Research on non-invasive blood glucose detection technology using near infrared spectroscopy

　　Diabetes is an endocrine disease. According to reports, in 1997, there were more than 120 million diabetes patients worldwide, and this number will grow to more than 220 million by 2010. The most effective treatment for diabetes is to control blood sugar concentration through frequent testing and insulin injections, thereby reducing or alleviating complications caused by diabetes.

　　The main method of testing blood sugar is to draw blood from the body and analyze it through biochemical testing. This is a invasive test, which brings pain and inconvenience to patients. Non-invasive blood sugar testing has attracted great attention. Its significance is: (1) reducing the pain of daily blood drawing and measurement for patients, improving the quality of life of patients; (2) increasing the number of measurements, improving the accuracy of blood sugar control, and reducing the risk of diabetic complications; (3) reducing the cost of each measurement; (4) it is possible to form a closed-loop circulation system containing a detector and insulin injection; (5) its measurement method and principle can be extended to the detection of other blood components. Infrared spectroscopy is the most commonly used method in non-invasive blood sugar testing research. By analyzing the spectral signal of a beam of infrared light passing through human tissue or reflected by it, the content of glucose in the tissue is determined. At present, the more effective spectral range is the near-infrared region (wavelength 0.7um-2.5um).

　　2 Principles and methods of infrared spectroscopy for glucose detection 2.1 Near-infrared absorption of glucose in aqueous solution The absorption of organic molecules in the near-infrared spectral region is mainly caused by the absorption of the harmonic and summed frequencies of the molecular vibration of hydrogen-containing groups [1]. The harmonic and summed frequency spectra of organic molecules can obtain information on the molecular structure and composition state. The near-infrared spectrum of organic matter is highly characteristic and less affected by the internal and external environment of the molecule, but the harmonic and summed frequency spectra are much wider than the fundamental frequency absorption bandwidth, which makes the near-infrared spectrum of multi-component samples seriously overlap in the spectral bands of different components, the spectral bands of different groups in the same component, and the harmonic and summed frequency bands of different forms of the same group, making the spectrum analysis of the near-infrared spectrum extremely difficult. It is difficult to separate the chemical components in a mixture into a single, non-overlapping absorption spectrum of each component. In the presence of strong background absorption of water, it is difficult to measure the content of low-concentration substances by conventional methods in biological mixtures. Water is the main component of biological tissues. It not only has a single infrared spectrum, but also has a rich summed frequency and harmonic frequency spectrum extending to the near-infrared region. The infrared spectrum analysis of water shows that water has a small absorption at a wavelength of 2.01um-2.5um, forming a region called the water transmission window, so the best analysis wavelength for aqueous solution substances is 2.0um-2.5um. The absorption rate of water above 3um is greater than 6AU/mm, making it difficult to measure other substances.

　　2.2 Specificity of glucose spectrum The fundamental frequency of glucose infrared absorption obtained in glucose solid and glucose solution has been reported long ago. Glucose stretching vibration can produce strong absorption bands of composite and overtones. The measurement of near infrared (2.0um-2.5um) spectrum of glucose aqueous solution has absorption peaks, and the spectrum of glucose is unique, but the composite and overtone spectrum of glucose in infrared region overlap with several composite and overtone frequencies of water, fat and hemoglobin electronic absorption bands, that is, covered by the spectrum of other components. This is the main interference of glucose infrared spectrum measurement. The overlap of absorption bands in near infrared region and diffuse reflection spectrum of organic mixture are not the superposition of spectra when each component exists alone. Tissue absorption also affects glucose measurement. In a small part like finger, near infrared light will weaken 3-4 absorption units, and the change of spectral absorption is about 10-5 absorption units when the glucose concentration changes by 5mmoL/L. Tissue light scattering also has a great impact on glucose measurement. The light intensity of tissue scattering, positioning error and the influence of various body factors are the most important measurement errors, all of which affect the application of near-infrared spectroscopy in blood glucose testing.

　　2.3 Spectral analysis method Chemometrics is very effective in infrared spectroscopy analysis. Chemometrics uses multivariate analysis to calibrate statistical methods and computational techniques to analyze chemical measurement data and calculate the content of each component of the sample from the infrared spectrum. Among the commonly used multivariate analysis correction methods, the partial least squares (PLS) method has a better effect on the spectral analysis of blood glucose detection. It uses the principal factor analysis method for quantitative calculation of the spectral group of known glucose concentrations, performs eigenvector analysis on the spectral matrix, and then uses multivariate linear regression to find the relationship between the smallest spectral changes and the analyte concentration, eliminates spectral variables unrelated to glucose, and obtains the corrected spectrum. The glucose concentration is determined by the inner product (i.e., dot product) of the corrected spectrum and the sample spectrum.

　　3 Research status of in vitro and in vivo detection 3.1 In vitro near-infrared spectroscopy measurement of mixed glucose solution Jonathon T. Olesberg et al. used 80 samples containing glucose, lactate, alanine, ascorbate, urea and acetic acid glycerol to measure the spectrum of glucose solution in the wavelength bandwidth range of 2.0um-2.5um, and used PLS to correct the spectrum to predict the concentration of solution components. The results showed that the standard deviation of the measurement prediction of glucose solution within 0-35mm was 0.39mm, lactate was 0.12mm, alanine was 0.53mm, ascorbate was 0.23mm, urea was 0.11mm, and acetic acid glycerol was 0.12mm, and the results were quite satisfactory. At present, the glucose concentration can be predicted in aqueous solutions with components ranging from simple to complex, but these solutions are still simple compared to blood or plasma, and the components studied are at most 5, so further research is needed on aqueous solutions with more components to simulate plasma or blood systems.

　　3.2 Plasma or whole blood near infrared spectroscopy glucose measurement Haahand obtained 4 different whole blood samples from the population and added glucose to them. For each individual, 20 blood samples with glucose concentrations ranging from (3-743) mg/dl were prepared, and then the near infrared spectrum of each sample was collected in the range of (1.5-2.3) um, and then these spectra were used to create a PLS calibration model using the reference glucose concentration. After studying the obtained spectra, it was shown that 2.0um-2.3um contained a lot of glucose information. Using this area, the cross-checked SEP value was 30.5mg/dL. This error is large, but it can be reduced by increasing the number of calibration samples and controlling the temperature of the samples during the scanning process. [page]

　　Amord et al. used digital filtering technology to determine the glucose concentration in bovine plasma. The bovine blood was centrifuged to obtain plasma, and different amounts of glucose were added to prepare 69 samples. The spectra of these samples were collected in the range of 2.01um-2.5um. By observing these spectra, it was found that some areas contained high noise. They introduced Fourier filtering to reduce noise and baseline offset. The SEP value was obtained through PLS calibration and prediction. The results show that near-infrared spectroscopy can be used to determine the glucose concentration in the plasma matrix, and the accuracy and precision are within the allowed error range.

　　We used disodium hydrogen phosphate and sodium dihydrogen phosphate to prepare glucose buffered aqueous solutions of different concentrations, with glucose concentrations ranging from 18mg/dL to 1800mg/dL. A total of 20 solution samples were prepared. In addition, a glucose solution with bovine serum albumin (BSA) was prepared. During the preparation, 70mg of BSA was added to the 900mg/dL glucose buffer solution to make samples. Blood samples with known glucose concentrations were collected clinically and studied using a MAGVA-AR560 near-infrared Fourier transform spectrometer in the near-infrared spectral range of 1.61xm-2.51xm. Good results were also obtained using PLS analysis.

　　3.3 In vivo near-infrared spectroscopy blood glucose measurement The key to in vivo near-infrared spectroscopy blood glucose measurement is to establish a corrected spectrum in an in vivo environment. Since there are many sources of error that affect the measurement, they need to be eliminated or compensated by calibration. Some errors that affect the measurement are not easily incorporated into the calibration. Such error sources mainly include detector positioning errors, the influence of temperature and pulse, mechanical pressure of the detection equipment, hydration, sweating, changes in blood volume and blood flow volume, etc. There are currently two main research methods. One is the experimental method, which non-invasively collects spectral signals from non-diabetic people and diabetic patients during the oral glucose tolerance test (OGTT), and measures blood glucose concentration using a invasive method. Finally, a model is established based on the relationship between the obtained blood glucose value and the non-invasively collected optical signal. This method cannot measure other metabolites, interfering substances, biological noise, or changes in the contact surface between the instrument and the body, but it can calculate the impact of these noises. Another method is the physical model method, in which the glucose signal is first measured in a set of standard glucose solutions. Then gradually increase the complexity of the standard solution to simulate human tissue, and describe the accuracy and correctness of each step, and then use mathematical models to correlate the data for light propagation in tissue, and finally apply the research measurement methods and systems to the human body. The resulting in vivo signal is then correlated with the traumatic data through chemical measurement technology. This method can identify the noise component, so this method is used to eliminate the influence of noise on the signal before using chemical measurement technology.

　　The near-infrared diffuse reflectance spectral characteristics of the skin on the back of the hand are similar to those of aqueous solutions. Human tissue also has a transmission window in the near-infrared region, so it is possible to measure the concentration of glucose at 2.0um-2.5um. A theoretical model containing fat and glucose has been used to simulate the light absorption of tissue glucose in the range of 2.0?m-2.5?m. The glucose concentration used in these studies is usually higher than the physiological concentration range. However, since several current technologies cannot well determine the measured signal, for an individual with a changing blood glucose concentration, the data from the oral glucose tolerance test can be used to establish a non-invasive measurement response for blood glucose concentration. The data generated during the test can also be used to predict blood glucose concentration in subsequent non-invasive measurements. Since the non-invasive measurement response may have non-sugar physiological effects, the clinical calibration determined by the relationship between the oral glucose tolerance test and the non-invasive measurement response will produce a calibration curve that is unique to the individual being tested. However, this calibration curve may need to be periodically updated through invasive testing. The oral glucose tolerance test and dietary tolerance test used for calibration produce a series of measurements that are continuous in time, but if random sampling is not possible, these time-dependent data will affect the results of multivariate calibration. In this way, the temporal distribution of spectral signals and noise may lead to incorrect associations with blood glucose. In vivo transdermal studies have shown that it is not possible to distinguish between directly measured glucose concentrations and accidental relationships within the data set. Therefore, the current level of research is still unacceptable for home blood glucose monitors.

　　4 Problems with detection Disadvantages of near-infrared in vivo detection of glucose concentration: (1) Low measurement accuracy; (2) Requires repeated calibration; (3) Affected by medication and many other interfering factors; (4) The absorption intensity of water in the near-infrared band is very sensitive to the concentration and temperature of the dissolved matter; (5) It has not yet been approved by the US FDA; (6) The instruments used are expensive and difficult to miniaturize.

　　5 Conclusions There is analytical information for detecting glucose in the near-infrared spectrum, and the optimal spectral range is 2.0um-2.5um. The spectrum of glucose is independent, but it overlaps with the spectra of other substances in the human body. Near-infrared light rarely penetrates the subcutaneous tissue, and the absorbance of glucose is very small, so it is difficult to measure glucose using near-infrared spectroscopy. There is indeed a correlation between the blood glucose level in the body and the measured optical signal, but there is no non-invasive experiment that can provide evidence to prove that the measured signal can be associated with the actual blood glucose concentration. It is necessary to research and develop more sensitive and more stable testing equipment. There is little basic data and clinical data for constructing models in China. Based on the above situation, we analyzed various methods for non-invasive blood glucose detection research, combined with the actual situation of technology and equipment, and used near-infrared transflection method to do some basic research and achieved good results.