A non-invasive blood glucose testing method based on group big data model

Abstract

The present invention discloses a non-invasive blood glucose testing method based on a group big data model, which relates to the technical field of blood glucose testing, and includes the following steps: using a near-infrared spectrometer to detect spectral data in skin and tissues, and obtaining the absorption peak of glucose molecules in the near-infrared range; performing preliminary feature extraction on the obtained absorption peak, and after performing abnormal analysis on the extracted features, preliminarily judging whether the absorption peak is abnormal, and screening out the abnormal absorption peak. The present invention uses a machine learning model to screen out abnormal absorption peaks through feature extraction and abnormal analysis, ensures the use of normal data, improves the accuracy of blood glucose prediction, reduces the risk of overestimation or underestimation, provides reliable monitoring for diabetic patients, adjusts drugs in time, and reduces the risk of complications. At the same time, differential analysis and standardized deviation calculation screen out abnormal prediction values, optimize data processing, improve system robustness and final prediction accuracy, and provide accurate non-invasive blood glucose monitoring solutions.

Description

Technical Field

[0001] The present invention relates to the technical field of blood sugar testing, and in particular to a non-invasive blood sugar testing method based on a group big data model. Background Art

[0002] Non-invasive blood glucose testing based on group big data models is a method of monitoring blood glucose levels using big data analysis and machine learning techniques. Traditional blood glucose testing usually requires piercing the skin to obtain blood samples, while non-invasive blood glucose testing uses non-invasive methods such as optical sensors and skin conductivity measurement to detect blood glucose levels. The principle of this method is that certain physiological parameters of the human body (such as light reflectance, transmittance, conductivity, etc.) will change with changes in blood glucose levels. These physiological parameters are collected by sensors and analyzed using big data models to infer blood glucose levels. This technology reduces the pain of patients and improves the convenience of testing.

[0003] Big data models play a vital role in non-invasive blood glucose testing. By collecting a large number of individual physiological data and corresponding blood glucose level data, a huge database is established, from which machine learning algorithms can identify complex associations and patterns. These models are continuously optimized during the training process to improve the accuracy and reliability of predictions. In addition, big data models can also be adjusted adaptively. As more data is added and time passes, the model can be continuously improved and refined, thereby providing personalized blood glucose monitoring solutions for different individuals. This non-invasive blood glucose testing technology based on group big data models not only provides a more comfortable testing experience for diabetic patients, but also provides an efficient and accurate monitoring method for the medical field.

[0004] When performing non-invasive blood glucose testing, existing technologies usually use spectral techniques, including near-infrared spectroscopy (NIR), mid-infrared spectroscopy (MIR), and Raman spectroscopy, to measure the light reflection and absorption characteristics in the skin, tissues, and blood. After light of different wavelengths penetrates the skin, it is absorbed and scattered by blood glucose molecules to different degrees. By analyzing the spectral characteristics of reflected or transmitted light, the blood glucose concentration can be indirectly inferred. The NIR spectrum range is usually between 700 and 2500 nanometers. Glucose molecules have specific absorption peaks in the near-infrared range. These absorption peaks correspond to changes in the vibration and rotation energy levels of glucose molecules. When near-infrared light passes through the skin and tissues, glucose molecules absorb NIR light of specific wavelengths. These absorption characteristics can be used to infer blood glucose concentration.

[0005] The prior art has the following deficiencies:

[0006] When using spectral technology to estimate blood sugar concentration, glucose molecules produce specific absorption peaks in the near-infrared range. If the absorption peak type of glucose molecules in the near-infrared range is abnormal and the existing technology fails to screen out these abnormal absorption peaks, the abnormal absorption peaks will cause the predicted results of blood sugar concentration to deviate from the actual value, which may overestimate or underestimate the blood sugar concentration. Misdiagnosis and mistreatment will cause diabetic patients to fail to adjust drug dosages in time, increase the risk of acute complications (such as hypoglycemic coma or hyperglycemic ketoacidosis), and long-term mistreatment will also lead to the aggravation of chronic complications (such as retinopathy, nephropathy, and cardiovascular disease);

[0007] Existing technologies usually perform a comprehensive analysis of all detected absorption peaks to estimate blood glucose concentrations. This method has several obvious shortcomings. First, the overall comprehensive analysis method requires processing a large amount of data, which requires a lot of calculations and increases the complexity and processing time of the system. Secondly, when the accuracy of blood glucose concentration estimation in certain local areas is poor, it is difficult for comprehensive analysis to detect these abnormal areas in time, making it impossible to effectively correct and screen them out. This will cause the overall blood glucose concentration prediction results to deviate from the actual value, and may overestimate or underestimate the blood glucose concentration, thereby affecting the patient's health management and treatment decisions.

[0008] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0009] The purpose of the present invention is to provide a non-invasive blood glucose testing method based on a group big data model. By performing feature extraction and abnormal analysis on absorption peaks, a pre-trained machine learning model is used to screen out abnormal absorption peaks, ensure the use of normal data for analysis, improve the accuracy of blood glucose concentration prediction, reduce the risk of overestimation or underestimation, provide reliable blood glucose monitoring for diabetic patients, adjust drug dosage in time, and reduce the risk of complications. At the same time, through difference analysis and standardized deviation calculation, abnormal blood glucose concentration values ​​are screened out, the data processing process is optimized, the system robustness and the accuracy of the final prediction are improved, and an accurate non-invasive blood glucose monitoring solution is provided to solve the problems in the above-mentioned background technology.

[0010] In order to achieve the above object, the present invention provides the following technical solution: a non-invasive blood glucose testing method based on a group big data model, comprising the following steps:

[0011] The spectral data in the skin and tissues are detected by a near-infrared spectrometer to obtain the absorption peak of glucose molecules in the near-infrared range;

[0012] Perform preliminary feature extraction on the acquired absorption peaks, and after abnormal analysis of the extracted features, preliminarily determine whether the absorption peaks are abnormal, and filter out the abnormal absorption peaks;

[0013] Include all absorption peaks that have not been screened out in a circular area to ensure that all relevant spectral data are included in the analysis range, and further divide the circular area into several sub-areas of the same size so that the absorption peak data in each sub-area is relatively independent;

[0014] After the sub-regions are divided, the absorption peaks in each sub-region are analyzed in detail, and the blood glucose concentration in each sub-region is predicted using the population big data model based on the absorption peak characteristics in each sub-region;

[0015] The predicted blood glucose concentration values ​​of all sub-regions are established into an analysis set, and the difference analysis of the blood glucose concentrations in the analysis set is performed to evaluate the differences in the predicted values ​​between the sub-regions. If it is found that the predicted values ​​of some regions are obviously inconsistent with the overall trend, the abnormal predicted blood glucose concentration values ​​in the analysis set are screened out based on the difference analysis;

[0016] The predicted blood glucose concentration values ​​after screening out abnormal data are averaged to calculate the final predicted blood glucose concentration.

[0017] Preferably, the absorption peak of glucose molecules in the near-infrared range is obtained by the following specific steps:

[0018] Before the actual measurement, a set of background spectrum data is collected;

[0019] Aim the light source of the near-infrared spectrometer at the skin or tissue sample and start the spectrum measurement program. The near-infrared light emitted by the light source penetrates the skin and tissue, part of the light is absorbed by the glucose molecules, and the other part is scattered and reflected. The detector of the spectrometer receives the light after passing through the sample and records the light absorption data of different wavelengths.

[0020] After collecting the spectral data of the sample, first, compare the sample spectral data with the background spectral data to eliminate background interference and noise. Then, use spectral processing software to extract the absorption peak characteristics in the spectrum. Finally, through feature analysis, identify the absorption peak related to glucose molecules to provide basic data for subsequent blood glucose concentration prediction.

[0021] Preferably, preliminary feature extraction is performed on the acquired absorption peak, and the extracted features include the offset of the absorption peak position and the symmetry of the absorption peak. Within the monitoring window, after abnormal analysis of the offset of the absorption peak position and the symmetry of the absorption peak, peak offset quantification values ​​and peak symmetry quantification values ​​are generated respectively. The peak offset quantification values ​​and peak symmetry quantification values ​​generated after the abnormal analysis are input into a pre-trained machine learning model, and a screening coefficient is generated by the machine learning model. The screening coefficient is used to preliminarily determine whether there is an abnormality in the absorption peak.

[0022] Preferably, the screening coefficient generated after abnormal analysis of each absorption peak under the monitoring window is compared and analyzed with a preset reference threshold of the screening coefficient to preliminarily determine whether the absorption peak is abnormal, and screen out the abnormal absorption peak. The specific steps are as follows:

[0023] If the screening coefficient is greater than or equal to the screening coefficient reference threshold, the absorption peak is judged as an abnormal absorption peak and the absorption peak is screened out;

[0024] If the screening coefficient is less than the screening coefficient reference threshold, the absorption peak is judged as a normal absorption peak and is retained.

[0025] Preferably, the absorption peak characteristics in each sub-region are used to predict the blood glucose concentration in the sub-region using a population big data model, and the specific steps are as follows:

[0026] After the sub-region division, the spectral data in each sub-region is first preprocessed;

[0027] After preprocessing, key features are extracted from the absorption peak data in each sub-region;

[0028] Aggregate the feature data in each sub-region so as to input it into the group big data model;

[0029] Using the aggregated feature data, a group big data model is used to predict blood glucose concentration.

[0030] Preferably, the predicted values ​​of blood glucose concentration in all sub-regions are integrated into an analysis set, and there are m sub-regions, and the predicted values ​​of each sub-region are , , , , , the predicted values ​​of all sub-regions are combined into a vector C, then: ,in, represents the predicted value of blood glucose concentration in the kth sub-region;

[0031] Calculate global statistics of the analysis set C, including the median and interquartile range , median is the median value of the sorted set, the interquartile range is the upper quartile and lower quartile The difference is used to measure the degree of dispersion of data;

[0032] Use global statistics to perform difference analysis and calculate the predicted value for each sub-region With median The absolute difference between , the calculation expression is: , where Represents the absolute deviation between the predicted value of the kth sub-region and the median;

[0033] The absolute deviation Standardized using the interquartile range To measure the relative difference, the expression of standardized absolute deviation is: , where represents the standardized deviation of the k-th sub-region, indicating the relative difference;

[0034] The standardized deviation of the kth sub-region is compared and analyzed with the pre-set deviation reference threshold. If the standardized deviation is greater than or equal to the deviation reference threshold, the predicted value of the sub-region is marked as abnormal and the predicted value of the sub-region is screened out. If the standardized deviation is less than the deviation reference threshold, the predicted value of the sub-region is marked as normal and the predicted value of the sub-region is retained.

[0035] Preferably, the remaining normal prediction values ​​after screening out the abnormalities are grouped as ,but: , where represents the predicted value after filtering out abnormal data, represents the total number of predicted values ​​after filtering out abnormal data, and the calculation expression for the final predicted blood glucose concentration is: , where represents the final predicted blood glucose concentration.

[0036] Preferably, within the monitoring window, after performing abnormal analysis on the shift of the absorption peak position, the specific steps of generating the peak shift quantization value are as follows:

[0037] Determine the reference wavelength range of the absorption peak of glucose molecules , and Respectively represent the minimum and maximum values ​​of the reference wavelength range;

[0038] In the monitoring window, the actual wavelength position is measured and the actual wavelength position is used express;

[0039] Calculate the offset of the actual wavelength position relative to the central wavelength. The calculation expression is: , where represents the center wavelength of the reference wavelength range, which is the expected peak position. , Indicates the offset of the actual wavelength position relative to the central wavelength;

[0040] Considering the offset direction and amplitude, positive and negative offset weights are introduced, and the expression for weight calculation is: , ,in, and It is a parameter that adjusts the offset sensitivity and controls the rate of change of the positive offset weight and the negative offset weight. and Represent the positive offset weight and negative offset weight respectively, represents the normalized offset, ;

[0041] Calculate the peak shift quantification value, the calculation expression is: , represents the absolute value of the normalized wavelength offset, Represents the difference between the positive bias weight and the negative bias weight, Represents the normalization coefficient, which is used to adjust the proportional relationship between the normalized offset and the actual measured wavelength. Represents the peak shift quantification value.

[0042] Preferably, within the monitoring window, after performing abnormal analysis on the symmetry of the absorption peak, the specific steps of generating the peak symmetry quantization value are as follows:

[0043] Determine the highest point of the absorption peak, that is, the point with the highest absorbance, search to the left from the highest point of the absorption peak to find the point where the absorbance drops to the base value, mark this point as L, search to the right from the highest point of the absorption peak to find the point where the absorbance drops to the base value, mark this point as R;

[0044] The peak is segmented, and a data set is established by taking the data points between L and the highest point of the absorption peak, and the data set is recorded as the left side data of the peak; and a data set is established by taking the data points between R and the highest point of the absorption peak, and the data set is recorded as the right side data of the peak;

[0045] The difference between the data on the left side of the peak and the data on the right side of the peak is calculated to compare the symmetry of the two sides. The calculation expression is: , where represents the difference value of the symmetric difference function at the i-th data point, represents the absorbance of the ith data point on the left side of the peak, Indicates the data on the right side of the peak The absorbance of the data points, n is the total number of data points on both sides, represents the symmetry difference function, Indicates the data on the left side of the peak, represents the data on the right side of the peak, and i represents the index of the data point;

[0046] The symmetry difference function is integrated to quantify the overall symmetry difference. The integral is calculated as: ,in, It indicates the degree of difference in the overall symmetry of the absorption peak;

[0047] Calculate the peak symmetry quantitative value, the calculation expression is: , where represents the quantitative value of peak symmetry, represents the absorbance function of the peak, x represents the wavelength, Represents the total area of ​​the absorption peak.

[0048] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0049] The present invention performs preliminary feature extraction on the acquired absorption peaks, performs abnormal analysis on these features within the monitoring window, and inputs the analysis results into a pre-trained machine learning model to preliminarily determine whether there are abnormalities in the absorption peaks, and screen out abnormal absorption peaks, ensuring that only normal spectral data that accurately reflects the characteristics of glucose molecules are used for analysis. In this way, the blood glucose concentration prediction error caused by abnormal absorption peaks is reduced, the overall prediction accuracy is improved, and the risk of overestimating or underestimating the blood glucose concentration is reduced, thereby providing more reliable blood glucose monitoring for diabetic patients, ensuring that diabetic patients can adjust drug dosages in a timely and accurate manner, and reducing the risk of acute and chronic complications.

[0050] After the present invention establishes the predicted blood glucose concentration values ​​of all sub-regions into an analysis set, it performs difference analysis on the blood glucose concentrations in the set and evaluates the differences in the predicted values ​​between the sub-regions. It can timely discover and identify those abnormal regions that are obviously inconsistent with the overall trend, and screen out the predicted abnormal blood glucose concentration values ​​by calculating the standardized deviation of each sub-region, thereby further improving the accuracy of blood glucose concentration prediction. Through this method, not only the negative impact on the overall result due to the poor prediction accuracy of the local area is reduced, but also the data processing flow can be optimized, and the robustness and reliability of the system can be improved. Finally, the predicted blood glucose concentration value after screening out the abnormal data is calculated by taking the average value to obtain the final predicted blood glucose concentration. This refined processing ensures the stability and accuracy of the final result, and provides a more accurate and reliable non-invasive blood glucose monitoring solution for diabetic patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0052] Figure 1 The present invention is a method flow chart of a non-invasive blood glucose testing method based on a group big data model. DETAILED DESCRIPTION

[0053] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0054] The present invention provides Figure 1 A non-invasive blood glucose testing method based on a group big data model is shown, comprising the following steps:

[0055] The spectral data in the skin and tissues are detected by a near-infrared spectrometer to obtain the absorption peak of glucose molecules in the near-infrared range;

[0056] This step involves the use of precise optical sensors and spectroscopic instruments to ensure high resolution and high signal-to-noise ratio of the spectral data, thereby accurately capturing the absorption properties of the glucose molecule.

[0057] To obtain the absorption peak of glucose molecules in the near-infrared range, the specific steps are as follows:

[0058] Before the actual measurement, a set of background spectrum data is collected;

[0059] Background spectrum data is obtained by measuring air or reference materials without samples. These data are used to correct and compensate for background interference caused by factors such as ambient light sources and instrument noise. The collection of background spectrum data can be repeated multiple times to improve the reliability and accuracy of the data.

[0060] Aim the light source of the near-infrared spectrometer at the skin or tissue sample and start the spectrum measurement program. The near-infrared light emitted by the light source penetrates the skin and tissue, part of the light is absorbed by the glucose molecules, and the other part is scattered and reflected. The detector of the spectrometer receives the light after passing through the sample and records the light absorption data of different wavelengths.

[0061] These data reflect the absorption properties of glucose molecules and other biomolecules within the sample.

[0062] After collecting the spectral data of the sample, first, compare the sample spectral data with the background spectral data to eliminate background interference and noise. Then, use spectral processing software to extract the absorption peak characteristics in the spectrum (including the position, intensity and shape of the absorption peak, etc.). Finally, through feature analysis, identify the absorption peak related to glucose molecules to provide basic data for subsequent blood glucose concentration prediction.

[0063] Perform preliminary feature extraction on the acquired absorption peaks, and after abnormal analysis of the extracted features, preliminarily determine whether the absorption peaks are abnormal, and filter out the abnormal absorption peaks;

[0064] A preliminary feature extraction is performed on the acquired absorption peak, and the extracted features include the offset of the absorption peak position (wavelength) and the symmetry of the absorption peak. Within the monitoring window, after an abnormal analysis of the offset of the absorption peak position (wavelength) and the symmetry of the absorption peak, a peak offset quantification value and a peak symmetry quantification value are generated respectively. The peak offset quantification value and the peak symmetry quantification value generated after the abnormal analysis are input into a pre-trained machine learning model, and a screening coefficient is generated by the machine learning model. The screening coefficient is used to preliminarily determine whether there is an abnormality in the absorption peak.

[0065] The shift of the absorption peak position (wavelength) deviates from the specific wavelength range, indicating that the absorption peak is abnormal, which will have a serious impact on the prediction accuracy of blood glucose concentration. The specific absorption peak position of glucose molecules in the near-infrared spectrum is based on the characteristic wavelength of its molecular vibration and rotation energy level changes, and these positions are crucial for accurately identifying and quantifying blood glucose concentration. When the absorption peak position is offset, the model may mistakenly identify the absorption characteristics of other molecules as glucose characteristics, causing the prediction model to obtain incorrect information in the feature extraction stage. This error will accumulate in the subsequent data processing and analysis process, causing the blood glucose concentration prediction results to deviate from the actual value, and may overestimate or underestimate the blood glucose concentration. This misdiagnosis and mistreatment will seriously affect the health management of diabetic patients, resulting in the inability to adjust drug dosage in time, increasing the risk of acute complications (such as hypoglycemic coma or hyperglycemic ketoacidosis), and long-term mistreatment may also lead to the aggravation of chronic complications (such as retinopathy, nephropathy, cardiovascular disease). Therefore, ensuring the accuracy and stability of the absorption peak position is crucial to the accuracy and reliability of non-invasive blood glucose monitoring.

[0066] After performing abnormal analysis on the shift of the absorption peak position (wavelength) within the monitoring window, the specific steps for generating the peak shift quantification value are as follows:

[0067] Determine the reference wavelength range of the absorption peak of glucose molecules , and Respectively represent the minimum and maximum values ​​of the reference wavelength range;

[0068] The reference wavelength range is obtained based on a detailed spectral analysis of glucose molecules under laboratory conditions. The absorption peak position of the glucose solution is measured multiple times under controlled variables using a high-precision near-infrared spectrometer. Combined with the standard absorption peak position reported in the literature, the wavelength range of the specific absorption peak of the glucose molecule in the near-infrared range is determined through statistical analysis to ensure the consistency and repeatability of the measured absorption peak.

[0069] In the monitoring window, the actual wavelength position is measured and the actual wavelength position is used express;

[0070] The actual wavelength position is the peak position detected by the spectrometer in the current sample measurement.

[0071] Calculate the offset of the actual wavelength position relative to the central wavelength. The calculation expression is: , where represents the center wavelength of the reference wavelength range, which is the expected peak position. , Indicates the offset of the actual wavelength position relative to the central wavelength;

[0072] Considering the offset direction (positive offset or negative offset) and amplitude, positive and negative offset weights are introduced, and the expression for weight calculation is: , ,in, and It is a parameter that adjusts the offset sensitivity and controls the rate of change of the positive offset weight and the negative offset weight. and Represent the positive offset weight and negative offset weight respectively, represents the normalized offset, ;

[0073] Calculate the peak shift quantification value, the calculation expression is: , It represents the absolute value of the normalized wavelength offset, indicating that no matter whether the offset is positive or negative, the final offset takes its absolute value, indicating the degree of offset regardless of direction. It represents the difference between the positive offset weight and the negative offset weight, and quantifies the influence of the absorption peak offset direction and amplitude on anomaly detection. The difference between the positive and negative weights can more accurately reflect the overall effect of the offset. Represents the normalization coefficient, which is used to adjust the proportional relationship between the normalized offset and the actual measured wavelength. Represents the peak shift quantification value.

[0074] It can be seen from the peak shift quantification value that within the monitoring window, the greater the performance value of the peak shift quantification value generated after abnormal analysis of the shift of the absorption peak position (wavelength), the greater the hidden danger of the absorption peak causing a serious impact on the prediction accuracy of blood glucose concentration. This is because the peak shift index quantifies the degree and direction of the absorption peak shift. The larger the peak shift quantification value, the greater the shift of the absorption peak position relative to the expected wavelength, and the more significant the impact of the shift direction and amplitude, which will cause the relationship between the eigenvalue extracted by the model and the actual blood glucose concentration to be distorted, thereby reducing the prediction accuracy. Conversely, the smaller the peak shift quantification value, the smaller the shift of the absorption peak position, and the smaller the impact on the prediction accuracy of blood glucose concentration, so the hidden danger is smaller.

[0075] The low symmetry of the absorption peak usually indicates that the absorption peak is abnormal, which will have a serious impact on the prediction accuracy of blood glucose concentration. The absorption peak should be symmetrical, usually with a Gaussian or Lorentzian shape as the ideal. However, the low symmetry of the absorption peak may mean the presence of interfering signals, noise, or overlapping absorption peaks of other molecules. These anomalies will cause distortion of key information in the spectral data, making the data inaccurate during feature extraction and model training, which in turn affects the prediction results of blood glucose concentration. Since the blood glucose concentration prediction model relies on accurate spectral features, when the absorption peak is asymmetric, the model may not be able to correctly identify and quantify the absorption characteristics of glucose molecules, resulting in overestimation or underestimation of blood glucose concentration. This error not only affects individual health management and treatment decisions, but also may lead to serious clinical consequences, such as misdiagnosis and mistreatment, and increase the risk of acute complications (such as hypoglycemic coma or hyperglycemic ketoacidosis) and chronic complications (such as retinopathy, nephropathy, cardiovascular disease). Therefore, ensuring the symmetry of the absorption peak is a key step to improve the accuracy of blood glucose concentration prediction.

[0076] In the monitoring window, after the abnormal analysis of the symmetry of the absorption peak, the specific steps to generate the peak symmetry quantitative value are as follows:

[0077] Determine the highest point of the absorption peak, that is, the point with the highest absorbance, search to the left from the highest point of the absorption peak to find the point where the absorbance drops to the basic value (baseline level), mark this point as L, search to the right from the highest point of the absorption peak to find the point where the absorbance drops to the basic value, mark this point as R;

[0078] The peak is segmented, and a data set is established by taking the data points between L and the highest point of the absorption peak, and the data set is recorded as the left side data of the peak; and a data set is established by taking the data points between R and the highest point of the absorption peak, and the data set is recorded as the right side data of the peak;

[0079] The difference between the data on the left side of the peak and the data on the right side of the peak is calculated to compare the symmetry of the two sides. The calculation expression is: , where represents the difference value of the symmetric difference function at the i-th data point, represents the absorbance of the ith data point on the left side of the peak, Indicates the data on the right side of the peak The absorbance of the data points, n is the total number of data points on both sides, represents the symmetry difference function, Indicates the data on the left side of the peak, represents the data on the right side of the peak, and i represents the index of the data point;

[0080] The symmetry difference function is integrated to quantify the overall symmetry difference. The integral is calculated as: ,in, It indicates the degree of difference in the overall symmetry of the absorption peak;

[0081] Calculate the peak symmetry quantitative value, the calculation expression is: , where represents the quantitative value of peak symmetry, represents the absorbance function of the peak, x represents the wavelength, represents the total area of ​​the absorption peak;

[0082] Absorbance function It is a function that describes the change in absorbance of an absorption peak at different wavelengths x. In spectral analysis, absorbance indicates the degree to which light is absorbed by a sample, and it is related to the logarithmic ratio between the intensity of incident light and the intensity of transmitted light. Specifically, absorbance A is defined as: ,in is the incident light intensity, is the intensity of the transmitted light. Absorbance function It describes the absorbance values ​​corresponding to different wavelengths x during the spectral measurement process, thereby reflecting the shape and characteristics of the absorption peak.

[0083] It can be seen from the peak symmetry quantification value that within the monitoring window, the greater the performance value of the peak symmetry quantification value generated after abnormal analysis of the symmetry of the absorption peak, the greater the potential risk of the absorption peak causing a serious impact on the prediction accuracy of blood glucose concentration. Specifically, a larger peak symmetry quantification value indicates that the symmetry of the absorption peak is poor, the difference in absorbance between the left and right sides is significant, and there may be noise or interference signals, thereby increasing the risk of the prediction model misjudging the blood glucose concentration. Conversely, a smaller peak symmetry quantification value indicates that the symmetry of the absorption peak is better, the data quality is high, and the potential risk of predicting blood glucose concentration is smaller, and the model can use these data more accurately for prediction.

[0084] The machine learning model is not specifically limited here, and can achieve the quantification of peak shift Sum peak symmetry quantification value Perform comprehensive analysis to generate screening coefficients In order to realize the technical solution of the present invention, the present invention provides a specific implementation method;

[0085] Screening coefficient The resulting calculation formula is: , where , The peak shift quantification values ​​are Sum peak symmetry quantification value The preset scaling factor of , Both are greater than 0.

[0086] It can be seen from the screening coefficient that, under the monitoring window, the larger the performance value of the peak offset quantization value generated after the abnormal analysis of the offset of the absorption peak position (wavelength), the larger the performance value of the peak symmetry quantization value generated after the abnormal analysis of the symmetry of the absorption peak, that is, the larger the performance value of the screening coefficient generated under the monitoring window, the greater the potential risk that the absorption peak will cause a serious impact on the prediction accuracy of blood glucose concentration, and vice versa.

[0087] The screening coefficient generated after abnormal analysis of each absorption peak in the monitoring window is compared with the preset screening coefficient reference threshold to preliminarily determine whether there is an abnormality in the absorption peak and screen out the abnormal absorption peak. The specific steps are as follows:

[0088] If the screening coefficient is greater than or equal to the screening coefficient reference threshold, the absorption peak is judged as an abnormal absorption peak and the absorption peak is screened out;

[0089] If the screening coefficient is less than the screening coefficient reference threshold, the absorption peak is judged as a normal absorption peak and is retained.

[0090] Include all absorption peaks that have not been screened out in a circular area to ensure that all relevant spectral data are included in the analysis range, and further divide the circular area into several sub-areas of the same size so that the absorption peak data in each sub-area is relatively independent;

[0091] The step of including all absorption peaks that have not been screened out in a circular area and further dividing the area into several sub-areas of the same size so that the absorption peak data in each sub-area are relatively independent is used to analyze the spectral data in a systematic and refined manner. By defining a circular area, it is ensured that all relevant absorption peak data are included in the analysis range to avoid missing any important spectral information. Dividing the area into multiple independent sub-areas can disperse and reduce the complexity of data processing. The independent analysis of each sub-area helps to accurately identify and quantify local absorption features. The division of sub-areas also facilitates parallel processing and calculation, improving data processing efficiency.

[0092] After the sub-regions are divided, the absorption peaks in each sub-region are analyzed in detail, and the blood glucose concentration in each sub-region is predicted using the population big data model based on the absorption peak characteristics in each sub-region;

[0093] Using the absorption peak characteristics in each sub-region, the group big data model is used to predict the blood glucose concentration in the sub-region. The specific steps are as follows:

[0094] After the sub-region division, the spectral data in each sub-region is first preprocessed;

[0095] This step includes denoising, baseline correction and normalization. Denoising is to eliminate random noise and interfering signals, baseline correction is to eliminate instrument drift and background signals, and normalization is to adjust the spectral data to a uniform scale range to facilitate subsequent feature extraction and analysis.

[0096] After preprocessing, key features are extracted from the absorption peak data in each sub-region;

[0097] These features include the position (wavelength), intensity (absorbance), peak width (full width at half maximum, FWHM) and peak shape of the absorption peak. By extracting these features, the main change trends and characteristics of the spectral data in each sub-region can be captured, providing basic data for subsequent analysis.

[0098] Aggregate the feature data in each sub-region so as to input it into the group big data model;

[0099] Aggregation methods may include weighted average of features, combination of feature vectors, etc. Through feature aggregation, the spectral information in the sub-region is integrated to form a unified feature vector representing the spectral characteristics of the sub-region.

[0100] Using the aggregated feature data, a group big data model is used to predict blood glucose concentration;

[0101] The group big data model can be a machine learning model such as regression analysis, neural network, support vector machine, etc. By training and optimizing these models, accurate prediction of blood glucose concentration in each sub-region can be achieved. During the selection and training of the model, a large amount of historical data needs to be used for verification and tuning to improve the prediction accuracy and robustness.

[0102] The predicted blood glucose concentration values ​​of all sub-regions are established into an analysis set, and the difference analysis of the blood glucose concentrations in the analysis set is performed to evaluate the differences in the predicted values ​​between the sub-regions. If it is found that the predicted values ​​of some regions are obviously inconsistent with the overall trend, the abnormal predicted blood glucose concentration values ​​in the analysis set are screened out based on the difference analysis;

[0103] The predicted values ​​of blood glucose concentration in all sub-regions are integrated into an analysis set. There are m sub-regions, and the predicted values ​​of each sub-region are , , , , , the predicted values ​​of all sub-regions are combined into a vector C, then: ,in, represents the predicted value of blood glucose concentration in the kth sub-region;

[0104] Calculate global statistics of the analysis set C, including the median and interquartile range , median is the median value of the sorted set, the interquartile range is the upper quartile and lower quartile The difference is used to measure the degree of dispersion of data;

[0105] The interquartile range is a measurement method used in statistics to describe the distribution of data, representing the range of the middle 50% of the data set. The upper quartile is the data point ranked at the 75th percentile, while the lower quartile is the data point ranked at the 25th percentile. By calculating the difference between these two values, you can understand the degree of variation in the data in the middle part. The larger the interquartile range, the greater the dispersion of the data, and vice versa, the more concentrated the data.

[0106] Use global statistics to perform difference analysis and calculate the predicted value for each sub-region With median The absolute difference between , the calculation expression is: , where Represents the absolute deviation between the predicted value of the kth sub-region and the median;

[0107] The absolute deviation Standardized using the interquartile range To measure the relative difference, the expression of standardized absolute deviation is: , where represents the standardized deviation of the k-th sub-region, indicating the relative difference;

[0108] It should be noted that the larger the standardized deviation of the kth sub-region, the worse the prediction accuracy of the sub-region; the smaller the standardized deviation of the kth sub-region, the better the prediction accuracy of the sub-region. Reflects the predicted value of the sub-region Relative to the overall median The degree of deviation was taken into account, and the discreteness of the data was taken into account (through the interquartile range Measure). The larger Indicates that the predicted value of the sub-region deviates significantly from the overall trend and may be affected by abnormal data or noise, thereby reducing the prediction accuracy; while smaller It means that the predicted value of this sub-region is consistent with the overall trend, the data quality is high, and the prediction accuracy is better. , which can effectively evaluate and improve the prediction accuracy of each sub-region.

[0109] The standardized deviation of the kth sub-region is compared and analyzed with the pre-set deviation reference threshold. If the standardized deviation is greater than or equal to the deviation reference threshold, the predicted value of the sub-region is marked as abnormal and the predicted value of the sub-region is screened out. If the standardized deviation is less than the deviation reference threshold, the predicted value of the sub-region is marked as normal and the predicted value of the sub-region is retained.

[0110] The predicted blood glucose concentration values ​​after the abnormal data are screened out are averaged to calculate the final predicted blood glucose concentration;

[0111] The remaining normal prediction values ​​after filtering out the abnormalities are set as ,but: , where represents the predicted value after filtering out abnormal data, represents the total number of predicted values ​​after filtering out abnormal data, and the calculation expression for the final predicted blood glucose concentration is: , where represents the final predicted blood glucose concentration.

[0112] The present invention performs preliminary feature extraction on the acquired absorption peaks, performs abnormal analysis on these features within the monitoring window, and inputs the analysis results into a pre-trained machine learning model to preliminarily determine whether there are abnormalities in the absorption peaks, and screen out abnormal absorption peaks, ensuring that only normal spectral data that accurately reflects the characteristics of glucose molecules are used for analysis. In this way, the blood glucose concentration prediction error caused by abnormal absorption peaks is reduced, the overall prediction accuracy is improved, and the risk of overestimating or underestimating the blood glucose concentration is reduced, thereby providing more reliable blood glucose monitoring for diabetic patients, ensuring that diabetic patients can adjust drug dosages in a timely and accurate manner, and reducing the risk of acute and chronic complications.

[0113] After the present invention establishes the predicted blood glucose concentration values ​​of all sub-regions into an analysis set, it performs difference analysis on the blood glucose concentrations in the set and evaluates the differences in the predicted values ​​between the sub-regions. It can timely discover and identify those abnormal regions that are obviously inconsistent with the overall trend, and screen out the predicted abnormal blood glucose concentration values ​​by calculating the standardized deviation of each sub-region, thereby further improving the accuracy of blood glucose concentration prediction. Through this method, not only the negative impact on the overall result due to the poor prediction accuracy of the local area is reduced, but also the data processing flow can be optimized, and the robustness and reliability of the system can be improved. Finally, the predicted blood glucose concentration value after screening out the abnormal data is calculated by taking the average value to obtain the final predicted blood glucose concentration. This refined processing ensures the stability and accuracy of the final result, and provides a more accurate and reliable non-invasive blood glucose monitoring solution for diabetic patients.

[0114] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0115] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0116] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A non-invasive blood glucose testing method based on a group big data model, characterized in that: The following steps are involved: The spectral data in the skin and tissues are detected by a near-infrared spectrometer to obtain the absorption peak of glucose molecules in the near-infrared range; Perform preliminary feature extraction on the acquired absorption peaks, and after abnormal analysis of the extracted features, preliminarily determine whether the absorption peaks are abnormal, and filter out the abnormal absorption peaks; Include all absorption peaks that have not been screened out in a circular area to ensure that all relevant spectral data are included in the analysis range, and further divide the circular area into several sub-areas of the same size so that the absorption peak data in each sub-area is relatively independent; After the sub-regions are divided, the absorption peaks in each sub-region are analyzed in detail, and the blood glucose concentration in each sub-region is predicted using the population big data model based on the absorption peak characteristics in each sub-region; The predicted blood glucose concentration values ​​of all sub-regions are established into an analysis set, and the difference analysis of the blood glucose concentrations in the analysis set is performed to evaluate the differences in the predicted values ​​between the sub-regions. If it is found that the predicted values ​​of some regions are obviously inconsistent with the overall trend, the abnormal predicted blood glucose concentration values ​​in the analysis set are screened out based on the difference analysis; The predicted blood glucose concentration values ​​after screening out abnormal data were averaged to calculate the final predicted blood glucose concentration.

2. A non-invasive blood glucose testing method based on a group big data model according to claim 1, characterized in that: To obtain the absorption peak of glucose molecules in the near-infrared range, the specific steps are as follows: Before the actual measurement, a set of background spectrum data is collected; Aim the light source of the near-infrared spectrometer at the skin or tissue sample and start the spectrum measurement program. The near-infrared light emitted by the light source penetrates the skin and tissue, part of the light is absorbed by the glucose molecules, and the other part is scattered and reflected. The detector of the spectrometer receives the light after passing through the sample and records the light absorption data of different wavelengths. After collecting the spectral data of the sample, first, compare the sample spectral data with the background spectral data to eliminate background interference and noise. Then, use spectral processing software to extract the absorption peak characteristics in the spectrum. Finally, through feature analysis, identify the absorption peak related to glucose molecules to provide basic data for subsequent blood glucose concentration prediction.

3. The non-invasive blood glucose testing method based on a group big data model according to claim 1, characterized in that: A preliminary feature extraction is performed on the acquired absorption peak, and the extracted features include the offset of the absorption peak position and the symmetry of the absorption peak. Within the monitoring window, after anomaly analysis of the offset of the absorption peak position and the symmetry of the absorption peak, peak offset quantification values ​​and peak symmetry quantification values ​​are generated respectively. The peak offset quantification values ​​and peak symmetry quantification values ​​generated after the anomaly analysis are input into a pre-trained machine learning model, and a screening coefficient is generated by the machine learning model. The screening coefficient is used to preliminarily determine whether there is an abnormality in the absorption peak.

4. The non-invasive blood glucose testing method based on a group big data model according to claim 3, characterized in that: The screening coefficient generated after abnormal analysis of each absorption peak in the monitoring window is compared with the preset screening coefficient reference threshold to preliminarily determine whether there is an abnormality in the absorption peak and screen out the abnormal absorption peak. The specific steps are as follows: If the screening coefficient is greater than or equal to the screening coefficient reference threshold, the absorption peak is judged as an abnormal absorption peak and the absorption peak is screened out; If the screening coefficient is less than the screening coefficient reference threshold, the absorption peak is judged as a normal absorption peak and is retained.

5. The non-invasive blood glucose testing method based on a group big data model according to claim 1, characterized in that: Using the absorption peak characteristics in each sub-region, the group big data model is used to predict the blood glucose concentration in the sub-region. The specific steps are as follows: After the sub-region division, the spectral data in each sub-region is first preprocessed; After preprocessing, key features are extracted from the absorption peak data in each sub-region; Aggregate the feature data in each sub-region so as to input it into the group big data model; Using the aggregated feature data, a group big data model is used to predict blood glucose concentration.

6. The non-invasive blood glucose testing method based on a group big data model according to claim 1, characterized in that: The predicted values ​​of blood glucose concentration in all sub-regions are integrated into an analysis set. There are m sub-regions, and the predicted values ​​of each sub-region are , , , , , the predicted values ​​of all sub-regions are combined into a vector C, then: ,in, represents the predicted value of blood glucose concentration in the kth sub-region; Calculate global statistics of the analysis set C, including the median and interquartile range , median is the median value of the sorted set, the interquartile range is the upper quartile and lower quartile The difference is used to measure the degree of dispersion of data; Use global statistics to perform difference analysis and calculate the predicted value for each sub-region With median The absolute difference between , the calculation expression is: , where It represents the absolute deviation between the predicted value of the kth sub-region and the median; The absolute deviation Standardized using the interquartile range To measure the relative difference, the expression of standardized absolute deviation is: , where represents the standardized deviation of the k-th sub-region, indicating the relative difference; The standardized deviation of the kth sub-region is compared and analyzed with the pre-set deviation reference threshold. If the standardized deviation is greater than or equal to the deviation reference threshold, the predicted value of the sub-region is marked as abnormal and the predicted value of the sub-region is screened out. If the standardized deviation is less than the deviation reference threshold, the predicted value of the sub-region is marked as normal and the predicted value of the sub-region is retained.

7. The non-invasive blood glucose testing method based on a group big data model according to claim 6, characterized in that: The remaining normal prediction values ​​after filtering out the abnormalities are set as ,but: , where represents the predicted value after filtering out abnormal data. represents the total number of predicted values ​​after filtering out abnormal data, and the calculation expression for the final predicted blood glucose concentration is: , where represents the final predicted blood glucose concentration.

8. The non-invasive blood glucose testing method based on a group big data model according to claim 3, characterized in that: After abnormal analysis of the shift of the absorption peak position in the monitoring window, the specific steps for generating the peak shift quantification value are as follows: Determine the reference wavelength range of the absorption peak of glucose molecules , and Respectively represent the minimum and maximum values ​​of the reference wavelength range; In the monitoring window, the actual wavelength position is measured and the actual wavelength position is used express; Calculate the offset of the actual wavelength position relative to the central wavelength. The calculation expression is: , where represents the center wavelength of the reference wavelength range, which is the expected peak position. , Indicates the offset of the actual wavelength position relative to the central wavelength; Considering the offset direction and amplitude, positive and negative offset weights are introduced, and the expression for weight calculation is: , ,in, and It is a parameter that adjusts the offset sensitivity and controls the rate of change of the positive offset weight and the negative offset weight. and Represent the positive offset weight and negative offset weight respectively, represents the normalized offset, ; Calculate the peak shift quantification value, the calculation expression is: , represents the absolute value of the normalized wavelength offset, Represents the difference between the positive bias weight and the negative bias weight, Represents the normalization coefficient, which is used to adjust the proportional relationship between the normalized offset and the actual measured wavelength. Represents the peak shift quantification value.

9. The non-invasive blood glucose testing method based on a group big data model according to claim 3, characterized in that: In the monitoring window, after the abnormal analysis of the symmetry of the absorption peak, the specific steps to generate the peak symmetry quantitative value are as follows: Determine the highest point of the absorption peak, that is, the point with the highest absorbance, search to the left from the highest point of the absorption peak to find the point where the absorbance drops to the base value, mark this point as L, search to the right from the highest point of the absorption peak to find the point where the absorbance drops to the base value, mark this point as R; The peak is segmented, and a data set is established by taking the data points between L and the highest point of the absorption peak, and the data set is recorded as the left side data of the peak; and a data set is established by taking the data points between R and the highest point of the absorption peak, and the data set is recorded as the right side data of the peak; The difference between the data on the left side of the peak and the data on the right side of the peak is calculated to compare the symmetry of the two sides. The calculation expression is: , where represents the difference value of the symmetric difference function at the i-th data point, represents the absorbance of the ith data point on the left side of the peak, Indicates the data on the right side of the peak The absorbance of the data points, n is the total number of data points on both sides, represents the symmetry difference function, Indicates the data on the left side of the peak, represents the data on the right side of the peak, and i represents the index of the data point; The symmetry difference function is integrated to quantify the overall symmetry difference. The integral is calculated as: ,in, It indicates the degree of difference in the overall symmetry of the absorption peak; Calculate the peak symmetry quantitative value, the calculation expression is: , where represents the quantitative value of peak symmetry, represents the absorbance function of the peak, x represents the wavelength, Represents the total area of ​​the absorption peak.

Images