is arranged on one side of the fingertip position, and an LED light source is designed on the other side relative to the fingertip position;

step two: a tunable filter of a Fabry-Perot interferometer is adapted in the spectrum sensor, and the optical receiving range of the tunable filter is adjusted to reach nm level;

step three: light rays emitted by an LED with the wavelength of 1530-1850 nm penetrate through human fingertip tissues and are collected by a spectral sensor with the wavelength of 1350-1650 nm;

step four: using blood flow volume spectroscopy subtraction: carrying out differential processing on two spectra with different blood flow volume quantities acquired in a very short time; at two times T1 and T2 with a time interval Δ T, the blood thickness at the same measurement position changes from L1 to L2 with a volume change Δ L, and when a beam of near infrared light irradiates the measurement position, the absorbance characteristics corresponding to the times T1 and T2 can be calculated according to Lambert-be's law:

A1＝lg〔I0/I(t1)〕＝lg(I0/I1)

A2＝lg〔I0/I(t2)〕＝lg(I0/I2)

a blood absorbance spectrum corresponding to the amount of change in optical path is obtained by referring to the spectra measured at two times:

ΔA＝lg〔I2/I1〕＝lg(I0/I1)-lg(I0/I2)＝A1-A2

obtaining a near-infrared volume difference spectrum;

step five: inputting two groups of near infrared spectrum data with different wavelengths as independent variable matrixes of a model, taking blood sugar values collected by a blood sugar instrument as dependent variables of the model, dividing a sample training set and a test set, and testing a plurality of groups of data of a user;

step six: sorting the screened spectral data according to the blood sugar values measured by a blood sugar instrument, taking two groups of spectral data measured by different wavelengths as independent variables, taking the blood sugar values as dependent variables, respectively normalizing, distributing the probability in the same range, setting SVM parameters to select the optimal kernel function, penalty factor coefficient (c) and parameter coefficient (g) of the kernel function, and obtaining the optimal values of c and g by adopting a cross validation method in the modeling process to train and predict to obtain a model between the spectral data and the blood sugar true value.

The spectrum sensor is a tunable filter combining single-point InGaAsPIN with a Fabry-Perot interferometer, and the type selection is KH-SE-01.

The LED adopts 1613nmLED, 1689nm and 1732nm wavelength, and the type selection is KH-LE-04.

The fingertip position is 6 mm from the front end of the finger fingertip.

The invention has the beneficial effects that: the noninvasive blood glucose detection method optimizes the selection of the detected part to the maximum extent, so that the detection error caused by individual difference is reduced to the minimum;

the most effective LED waveform selection is combined, and the accuracy in blood sugar detection is improved to the maximum extent by accurately selecting wavelength information;

through the appropriate type selection of the LED and the spectrum sensor, the fluctuation of the voltage and current values is reduced to the maximum extent, the detection value of the voltage and current values is further stable, and therefore the voltage and current values are closer to the real blood sugar value, and the voltage and current values can establish a real and stable data relation with the blood sugar value;

the error of blood sugar detection is reduced to the greatest extent through the improvement, so that the accurate detection of non-invasive blood sugar is greatly improved.

Drawings

FIG. 1 is a schematic view of the detection principle of the present invention;

FIG. 2 is a schematic diagram of the wavelength comparison during testing according to the present invention;

fig. 3 is a schematic diagram of blood vessel distribution in the palm.

Detailed Description

The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it will be understood that they are described herein for the purpose of illustration and explanation and not limitation.

The invention effectively solves the problems in three directions during the specific implementation, realizes that the error of the blood sugar test can be controlled to be less than +/-15% through the optimization of the problems in the three directions, the measured blood sugar value is a real voltage and current value with small fluctuation, and the real blood sugar value waveform is displayed through the subsequent calibration of the voltage and current value.

The first major problem, the waveform selection problem;

the propagation of light in tissue follows the law of absorption of light, the beer Lambert law, and the intensity of a beam is set as I₀The parallel monochromatic light is incident into a sample, and as part of the light is absorbed, the transmitted light intensity is It, and then

Wherein A is the degree of absorption of the sample to incident light (referred to as absorbance); ε is the absorption coefficient of the absorber, d is the optical path length of the sample, and c is the concentration of the absorber.

According to the absorption characteristic of blood sugar to near infrared light, the blood sugar has strong absorption to 1550nm wavelength and can carry blood sugar concentration information in tissues; the absorption to 1300nm wavelength is weaker, so that the method can be used in subsequent data processing, and the influence of the tissue structure on the detection of the blood glucose concentration is reduced;

in blood glucose measurement, transmitted light having a fixed wavelength λ 1 of 1550nm and a fixed wavelength λ 2 of 1300nm is measured. Let LD1, LR1 be the measured light intensity and reference light intensity of 1550nm wavelength, respectively, and LD2, LR2 be the measured light intensity and reference light intensity of 1300nm wavelength, respectively. Defining independent variables

Defining the invasive blood sugar concentration value Y as a dependent variable.

According to the beer-lambert law, Y and X are in a linear relation, and the linear regression equation is that Y is α + β X. And (3) establishing a calibration model by using a proper chemometric method for the blood glucose concentration data and the near-infrared light intensity value measured by a standard method, measuring the near-infrared light intensity data of a sample with unknown concentration, and performing prediction analysis on the blood glucose concentration by using the established calibration model.

The constructed blood sugar noninvasive detection optical subsystem consists of an LED light source and a spectrum sensor; therefore, KH-LE-04 and KH-LE-03 provided by custom LED manufacturers are selected, the wavelengths are 1650nm LED and 1550nm, the sensing waveband adopted by the spectral sensor is 1350nm to 1650nm, and the spectral sensor has the characteristics of quick response, high sensitivity and low noise.

The finger transmission measurement method is adopted as shown in fig. 1, namely, the spectrum sensor and the LED light source are respectively positioned at two sides of the finger, the spectrum sensor receives the transmitted light of the tissue, the optical path of the light transmitted in the tissue is long, and more concentration information is carried. Meanwhile, the anatomical structure of the finger tip of the human finger is relatively simple, the superficial capillary vessel network is abundant, and the finger is exposed and measured conveniently. The detection experiments of the light path system are respectively carried out by using LEDs with different wavelengths (no light contrast group, 940nm group, 1613nm group and 1732nm group) (wherein the actual detection of the 940nm group is finger pulse wave, and the waveform characteristics of the pulse wave are very obvious, the contrast detection of the group can intuitively reflect whether the light path of the system is turned on or not), and the data sampling frequency is 1000 Hz. Blank signal, 1613nm signal, and 1732nm signal were recorded for 30, 50, and 75 seconds, respectively. Firstly, carrying out 50Hz notch and 100Hz low-pass filtering processing, and then respectively segmenting the signals by taking 1s as an interval to obtain a blank signal 30 segment, a 1613nm signal 50 segment and a 1732nm signal 75 segment; then, each signal is averaged to obtain 30 values of blank signal, 50 values of 1613nm signal and 75 values of 1732nm signal. The overall distribution of the light group (1613nm group or 1732nm group) and the blank non-light group was analyzed by the rank-sum test to see if there was a significant difference. m and n are the sample sizes of the light and no light groups, respectively.

X, Y, fx (x), fy (x) and fx are distribution functions for the light and blank light groups, respectively, and H0, fx (x) fy (x), H1, fx (x) not equal to fy (x) (3) since n is 30 < m, sample rank and T of Y are chosen as test statistics. When α is 0.05, the rejection region is

The value of the sample rank sum T is calculated to be 795, thereby

If the two global distributions are in the rejection domain, the original hypothesis is rejected, and the alternative hypothesis is accepted, namely the two global distributions are considered to have obvious difference.

And calculating a chi 2 sample value according to the hypothesis test of the overall distribution function, and judging whether the distribution falls in a rejection region or not so as to test whether the distribution belongs to normal distribution or not. The hypothesis test of the overall distribution function uses a goodness-of-fit χ 2 test, whose basic idea is to try to determine a quantity characterizing the degree of fit between the observed data X1, X2, …, Xn and the theoretical distribution F0(X), i.e., the "goodness-of-fit", which, when exceeding a certain limit, indicates that the degree of fit is not high, H0 should be rejected, otherwise H0 should be accepted.

The recorded signals are shown in fig. 2, (a) is a blank control group, and it can be seen that there is a large amount of noise signals in the near-infrared detection, and the near-infrared signals are heavily buried in the background signals. This is because water absorbs near infrared light very severely, and more than 90% of blood is water, while the contents of glucose, cholesterol, triglyceride and other components are very low. At the same time, human tissue (e.g., skin, muscle, blood vessels) is a strong near-infrared absorber, and information that can be used for analysis is buried in these strong backgrounds. Fig. 2(b) obtains the waveform of the finger pulse wave, which can visually reflect the unobstructed optical path of the system and observe the influence of noise on the signal. Fig. 2(c) and fig. 2(d) carry different blood glucose information according to different detection principles, and the waveform difference from the blank group is clearly seen.

The rank sum test results are shown in table 1. Both μ values for the 1613nm and 1732nm groups were in the rejection region, indicating that the data was significantly different from the blank signal. And the blood glucose concentration information contained in the signal can be judged by combining the theory of experiment and the law of beer-Lambert.

TABLE 1 rank sum test results

According to the experimental data, the method comprises the following steps of,

1613nm group light intensity experiment sample value

The calculation of the χ 2 sample value is shown in table 2 by hypothesis testing of the global distribution function.

1732nm group light intensity experiment sample value

The calculation of the χ 2 sample value by hypothesis testing of the global distribution function is shown in table 3.

Table 21613 nm group light intensity χ 2 sample value calculation

Table 31732 nm group light intensity signal chi 2 sample value calculation

By verifying the statistical difference between the transmitted light intensity of the two wavelengths 1613nm and 1732nm and the blank group, the absorption condition of fingertip tissues to different near-infrared light under the beer-Lambert law is verified, and meanwhile, the transmitted light of the two wavelengths 1613nm and 1732nm is also shown to carry respective corresponding blood sugar information. By processing and analyzing the light intensity signals, effective blood sugar information in the light intensity signals is utilized, so that near-infrared noninvasive blood sugar detection is possible.

The first major problem, namely that the superficial capillary vessel network is abundant in the waveform selection problem, and the finger exposure is convenient to measure, but how to select the detection part can further reduce the influence of individual difference on the light wave detection, as shown in fig. 3, the blood vessel distribution of the palm is observed, and the finger tip is distributed with abundant capillary vessel networks, so that the blood sugar content condition in the human body can be effectively reflected, and the position has obvious blood volume change characteristics; the muscle and skeletal tissue of the finger is relatively thin, so background interference information has relatively little influence; in addition, the front end of the finger is convenient to measure, and the examinee has no psychological burden, so that a stable high signal-to-noise ratio spectrum signal can be obtained; the invention finally selects the fingertips of the ring finger and the fingertips of the little finger as the measuring parts.

In order to extract effective signals from a spectrum containing strong background interference, a scheme of adopting close background deduction is effective, and the interference such as background change can be effectively eliminated. However, the human body belongs to an abnormally complex time-varying living body, and it is difficult to find a standard plate whose physicochemical properties are similar to those of human tissue in time and space to eliminate the influence caused by the change of tissue background. However, if the spectrum of the human body itself at a certain time is used as a standard plate, background subtraction can be effectively realized.

The blood flow volume in the human tissue will change periodically with the heart beat, with different blood flow volumes corresponding to different blood thicknesses. Within one or several cycles (in the order of seconds) of the change in the blood flow volume, the physicochemical properties of the background tissues such as skin and muscle and the information on the blood component content are not substantially changed. Therefore, with the blood flow volume spectral subtraction method: the two spectra with different blood volume contents acquired in a very short time are subjected to differential processing, and the background interference of tissues can be effectively deducted.

At two times T1 and T2 with a time interval Δ T (in the order of seconds), the blood thickness changes from L1 to L2 at the same measurement location by an amount of Δ L in volume. When one beam of near-infrared light irradiates the measurement part, the absorbance characteristics corresponding to the time t1 and the time t2 can be calculated according to the Lambert-Beer law:

A1＝lg〔I0/I(t1)〕＝lg(I0/I1)

A2＝lg〔I0/I(t2)〕＝lg(I0/I2)

because the delta T is extremely short, the tissue characteristics of the skin, muscle and the like of the measurement part and the content information of blood components are basically not changed, and the change is only the optical path of blood, therefore, the spectrums measured at the two moments can be mutually referred to, and the absorbance spectrum of the blood corresponding to the change amount of the optical path is obtained:

ΔA＝lg〔I2/I1〕＝lg(I0/I1)-lg(I0/I2)＝A1-A2

the above process is equivalent to differentiating the human body spectra measured at two moments at an interval of Δ T, i.e. obtaining a near-infrared volume difference spectrum. The analysis formula shows that when the blood composition is analyzed by the blood flow volume spectrum subtraction method, the blood spectrum information with the volume difference delta L can be obtained only by directly calculating the logarithm of the ratio of the light signals with different intensities measured at the time t1 and the time t 2.

After the spectrum subtraction method is used for processing, human tissue background information is not carried in the volume difference spectrum, and meanwhile, interference factors such as individual difference, different contact pressures and the like which are related to the tissue background and are easy to change in the measurement process are effectively eliminated. The effective signal content related to blood components in the volume difference spectrum is greatly improved, and the near infrared spectrum noninvasive detection blood glucose analysis precision is favorably ensured.

Aiming at the problem of formation of blood glucose waveform data, the invention carries out targeted filtering and amplification on a signal through a signal processing module, and improves the signal-to-noise ratio and the peak value of a signal waveform. And then, the signal is lifted and limited, so that the signal conforms to the acquisition range of the ARM. Then, an ADC acquisition function of the ARM is used, analog signals from the filter circuit are converted into digital signals for subsequent use, and the ARM processes the acquired digital signals through a series of algorithm software to finally obtain processed waveform peak data. After data are transmitted to APP in electronic equipment (such as a mobile phone, a tablet personal computer and a notebook computer) with a Bluetooth function through Bluetooth, a blood glucose meter with a medical instrument license is used for testing a blood glucose value of a user, the blood glucose value is input into the APP to serve as a reference value for modeling, and the reference value is matched with waveform data and provided for MATLAB to establish a support vector machine model. Executing the prediction function of the acquired peak data in MATLAB every time when the predicted blood sugar model is used, calculating to obtain the predicted blood sugar value,

the system hardware design scheme of the invention mainly comprises: designing and manufacturing an LED with a finger bin and a spectrum sensor; the design and manufacture of a spectrum sensor signal amplifying and filtering circuit, a power supply circuit and an ARM-based data acquisition, storage, processing and display circuit.

A finger bin module: the LED lamp is controlled to be on or off through the IO port of the single chip microcomputer, the constant current power supply is used for supplying power to the LED lamp, the brightness of the LED lamp is stable, and when specific infrared light penetrates through finger fingertips, the spectrum sensor receives light signals and converts the light signals into current signals.

The signal processing module: the method comprises the steps of firstly converting weak current signals into voltage signals through an operational amplifier and a resistor, then carrying out band-pass filtering, wherein low frequency filtering is used for reducing ambient light interference, high frequency filtering is used for reducing clutter interference, the voltage signals are boosted through an amplifier, and because negative voltage signals can exist and a single chip AD cannot convert the negative voltage signals, the amplification factor needs to be controlled, so that the output voltage signals are between 0 and 3.3V.

A CPU processing module: convert the voltage signal of front end input through the ADC module, then communicate the signal to APP through a plurality of bluetooth equipment, obtain blood sugar data after APP will use cloud computing to call algorithm processing, return to APP again and show blood sugar numerical value and data waveform.

The above embodiments are preferred embodiments of the present invention, and those skilled in the art can make variations and modifications to the above embodiments, therefore, the present invention is not limited to the above embodiments, and any obvious improvements, substitutions or modifications made by those skilled in the art based on the present invention are within the protection scope of the present invention.

Claims (4)

1. A noninvasive blood glucose determination method is characterized by comprising the following steps: the method comprises the following steps: the method comprises the following steps: a spectrum sensor is arranged on one side of the fingertip position, and an LED light source is designed on the other side relative to the fingertip position;

step two: a tunable filter of a Fabry-Perot interferometer is adapted in the spectrum sensor, and the optical receiving range of the tunable filter is adjusted to reach nm level;

step three: collecting light rays emitted by an LED with the wavelength of 1530-1850 nm by a spectral sensor with the wavelength of 1350-1650 nm after penetrating through fingertip tissues of a human body;

step four: using blood flow volume spectroscopy subtraction: carrying out differential processing on two spectra with different blood volume quantities acquired in a very short time; at two time points T1 and T2 with a time interval Δ T, the blood thickness at the same measurement position changes from L1 to L2, the volume change is Δ L, and when a beam of near infrared light irradiates the measurement position, the absorbance characteristics corresponding to the time points T1 and T2 can be calculated according to Lambert-be law:

A1＝lg〔I0/I(t1)〕＝lg(I0/I1)

A2＝lg〔I0/I(t2)〕＝lg(I0/I2)

a blood absorbance spectrum corresponding to the amount of change in optical path length is obtained by referring to the spectra measured at two times:

ΔA＝lg〔I2/I1〕＝lg(I0/I1)-lg(I0/I2)＝A1-A2

obtaining a near-infrared volume difference spectrum;

step five: inputting two groups of near infrared spectrum data with different wavelengths as independent variable matrixes of a model, taking blood sugar values collected by a blood sugar instrument as dependent variables of the model, dividing a sample training set and a test set, and testing a plurality of groups of data of a user;

step six: sorting the screened spectral data according to the blood sugar values measured by a blood sugar instrument, taking two groups of spectral data measured by different wavelengths as independent variables and blood sugar values as dependent variables, respectively normalizing, distributing the probability in the same range, setting SVM parameters to select the optimal kernel function, penalty factor coefficient (c) and parameter coefficient (g) of the kernel function, training and predicting the optimal values of c and g by adopting a cross validation method in the modeling process, and obtaining a model between the spectral data and the blood sugar true value.

2. The method of noninvasive blood glucose sensing based on spectral sensor of claim 1, characterized in that: the spectral sensor is a tunable filter combining single-point InGaAsPIN with a Fabry-Perot interferometer, and the type selection is KH-SE-01.

3. The method of noninvasive blood glucose sensing based on spectral sensor of claim 1, characterized in that: the LED adopts 1613nmLED, 1689nm and 1732nm wavelength, and the type selection is KH-LE-04 and KH-LE-03.

4. The method of noninvasive blood glucose sensing based on spectral sensor of claim 1, characterized in that: the fingertip position is 6 mm from the front end of the finger fingertip.

Images