Diabetes metabolism risk modeling analysis system driven by physiological signals

Abstract

The invention relates to the technical field of intelligent health monitoring, in particular to a physiological signal driven diabetes mellitus metabolism risk modeling analysis system which comprises a variable sorting module, an electrocardio shearing module, a distribution edge forming module, a risk modeling module and a regulation and control deconstruction module. According to the method, the fluctuation response time difference between the skin temperature and the blood oxygen signal is obtained, distribution density analysis is conducted on the time difference, the samples are divided into response difference groups, and layered recognition of the signal time dynamic characteristics is achieved; a peak-valley amplitude sequence in the electrocardiogram periodic signal is further extracted, an amplitude withdrawing ratio is calculated, a key turning point is identified according to a threshold value, and the corresponding periodic signal is divided into different electrocardiogram segments, so that the micro electrocardiogram fluctuation response is effectively positioned; on the basis, according to the time index, the asynchronous distribution section of the skin temperature and the blood oxygen is backtracked, a cross boundary graph between the asynchronous signals is constructed, and the asymmetric response mode of skin temperature and blood oxygen changes can be analyzed in a refined mode.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent health monitoring, and particularly to a diabetes metabolic risk modeling and analysis system driven by physiological signals. Background Art

[0002] The technical field of intelligent health monitoring includes various systems and methods for real-time collection, analysis, and modeling of human physiological parameters using information technology, biosensing technology, and artificial intelligence methods. The core content of this technical field is to use multi-source physiological data as input, and through means such as data fusion and feature construction, quantitatively evaluate the individual health status and predict risks, thereby supporting personalized medical intervention and chronic disease management.

[0003] Among them, a diabetes metabolic risk modeling and analysis system driven by physiological signals refers to a system that quantitatively models the individual metabolic risk status based on multi-dimensional physiological signal data. This system mainly focuses on the state recognition and risk level assessment of individuals at high risk of diabetes, including: establishing a sequence of individual physiological state indicators based on electrophysiological signals such as electrocardiogram and skin electrical response; constructing an individual metabolic state feature set according to physiological dimensions such as respiratory rate, skin temperature, and blood oxygen saturation; using a supervised learning model to perform sample clustering and classification discrimination on the extracted features to generate diabetes metabolic risk assessment labels; and finally extracting relevant feature variables from historical population data based on statistical learning rules and completing the model training process.

[0004] Traditional methods mainly rely on the construction of static feature vectors and fixed-dimensional parameter combinations, and have insufficient recognition accuracy in the time-series differences of signal fluctuations between samples; due to the lack of in-depth analysis of the signal response time-series structure, the microscopic response delay between skin temperature and blood oxygen is often ignored, affecting the discovery efficiency of abnormal patterns. For example, the asymmetric alternation between a rapid increase in skin temperature and a slow decrease in blood oxygen often cannot be captured by static features, resulting in an underestimation of early abnormalities. In addition, electrocardiogram signals are equally processed in classification, and the signal features of their retraction amplitude in metabolic fluctuations are not effectively utilized, resulting in some samples with significant physiological abnormalities being classified into lower-risk categories, affecting the accuracy of model judgment. For samples in the risk marginal state, existing methods lack a targeted structural adjustment mechanism in boundary recognition and sample trend analysis, resulting in frequent fluctuations of these samples in the classification marginal zone, causing model instability, and further reducing the overall application credibility and risk warning efficiency. The above limitations cause problems such as weak model generalization ability, large deviation in early risk recognition, and fuzzy processing of marginal states in practical applications. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art and propose a diabetes metabolic risk modeling and analysis system driven by physiological signals.

[0006] To achieve the above object, the present invention adopts the following technical solutions: a physiological signal-driven diabetes metabolic risk modeling and analysis system, the system comprising: a variable sorting module, which monitors the skin temperature and blood oxygen saturation signals of diabetic patients, obtains the fluctuation response time difference of each pair of signals in the sample, divides the sample into multiple response difference groups, and generates a diabetes physiological signal grouping result; An electrocardiogram shearing module, based on the sorted time period in the diabetes physiological signal grouping result, obtains the peak-to-valley amplitude of each cycle and calculates the amplitude drawdown rate, marks the cutting points exceeding the drawdown threshold, and shears them into corresponding electrocardiogram sub-segments, generating a set of drawdown-marked electrocardiogram fluctuation sub-segments; A distribution border construction module, obtains the electrocardiogram sub-segment numbers in the set of drawdown-marked electrocardiogram fluctuation sub-segments, extracts the cross features of skin temperature and blood oxygen distribution, and establishes an asynchronous physiological signal cross-border map; A risk modeling module, obtains the skin temperature and blood oxygen cross-border structure defined in the asynchronous physiological signal cross-border map, sets cross-region markers, and establishes a metabolic risk classification model for distinguishing asynchronous mutation paths.

[0007] The present invention is improved in that the diabetes physiological signal grouping result includes the skin temperature first fluctuation time point index, the blood oxygen first fluctuation time point index, the fluctuation response time difference delay sequence, and the response difference grouping label. The set of drawdown-marked electrocardiogram fluctuation sub-segments is specifically the electrocardiogram sub-segment number, the amplitude drawdown rate parameter sequence, and the drawdown threshold trigger position mark. The asynchronous physiological signal cross-border map includes the skin temperature distribution map, the blood oxygen distribution map, the cross-point density slope, the distribution overlap region, and the distribution peak offset direction. The metabolic risk classification model includes the skin temperature change amplitude, the blood oxygen cycle shortening amount, the electrocardiogram drawdown rate, the cross-region marker, and the target mapping relationship.

[0008] The present invention is improved in that the variable sorting module includes: A fluctuation detection sub-module, which obtains the skin temperature and blood oxygen saturation signals of diabetic patients, based on the sampling data within a fixed time window, detects the time point corresponding to the first continuous positive slope segment in the skin temperature rate change curve as the skin temperature first fluctuation time point, and detects the starting time point of the continuous compression nodes in the blood oxygen steady cycle as the blood oxygen first fluctuation time point, and respectively calibrates the index positions of the two types of signal fluctuation time points on the unified time axis, generating a time point index sequence; A time difference calculation sub-module, which calls the respective time indexes of skin temperature and blood oxygen in the time point index sequence, calculates the time interval difference between the skin temperature first fluctuation time point and the corresponding blood oxygen first fluctuation time point for each group of sample signal data, writes the difference as a sorting factor into the unified sequence, and establishes a delayed response order sequence according to the numerical sorting of the time interval differences in the sequence; A response grouping sub-module determines the distribution aggregation of each sample within a unified sorting interval according to the numerical range of the time interval differences in the delayed response order sequence, calls the adjacent distribution density of each sorting factor within the overall interval, divides the continuous distribution interval by means of density comparison, and uses the interval boundary difference value as the clustering division criterion to obtain the grouping result of diabetic physiological signals.

[0009] The improvement of the present invention is that the electrocardiogram shearing module includes: A signal extraction sub-module extracts the electrocardiogram cycle signals synchronously collected within each time period based on the sorting time periods recorded in the grouping result of diabetic physiological signals, calls the voltage change data recorded within each cycle, locates the maximum voltage peak point and the minimum voltage valley point within each cycle, extracts the continuous amplitude values between adjacent peaks and valleys and arranges them in chronological order to generate a peak-valley amplitude sequence. A retracement ratio calculation sub-module calculates the ratio of the retracement value to the corresponding growth value by calling the amplitude growth value from peak to valley within the previous cycle according to the amplitude retracement value from peak to valley in each segment of the peak-valley amplitude sequence, and sets each group of ratios as retracement rate parameters and records them in a unified parameter sequence in turn to establish a retracement rate parameter sequence. A segment division sub-module calls the parameter values in the retracement rate parameter sequence, compares them with the set amplitude retracement rate threshold, determines the time points where the ratios exceeding the retracement rate threshold are located, sets the target time points as vector cutting points, calls the electrocardiogram cycle signals corresponding within the sorting time period, performs signal segmentation operations according to the cutting point positions, records each segment of the signal as an independent sub-segment and numbers them to obtain a set of retracement-marked electrocardiogram fluctuation sub-segments.

[0010] The improvement of the present invention is that the distribution border construction module includes: An identification tracing sub-module obtains the electrocardiogram sub-segment numbers recorded in the set of retracement-marked electrocardiogram fluctuation sub-segments, calls the skin temperature fluctuation record data and the blood oxygen cycle record data based on the time index intervals corresponding to the numbers, and extracts the skin temperature variation range and the blood oxygen fluctuation cycle that overlap with the time periods of the electrocardiogram sub-segment numbers respectively to obtain the corresponding skin temperature and blood oxygen segments. An intersection extraction sub-module constructs a skin temperature skewed distribution map and a blood oxygen skewed distribution map respectively according to the skin temperature variation data and the blood oxygen cycle variation data in the corresponding skin temperature and blood oxygen segments, obtains the position of the intersection point of the two distribution maps, calculates the density change slope of the intersection point area and judges the overlapping range area between regions, as well as the offset direction and distance between the two distribution peaks to obtain the parameter distribution intersection characteristics. A structure building sub-module, which calls the cross-point density slope, regional overlap area, and peak offset direction in the parameter distribution cross-feature, determines the boundary points of the asynchronous response interval between skin temperature and blood oxygen distribution according to the slope change trend and the regional position drift amplitude, constructs a parameter cross-relationship boundary structure on the unified time axis, and establishes an asynchronous physiological signal cross-boundary map.

[0011] The improvement of the present invention is that the risk modeling module includes: A factor construction sub-module, which obtains the skin temperature and blood oxygen cross-boundary structure delimited in the asynchronous physiological signal cross-boundary map, calls the electrocardiogram segment numbers in the associated drawdown marked electrocardiogram fluctuation sub-segments, extracts the corresponding amplitude drawdown rate parameters, skin temperature amplitude change rate, and blood oxygen cycle compression duration according to the numbers, combines the three parameters into the same structure, and generates a physiological signal association set; A target marking sub-module, based on each group of signals in the physiological signal association set, sets corresponding cross-region marks according to the time period range corresponding to the signal cross-boundary, adds the marks as target factors to the signal combination, and establishes the mapping attribute between the target factors and the signal features to obtain a cross-marked target factor set; A path classification sub-module, according to the target factors and corresponding signal features recorded in the cross-marked target factor set, constructs the corresponding relationship between the target factors and the sample set, and classifies and segments the path arrangement forms of all target factor combination methods and numerical structures inside the sample set to establish a metabolic risk classification model.

[0012] The improvement of the present invention also includes a regulation and deconstruction module. The regulation and deconstruction module extracts the classification results of the sample paths in the metabolic risk classification model, redefines the boundary interval adjustment logic of the model output categories, performs an asymmetric structure split on the classification boundary, and generates a trend risk boundary adjustment result; The trend risk boundary adjustment result specifically refers to the sample numbers in the risk marginal area, the trend variation feature group, the trend starting rhythm, the number of amplitude turning points, the signal start sequence, the continuous intensity comparison, and the updated classification boundary structure.

[0013] The improvement of the present invention is that the regulation and deconstruction module includes: An edge recognition sub-module, based on the classification results of all sample paths in the metabolic risk classification model, screens and determines the sample numbers in the risk marginal area, extracts the amplitude drawdown rate mutation points in the electrocardiogram band corresponding to the numbers, and synchronously obtains the skin temperature fluctuation rate and blood oxygen cycle change direction in the time intervals before and after the band to establish a risk section positioning result; The trend construction submodule calls the signal fragments in the risk segment positioning results, reconstructs the complete skin temperature, blood oxygen and ECG trend sequence under each sample number, calculates the starting rhythm and amplitude turning number of each trend, and obtains the trend variation feature indicator set; The boundary adjustment submodule determines the classification boundary interval into which the feature combination falls based on the comparison between the signal start-up sequence and the continuous intensity parameter in the trend variation feature indicator set, redefines the boundary demarcation rules for the category corresponding to each group of combinations, performs asymmetric structural splitting of the original classification boundary, and generates a trend risk boundary adjustment result.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by obtaining the fluctuation response time difference between skin temperature and blood oxygen signal and performing distribution density analysis on the time difference, the samples are divided into response difference groups, thereby realizing hierarchical recognition of the temporal dynamic characteristics of the signal; further extracting the peak-to-valley amplitude sequence in the ECG cycle signal, calculating the amplitude retracement ratio and identifying the key turning point based on the threshold, the corresponding periodic signal is divided into different ECG sub-segments, so that the tiny ECG fluctuation response is effectively located; on this basis, the asynchronous distribution segments of skin temperature and blood oxygen are traced back according to the time index, and their cross-features are extracted. The density change slope, distribution overlap area and other factors are used to identify the key turning points. The domain and peak offset parameters are used to construct the cross-boundary map between asynchronous signals, which can refine the analysis of the asymmetric response pattern of skin temperature and blood oxygen changes; such cross-boundaries are fused with the cardiac segments of the retracement mark to form a signal-related feature set, and the precise modeling of the temporal relationship between signals is achieved through target mapping, making the differentiation of metabolic risk paths more targeted and high-resolution; at the same time, the trend reconstruction and boundary splitting of the sample paths in the risk edge area are carried out, and the classification intervals are redefined through variables such as the starting rhythm and the continuous intensity of the signal, which enhances the model's adaptability to samples in the edge state and reduces the misjudgment rate. In the whole process, multi-dimensional innovative participants such as signal response time difference, peak-valley retracement characteristics, distribution cross parameters and trend variation structure are deeply integrated to form a high-precision and highly plastic metabolic risk modeling path, which improves the system's sensitive recognition of metabolic abnormalities and the dynamic adjustment ability of risk classification. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a system module diagram of the present invention; Figure 2 It is a system framework diagram of the present invention; Figure 3 A schematic diagram of a variable sorting module of the present invention; Figure 4 is a schematic diagram of the ECG shearing module of the present invention; Figure 5 It is a schematic diagram of the distributed edge construction module of the present invention; Figure 6 Schematic diagram of the risk modeling module of the present invention; Figure 7 Schematic diagram of the regulation and deconstruction module of the present invention. Specific implementation manners

[0016] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0017] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more unless otherwise specifically defined; Please refer to Figure 1 , the present invention provides a technical solution: a diabetes metabolic risk modeling and analysis system driven by physiological signals: The system includes: a variable sorting module, which monitors the skin temperature and blood oxygen saturation signals of diabetic patients, obtains the fluctuation response time difference of each pair of signals in the sample, divides the sample into multiple response difference groups, and generates a diabetes physiological signal grouping result; An electrocardiogram shearing module, which obtains the peak-valley amplitude of each cycle and calculates the amplitude retracement rate based on the sorted time period in the diabetes physiological signal grouping result, marks the cutting points exceeding the retracement threshold, and shears them into corresponding electrocardiogram sub-segments to generate a set of retracement-marked electrocardiogram fluctuation sub-segments; A distribution border construction module, which obtains the electrocardiogram sub-segment numbers in the set of retracement-marked electrocardiogram fluctuation sub-segments, extracts the cross features of skin temperature and blood oxygen distribution, and establishes an asynchronous physiological signal cross-border map; A risk modeling module, which obtains the skin temperature and blood oxygen cross-border structure defined in the asynchronous physiological signal cross-border map, sets cross-region markers, and establishes a metabolic risk classification model for distinguishing asynchronous mutation paths; The grouping results of diabetic physiological signals include the index of the first fluctuation time point of skin temperature, the index of the first fluctuation time point of blood oxygen, the delay sequence of the fluctuation response time difference, the response difference grouping label. The specific content of the retraction marked electrocardiogram fluctuation sub-segment set is the electrocardiogram sub-segment number, the amplitude retraction rate parameter sequence, and the retraction threshold trigger position mark. The asynchronous physiological signal cross-boundary map includes the skin temperature distribution map, the blood oxygen distribution map, the cross-point density slope, the distribution overlap area, and the distribution peak offset direction. The metabolic risk classification model includes the skin temperature change amplitude, the blood oxygen cycle shortening amount, the electrocardiogram retraction rate, the cross-region mark, and the target mapping relationship.

[0018] Please refer to Figure 2 and Figure 3 , the variable sorting module includes: The fluctuation detection sub-module obtains the skin temperature and blood oxygen saturation signals of diabetic patients. Based on the sampling data within a fixed time window, it detects the time point corresponding to the first continuous positive slope segment in the skin temperature rate change curve as the first fluctuation time point of skin temperature, and detects the starting time point of the continuous compression nodes in the blood oxygen stable period as the first fluctuation time point of blood oxygen. It respectively calibrates the index positions of the two types of signal fluctuation time points on the unified time axis to generate a time point index sequence; First, collect the skin temperature data and blood oxygen saturation data continuously recorded by multiple groups of subjects in the resting state. The two signals are synchronously recorded at a sampling interval of 0.5 seconds to form two columns of original time series signals. For the skin temperature signal, first calculate the temperature change rate at each time point. For example, assume that the skin temperature at the first time point is 30°C, the second time point is 30.02°C, and the third time point is 30.04°C. The rates are: (30.02 - 30) / 0.5 = 0.04°C / s, (30.04 - 30.02) / 0.5 = 0.04°C / s. If the temperature rates of three consecutive sampling points are all positive and conform to the positive change, then the first positive change point is taken as the first fluctuation time point of skin temperature. For example, the 240th sampling point is the first fluctuation time point of skin temperature. For the blood oxygen signal, assume that the blood oxygen values of 5 consecutive samples are 96.5%, 96.3%, 96.0%, 95.9%, and 95.8%, and the amplitude changes are all less than 0.5% and the time is less than 2 seconds. Then this segment is regarded as a compression node. If the starting point of this segment is the 258th sampling point, then this point is taken as the first fluctuation time point of blood oxygen. Finally, with the first fluctuation time point of skin temperature being the 240th sampling point and the first fluctuation time point of blood oxygen being the 258th sampling point, a corresponding time point index sequence is generated.

[0019] The time difference calculation sub-module calls the respective time indices of skin temperature and blood oxygen in the time point index sequence, calculates the time interval difference between the first fluctuation time point of skin temperature and the first fluctuation time point of the corresponding blood oxygen for the signal data within each group of samples, writes the difference as a sorting factor into the unified sequence, and establishes a delay response order sequence according to the numerical sorting of the time interval differences in the sequence; Calculate the time interval between the first fluctuation time point of skin temperature and the first fluctuation time point of blood oxygen for each sample. Suppose in sample A, the first fluctuation time point of skin temperature is the 240th sampling point, and the first fluctuation time point of blood oxygen is the 258th sampling point. Then the time difference between them is 258 - 240 = 18 sampling points. Considering that the interval between each sampling point is 0.5 seconds, the time difference is 18 * 0.5 = 9 seconds. Take the time difference of each sample as a sorting factor and put it into the unified sequence. For example, the time difference of sample C is 3 seconds, the time difference of sample B is 7 seconds, and the time difference of sample A is 9 seconds. The delay response order sequence obtained by arranging them in ascending order is: sample C (3 seconds), sample B (7 seconds), sample A (9 seconds). This sorting process will provide basic data for subsequent density analysis and sample classification.

[0020] The response grouping sub-module judges the distribution aggregation situation of each sample within the unified sorting interval according to the numerical range of the time interval difference in the delay response order sequence, calls the adjacent distribution density of each sorting factor within the overall interval, divides the continuous distribution interval by density comparison means, and takes the interval boundary difference value as the clustering division benchmark to obtain the grouping result of diabetic physiological signals; Call the sorting position of each sample and the corresponding time difference, and perform density aggregation analysis on all samples on the sorting axis. Suppose the density analysis window is set to 5 samples, and gradually slide the window and calculate the change of the time difference between adjacent samples within the window. For example, the time difference sequence within a certain window is [5 seconds, 6 seconds, 5 seconds, 7 seconds, 6 seconds], and calculate its mean square deviation as seconds. If this value is lower than the set threshold of 2 seconds, it is determined as a high-density area. Next, identify the starting sample and the ending sample of this high-density area. For example, if the starting point of this high-density area is sample B and the ending point is sample C, then construct a new response distribution set. Through similar analysis of all samples and according to their density division boundaries, classify them into different response intervals, and finally obtain the grouping result of diabetic physiological signals.

[0021] Please refer to Figure 2 and Figure 4 , the electrocardiogram shearing module includes: The signal extraction sub-module extracts the electrocardiogram cycle signals synchronously collected within each time period based on the sorted time periods recorded in the diabetes physiological signal grouping results, calls the voltage change data recorded within each cycle, locates the maximum voltage peak point and the minimum voltage trough point within each cycle, extracts the continuous amplitude values between adjacent peaks and troughs and arranges them in chronological order to generate a peak-trough amplitude sequence; Extract the electrocardiogram cycle signals synchronously collected within each sorted time period. Assume that the signals of each electrocardiogram cycle within a sample time period are recorded within 1 second, the sampling frequency is 1000 Hz, and 1000 data points are collected. For the data within each electrocardiogram cycle, by analyzing the voltage change signal, first determine the positions of the maximum voltage peak and the minimum voltage trough. For example, assume that the signal reaches the maximum voltage peak of +1.5 mV at the 500th data point and reaches the minimum voltage trough of -1.0 mV at the 800th data point. Next, extract the amplitude change value between the maximum peak and the minimum trough, that is, the amplitude retracement value is 1.5 mV - (-1.0 mV) = 2.5 mV, and record the change in this section of the amplitude. After that, arrange the amplitude retracement values between all adjacent peaks and troughs in chronological order to generate a peak-trough amplitude sequence. This sequence contains the amplitude changes between all adjacent voltage peaks and troughs in this electrocardiogram cycle, such as: [2.5 mV, 3.0 mV, 2.2 mV, 3.5 mV], and organize these data into a complete amplitude sequence to provide a basis for subsequent analysis.

[0022] The retracement ratio calculation sub-module calculates the ratio of the retracement value to the corresponding growth value by calling the amplitude growth value from the peak to the trough in the previous cycle according to the amplitude retracement value from the peak to the trough in each section of the peak-trough amplitude sequence, and sets each group of ratios as the retracement rate parameter and records them in the unified parameter sequence in turn to establish a retracement rate parameter sequence; According to the extracted peak-trough amplitude sequence, calculate the amplitude retracement value from the peak to the trough in each section. Assume that the retracement amplitude from the peak to the trough in the first section of this sequence is 2.5 mV. Then, call the amplitude growth value from the peak to the trough in the previous cycle for ratio calculation. For example, assume that the amplitude growth value from the maximum peak to the minimum trough in the previous cycle is 3.0 mV. Then, the ratio of the retracement value to the growth value is retracement value / growth value = 2.5 mV / 3.0 mV = 0.83. This ratio is used to measure the degree of amplitude retracement of the current signal compared with the change in the previous cycle. Next, record the ratios of the amplitude retracement values from each peak-trough section to the amplitude growth value in the previous cycle in the unified retracement rate parameter sequence in turn, such as: [0.83, 0.75, 0.91, 0.79], and assign corresponding time points to each ratio to form a time series, preparing to provide basic data for subsequent segment division operations.

[0023] The segment division sub-module calls the parameter values in the drawdown rate parameter sequence, compares them with the set amplitude drawdown rate threshold, determines the time points where the ratio exceeding the drawdown rate threshold is located, sets the target time point as the vector cutting point, calls the corresponding electrocardiogram cycle signals within the sorted time period, performs signal segmentation operations according to the cutting point position, records each segment of the signal as an independent sub-segment and numbers it, and obtains the drawdown-marked electrocardiogram fluctuation sub-segment set; According to the ratio in the drawdown rate parameter sequence, it is compared with the set amplitude drawdown rate threshold. The setting of the drawdown rate threshold is based on the historical statistical analysis of sample data and domain experience. Usually, by analyzing the drawdown amplitudes of a large number of normal and abnormal electrocardiogram cycle data, a cut-off value is determined to distinguish normal signal fluctuations from fluctuations that may have abnormal changes. The setting of this threshold can be adjusted according to the actual scenario to ensure that abnormal electrocardiogram signals can be effectively identified. For example, assume that by analyzing the electrocardiogram signal data of a large number of diabetic patients, it is found that the drawdown rate is usually between 0.5 and 0.7 under normal circumstances, and when abnormal electrocardiogram fluctuations occur, the drawdown rate may rise above 0.8. Based on these data, the drawdown rate threshold can be set to 0.8. The setting of this threshold reflects the boundary between normal and abnormal fluctuations. When the ratio in the drawdown rate parameter sequence exceeds the set threshold, it means that there may be abnormal fluctuations in this segment of the signal, which requires attention and further analysis. In this way, the drawdown rate threshold helps to accurately distinguish normal fluctuations from abnormal fluctuations, ensuring that subsequent signal segmentation operations can effectively identify key physiological signal changes. Assume that the drawdown rate threshold is set to 0.8. When the ratio of a certain segment in the drawdown rate parameter sequence exceeds 0.8, this segment is considered as a signal section that needs attention. For example, if the ratio of a certain segment in the drawdown rate sequence is 0.85, then this segment exceeds the set drawdown rate threshold, and the time point corresponding to this ratio is set as the vector cutting point. Assume that the position of this time point is the 700th sampling point, then the signal segmentation operation will be performed at this time point, and the electrocardiogram signal will be segmented from this cutting point to form independent sub-segments. For example, if the signal is cut at the 700th sampling point, then the part before the 700th is the first signal sub-segment, and the part after the 700th is the second signal sub-segment. Each signal sub-segment will be numbered and recorded as part of the drawdown-marked electrocardiogram fluctuation sub-segment set. All segmented signal sub-segments will be numbered, and finally a complete drawdown-marked electrocardiogram fluctuation sub-segment set is obtained, which is convenient for subsequent analysis and processing.

[0024] Please refer to Figure 2 and Figure 5 , the distribution edge construction module includes: The number tracing sub-module obtains the electrocardiogram segment numbers recorded in the recalled marked electrocardiogram fluctuation sub-segments. Based on the time index interval corresponding to the numbers, it calls the skin temperature fluctuation record data and the blood oxygen cycle record data, extracts the skin temperature amplitude change interval and the blood oxygen fluctuation cycle that overlap with the electrocardiogram segment number time period respectively, and obtains the corresponding skin temperature and blood oxygen segments; First of all, it is necessary to determine the corresponding time interval according to the electrocardiogram segment number. For example, the time interval of a certain electrocardiogram fluctuation record may be from a certain moment T1 to T2. Then, through this time interval, the skin temperature amplitude change and blood oxygen fluctuation cycle data that overlap with it are searched. Suppose the time range covered by the skin temperature record data is from T3 to T4, and the cycle of the blood oxygen data is between T5 and T6. At this time, through the intersection operation, the skin temperature amplitude change and blood oxygen cycle time periods that overlap with the electrocardiogram segment number can be obtained. For example, it may be the intersection part of T3 to T4 and T5 to T6. These intersection parts are the skin temperature and blood oxygen data that need to be further extracted and analyzed. Through these overlapping segments, the skin temperature amplitude change data and the blood oxygen fluctuation cycle data are extracted respectively for subsequent analysis. Suppose in a specific example, the time interval of the electrocardiogram segment number is from T1 to T2, the time interval of the skin temperature data is from T3 to T4, and the time interval of the blood oxygen cycle data is from T5 to T6. The intersection segment is the overlapping time period of T3 to T4 and T5 to T6. Finally, the corresponding skin temperature and blood oxygen segments obtained are the intersection part of these two time periods.

[0025] The cross-extraction sub-module constructs a skin temperature skewed distribution map and a blood oxygen skewed distribution map respectively according to the skin temperature amplitude change data and the blood oxygen cycle change data in the corresponding skin temperature and blood oxygen segments, obtains the position of the intersection point of the two distribution maps, calculates the density change slope of the intersection point area, and judges the overlapping range area between regions, as well as the offset direction and distance between the two distribution peaks, to obtain the cross-characteristic of the parameter distribution; First, the skin temperature amplitude change data and the blood oxygen cycle change data need to be processed, and a skin temperature skewed distribution map and a blood oxygen skewed distribution map are drawn respectively. Suppose the skin temperature amplitude change data is [36.1°C, 36.3°C, 36.5°C, 36.7°C, 36.9°C], and the blood oxygen cycle change data is [98%, 99%, 97%, 98%, 100%]. On this basis, the skewed distributions of these two data sets are first obtained through statistical calculations.

[0026] The skewed distribution of the skin temperature amplitude change data can be calculated by the Pearson skewness coefficient, and the formula is as follows: ; where, is the skewness, which is used to measure the skewness of the data distribution. is the total number of data points, which is 5 here (because there are 5 skin temperature data points). is each data point in the data set, representing the specific skin temperature value, such as 36.1°C, 36.3°C, etc., is the data point index. is the mean of the data set, that is, the average of the data points, assumed to be 36.5 °C here. is the standard deviation, reflecting the degree of dispersion of the data, assumed to be 0.327 °C.

[0027] Assume the mean °C, standard deviation °C, calculate the difference between each data point and the mean, and calculate the skewness: ; The calculation results in: ; Thus, the skewness value of the skin temperature data is 0, indicating that the data shows a symmetric distribution.

[0028] For the skewed distribution map of blood oxygen data, a similar method is also used to calculate the skewness. Assume the mean of the blood oxygen data is , standard deviation , substitute into the formula to calculate the skewness: ; The calculation results in: ; Therefore, the skewness value of the blood oxygen data is also 0, indicating that the data is also symmetrically distributed.

[0029] The density change slope of the intersection point is mainly calculated by calculating the change rates of the two distributions. Assume that the intersection point of the distribution maps of skin temperature and blood oxygen is at time point T1, the corresponding skin temperature is 36.6 °C, and the blood oxygen is 98%. When calculating the density change slope of the intersection point area, the change trends of the two distributions are required. Assume the following distribution densities corresponding to the skin temperature variation data and blood oxygen fluctuation data: Skin temperature distribution density : , blood oxygen distribution density : .

[0030] Assume the form of this function is an exponential decay function: , where: is the constant term of the skin temperature distribution, usually set by the overall shape or maximum value of the data. is the decay coefficient of the skin temperature, determining the decay rate of the skin temperature over time, usually obtained by fitting experimental data. : Blood oxygen distribution density. It represents the density of blood oxygen changing over time, and is also a probability density changing over time. It describes the changes of blood oxygen at different time points. Assume the form of this function is also an exponential decay function: , where: is the constant term of the blood oxygen distribution, usually set by the overall shape or maximum value of the data. is the decay coefficient of the blood oxygen, determining the decay rate of the blood oxygen over time, usually obtained by fitting experimental data. Assume the decay coefficient of the skin temperature , the attenuation coefficient of blood oxygen , which means that the rate of change of skin temperature is faster than that of blood oxygen. is the base of the exponential function, used to represent that a variable decays exponentially over time.

[0031] The calculation formula for the slope of density change is: ; Among them, : the slope of density change. Represents the rate of density change in the intersection area. It reflects the rate of change of the skin temperature and blood oxygen distribution maps at the intersection position, and is a key parameter for analyzing the differences in the responses of skin temperature and blood oxygen signals. : the time derivative. Represents taking the derivative of the included function, that is, calculating the rate of change over time. The meaning of the derivative symbol is "taking the derivative with respect to time (t)", and in this formula, it represents calculating the rate of change of the product of the skin temperature and blood oxygen distribution density over time. : the skin temperature distribution density. Represents the density of the skin temperature changing over time, which is a probability density that changes over time. It describes the changes in skin temperature at different time points. : the time variable. Represents a certain moment on the time axis. Time is the independent variable for the changes of all signals (skin temperature and blood oxygen). It controls the change process of skin temperature and blood oxygen over time. : the intersection moment. Represents the specific time point when the intersection of the skin temperature and blood oxygen distribution maps occurs. At the intersection point, the distribution maps of skin temperature and blood oxygen will have an intersection or overlap, and usually this moment is the reference time point when calculating the slope of density change.

[0032] First, calculate the densities of skin temperature and blood oxygen: ; At the intersection point T1, calculate the slope of density change: ; Assume T1 is the moment of 0.1, substitute it into the formula: ; Therefore, the slope of density change in the intersection area is -0.338.

[0033] coefficient , , and are usually set through data analysis. The constants and are set to 1 because the maximum values of these two distributions are usually 1 per unit time, so standardized constants are used. The attenuation coefficients and It is set according to the attenuation rates of skin temperature and blood oxygen signals over time, usually obtained by fitting the attenuation trends of historical data. The settings of these coefficients are based on the historical distribution of the data and experimental measurements, ensuring the accuracy and reliability of the model. In this way, the density change slope in the cross-region of skin temperature and blood oxygen data can be calculated, which reflects the difference in the change rates between skin temperature and blood oxygen, further providing a basis for subsequent analysis.

[0034] The structure establishment sub-module calls the cross-point density slope, regional overlap area, and peak offset direction in the cross-features of parameter distribution. According to the slope change trend and the amplitude of regional position drift, it determines the boundary points of the asynchronous response interval between the skin temperature and blood oxygen distributions, constructs the boundary structure of the parameter cross-relationship on the unified time axis, and establishes the asynchronous physiological signal cross-boundary map. First, according to the density change slope results obtained in the previous processing step, extract the values corresponding to each cross-region, and then extract the overlapping area values of these regions and the time point difference and direction of the peaks in the skin temperature and blood oxygen distribution maps. For example, if the peak of the skin temperature map appears at an earlier time and the peak of the blood oxygen map appears at a later time, the offset direction is backward, and the offset distance is the interval length between the two time points. Then, analyze the change trend of the density change slope values of multiple cross-regions to determine whether there is a trend from negative to positive or from large fluctuations to becoming flat. At the same time, combine whether the peak offset direction within the region is continuously consistent and whether the offset distance expands or shrinks over time to analyze whether there is a trend of asynchronous response between these signals during a certain period. To determine the boundary interval of asynchronous response, by setting judgment criteria, when the density change slope exceeds a certain fixed threshold range and the peak offset distance exceeds a specified value, record the start and end times of this cross-region as the boundary of asynchronous response. Traverse all cross-regions in turn and make judgments, extract the time periods that meet the conditions, generate a set of boundary point sequences, map them at fixed time intervals on the unified time axis, connect each boundary segment in sequence to construct the overall cross-relationship boundary structure, and distinguish between asynchronous and synchronous segments through annotation, and output the asynchronous physiological signal cross-boundary map on the time axis to clearly show the temporal feature distribution of whether the skin temperature and blood oxygen signals respond consistently in different time periods.

[0035] Please refer to Figure 2 and Figure 6 , the risk modeling module includes: The factor construction sub-module obtains the cross-boundary structure of skin temperature and blood oxygen delimited in the asynchronous physiological signal cross-boundary map, calls the electrocardiogram sub-segment numbers in the associated drawdown-marked electrocardiogram fluctuation sub-segment set, and extracts the corresponding amplitude drawdown rate parameters, skin temperature amplitude change rate, and blood oxygen cycle compression duration according to the numbers. Combine the three parameters into the same structure to generate the physiological signal association set. First, traverse the time intervals in the cross-boundary graph, extract the start and end time nodes corresponding to each cross-section, and then, according to each cross-interval time period, call the associated retracement-marked electrocardiogram fluctuation sub-segment set to extract the electrocardiogram sub-segment numbers marked within this time period. For example, when the cross-section time range is between the 120th second and the 135th second, obtain electrocardiogram sub-segments with numbers such as E01, E02, etc. Use each number as an index item to query the corresponding electrocardiogram signal data and calculate the amplitude retracement rate. The specific calculation method is to subtract the amplitude value at the end point from the maximum amplitude value within this segment, and then divide by the maximum amplitude value. If the maximum amplitude of segment E01 is 2.0 mV and the amplitude at the end point is 1.2 mV, then the retracement rate is 0.4. Next, extract the rate of change of skin temperature amplitude, which is obtained by dividing the difference between the end value and the start value within the change interval by the duration. Suppose the skin temperature changes from 36.8 °C to 36.2 °C within this segment and the duration is 15 seconds, then the rate of change is -0.04 per second. Subsequently, obtain the compression duration of the blood oxygen cycle, which is obtained by comparing the difference between the original blood oxygen cycle and the compressed cycle within the cross-section time. If the original blood oxygen cycle is 8 seconds and the actual cycle within the cross-section is 6 seconds, then the compression duration is 2 seconds. Finally, combine the above three parameters of the retracement rate, the rate of change of skin temperature amplitude, and the compression duration of the blood oxygen cycle according to the corresponding numbers to generate a structured physiological signal association set.

[0036] The target calibration sub-module, based on each group of signals in the physiological signal association set, according to the time period range corresponding to the signal cross-boundary, sets the corresponding cross-region mark, adds the mark as a target factor to the signal combination, and establishes the mapping attribute between the target factor and the signal characteristics to obtain the cross-marked target factor set; Read the established data structure records one by one. On the basis of reading each electrocardiogram sub-segment number and its corresponding three parameters, first extract its time period information. By comparing the cross-boundary structure, determine the cross-boundary paragraph to which this group of data belongs, and thus assign a cross-region mark value. For example, assign the mark B1 to the cross-section from the 135th second to the 150th second, and then merge the B1 mark with the signal combination structure to form a marked factor data item. Continue to loop through all subsequent signal combination data, bind the signal structure of each data record to the corresponding cross-mark, and form a signal combination set containing target factors. Subsequently, set the mapping attribute for each combination structure, that is, add a field in the data structure representing the corresponding relationship between the cross-mark and the three parameters of the amplitude retracement rate, the skin temperature amplitude rate, and the blood oxygen compression duration, and establish a two-way index relationship for the field content. When reading any group of target factors, the signal parameters can be traced back, and vice versa. If the number of processed groups is one hundred in the whole process, then one hundred signal structure combination data with target factor attributes are generated, and finally the cross-marked target factor set is output.

[0037] A path classification sub-module constructs a correspondence between target factors and a sample set based on the target factors and corresponding signal characteristics recorded in the cross-marked target factor set. By classifying and segmenting the path arrangement forms of all target factor combination methods and numerical structures within the sample set, a metabolic risk classification model is established. Extract the cross-marked information and three physiological parameters from each group of data item by item, and sort out the signal structures to which all target factors belong. Use the parameter combination method and specific numerical configuration as discriminant features to establish a data set of sample sets. Within this set, perform an arrangement structure analysis on the signal combination paths of all data items. According to the numerical ranges of the amplitude retracement rate, skin temperature wave amplitude rate, and blood oxygen cycle compression duration, group similar features together. For example, data combinations with an amplitude retracement rate in the range of 0.3 to 0.5, a skin temperature rate in the range of -0.05 to -0.02, and a blood oxygen compression within 2 seconds are grouped into one category. If there are 23 groups of paths that meet this feature, they are marked as path A. Then repeat the above operation for other types of parameter structures. For example, paths with an amplitude retracement rate exceeding 0.6 or lower than 0.2 are classified into path B or path C. Eventually, multiple sub-path structures are formed. Each sub-path forms a mapping relationship with its internal combination features. Based on these classification results, a metabolic risk classification model is established, and different path markings are mapped to high-risk, medium-risk, or low-risk level fields to form a final structure model set for signal classification.

[0038] The overall process ends in the risk modeling module, completing the metabolic risk modeling for diabetic patients. The regulation and deconstruction module is a further optimization module for this model, used to identify marginal samples and adjust the classification boundary.

[0039] Please refer to Figure 2 and Figure 7 It also includes a regulation and deconstruction module. The regulation and deconstruction module extracts the classification results of the sample paths in the metabolic risk classification model, redefines the boundary interval adjustment logic for the model output categories, performs an asymmetric structure split on the classification boundary, and generates a trend risk boundary adjustment result. The trend risk boundary adjustment result specifically refers to the sample numbers in the risk marginal area, the trend variation feature group, the trend starting rhythm, the number of amplitude turning points, the signal start sequence, the continuous intensity comparison, and the updated classification boundary structure. The regulation and deconstruction module includes: A marginal recognition sub-module, based on the classification results of all sample paths in the metabolic risk classification model, screens and determines the sample numbers in the risk marginal area, extracts the amplitude retracement rate mutation points in the corresponding electrocardiogram bands of the numbers, and synchronously obtains the skin temperature fluctuation rate and the blood oxygen cycle change direction within the time intervals before and after the bands to establish a risk section positioning result. Based on the classification results of all sample paths in the metabolic risk classification model, perform a screening operation on the sample numbers of the samples whose determination results are marked as the risk marginal zone. In actual operation, set the determination threshold of the risk margin to samples with a risk probability greater than 0.6 but less than 0.75 as marginal samples. If the sample numbers S023, S037, and S044 meet the conditions, they will be included in the screening results. Subsequently, read the corresponding electrocardiogram band signals of these samples respectively, and locate the amplitude retracement rate mutation points therein. The specific method is to identify the moments when the change amplitude of the retracement rate between consecutive time points in the band exceeds 0.15. For example, in sample number S023, the amplitude retracement rate from the 84th second to the 85th second drops from 0.3 to 0.12, then record the 85th second as the mutation point. Continue to extract the skin temperature change data and blood oxygen cycle data from the time period within 10 seconds before and after the mutation point. By calculating the skin temperature change rate per unit time, for example, the skin temperature rises from 36.6 °C to 36.9 °C in the first 10 seconds, then the rate is 0.03 per second, and it drops to 36.4 °C in the next 10 seconds, which is -0.05 per second, to obtain the change in the fluctuation direction from rising to falling; synchronously calculate the change direction of the blood oxygen cycle. If the cycle shortens from 8 seconds to 6 seconds, it is determined as cycle compression, otherwise it is expansion. Finally, uniformly associate the skin temperature rate, blood oxygen cycle direction, and electrocardiogram mutation points within these time periods to form a structured record, and generate the risk section positioning result in combination with the time axis annotation information.

[0040] The trend construction sub-module calls the signal segments in the risk section positioning result to reconstruct the complete skin temperature, blood oxygen, and electrocardiogram trend sequences for each sample number, calculates the starting rhythm and the number of amplitude turning points of each trend, and obtains the trend variation characteristic index set; Call the signal segments in the risk section positioning result, read each electrocardiogram, skin temperature, and blood oxygen data segment containing the mutation point and its front and back time windows, and perform time axis unified splicing construction on all signal segments for each sample number respectively to form the complete trend sequence corresponding to the sample. During the trend reconstruction process, with a sampling frequency of 1 second, connect the adjacent signal segments end to end and fill in the missing time point data. If the 91st second data of a certain skin temperature signal is missing, interpolation is used for valuation. Suppose the 90th second is 36.5 °C and the 92nd second is 36.7 °C, then the valuation of the 91st second is 36.6 °C. After completing the trend reconstruction of the three types of signals, scan the signal sequence in turn to identify the starting rhythm characteristics. For example, in the electrocardiogram sequence, whether there is a constant interval between consecutive wave peaks within the first ten seconds is used as the initial rhythm standard. If the interval time is 0.8 seconds each time, it is marked as a stable rhythm. Continue to identify the number of turning points of the waveform amplitude in the trend. The turning point is defined as the position where the signal change direction changes from rising to falling or from falling to rising. If there are 6 direction changes in the skin temperature trend, it is recorded as 6 turning points, and 4 in the blood oxygen trend is recorded as 4. Finally, combine the starting rhythm types and the number of turning points of the three types of signals of all samples respectively to form the trend variation characteristic index set.

[0041] The boundary adjustment submodule determines the classification boundary interval that the feature combination falls into based on the comparison between the signal start sequence and the continuous strength parameter in the trend variation feature indicator set, redefines the boundary demarcation rules for the category corresponding to each combination, performs asymmetric structural splitting of the original classification boundary, and generates trend risk boundary adjustment results; The start-up order and continuous intensity parameter of each group of signals recorded in the trend variation characteristic index set are compared and analyzed. The signal start-up order refers to the order of the time points corresponding to the positions where the three types of signals first change in the trend segment. For example, in sample number S037, the skin temperature changes at the 40th second, the blood oxygen at the 42nd second, and the ECG change at the 45th second. The order of this group is recorded as skin temperature-blood oxygen-ECG. The continuous intensity parameter is the total amount of amplitude change of the signal over a period of time divided by the length of the period. For example, the skin temperature change value between the 40th and 60th seconds is 1.2 units, and the average intensity is 0.06 per second. All signal combinations are sorted by start-up order and intensity value. Afterwards, it is matched with the boundary interval of the original classification model. If the order of a group of signal combinations is inconsistent with the existing model classification rules, or the continuous intensity value falls into the intersection area of ​​multiple risk levels, it is marked as a boundary fuzzy type, and then the degree of deviation is used to determine whether the classification needs to be reset. If most fuzzy groups are concentrated above or below a certain boundary point, an asymmetric structure split is performed at this position. For example, the original high-risk interval from 0.7 to 1.0 is divided into two sub-segments of 0.7 to 0.85 and 0.85 to 1.0, and different trend feature combination rules are set respectively. The original boundary judgment logic is updated, and finally a new classification boundary definition structure is output and a trend risk boundary adjustment result is generated.

[0042] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A physiological signal-driven diabetes metabolic risk modeling and analysis system, characterized in that: The system includes: A variable sorting module that monitors the skin temperature and blood oxygen saturation signals of diabetic patients, obtains the fluctuation response time difference of each pair of signals in the sample, divides the sample into multiple response difference groups, and generates a grouping result of diabetic physiological signals; An electrocardiogram shearing module that, based on the sorted time periods in the grouping result of diabetic physiological signals, obtains the peak-to-valley amplitude of each cycle and calculates the amplitude retraction rate, marks the cutting points exceeding the retraction threshold, and cuts them into corresponding electrocardiogram sub-segments to generate a set of retraction-marked electrocardiogram fluctuation sub-segments; A distribution border construction module that obtains the electrocardiogram sub-segment numbers in the set of retraction-marked electrocardiogram fluctuation sub-segments, extracts the cross features of skin temperature and blood oxygen distribution, and establishes an asynchronous physiological signal cross-border map; A risk modeling module that obtains the skin temperature and blood oxygen cross-border structure defined in the asynchronous physiological signal cross-border map, sets cross-region markers, and establishes a metabolic risk classification model for distinguishing asynchronous mutation paths.

2. The physiological signal-driven diabetes metabolic risk modeling and analysis system according to claim 1, wherein: The grouping result of diabetic physiological signals includes the skin temperature first fluctuation time point index, the blood oxygen first fluctuation time point index, the fluctuation response time difference delay sequence, and the response difference grouping label. The set of retraction-marked electrocardiogram fluctuation sub-segments is specifically the electrocardiogram sub-segment number, the amplitude retraction rate parameter sequence, and the retraction threshold trigger position marker. The asynchronous physiological signal cross-border map includes the skin temperature distribution map, the blood oxygen distribution map, the cross-point density slope, the distribution overlap region, and the distribution peak offset direction. The metabolic risk classification model includes the skin temperature change amplitude, the blood oxygen cycle shortening amount, the electrocardiogram retraction rate, the cross-region marker, and the target mapping relationship.

3. The physiological signal-driven diabetes metabolic risk modeling and analysis system according to claim 1, characterized in that: The variable sorting module includes: A fluctuation detection sub-module that obtains the skin temperature and blood oxygen saturation signals of diabetic patients, and based on the sampling data within a fixed time window, detects the time point corresponding to the first continuous positive slope segment in the skin temperature rate change curve as the skin temperature first fluctuation time point, and detects the starting time point of the continuous compression nodes in the blood oxygen stable cycle as the blood oxygen first fluctuation time point. Index positions of the two types of signal fluctuation time points are calibrated on the unified time axis respectively to generate a time point index sequence; A time difference calculation sub-module that calls the respective time indexes of skin temperature and blood oxygen in the time point index sequence, calculates the time interval difference between the skin temperature first fluctuation time point and the corresponding blood oxygen first fluctuation time point for each group of sample signal data, writes the difference as a sorting factor into the unified sequence, and establishes a delayed response order sequence according to the numerical sorting of the time interval differences in the sequence; A response grouping sub-module that, according to the numerical range of the time interval differences in the delayed response order sequence, judges the distribution aggregation situation of each sample in the unified sorting interval, calls the adjacent distribution density of each sorting factor in the overall interval, divides the continuous distribution interval by density comparison means, and takes the interval boundary difference value as the clustering division benchmark to obtain the grouping result of diabetic physiological signals.

4. The physiological signal-driven diabetes metabolic risk modeling and analysis system according to claim 1, wherein: The electrocardiogram shearing module includes: A signal extraction sub-module extracts the electrocardiogram cycle signals synchronously collected within each time period based on the sorted time periods recorded in the diabetes physiological signal grouping result, calls the voltage change data recorded within each cycle, locates the maximum voltage peak point and the minimum voltage valley point within each cycle, extracts the continuous amplitude values between adjacent peaks and valleys and arranges them in chronological order to generate a peak-valley amplitude sequence; A drawdown ratio calculation sub-module calculates the ratio of the drawdown value to the corresponding growth value by calling the amplitude growth value from peak to valley within the previous cycle according to the amplitude drawdown value from peak to valley in each segment of the peak-valley amplitude sequence, and sets each group of ratios as drawdown rate parameters and records them in a unified parameter sequence in turn to establish a drawdown rate parameter sequence; A segment division sub-module calls the parameter values in the drawdown rate parameter sequence, compares them with the set amplitude drawdown rate threshold, determines the time points where the ratios exceeding the drawdown rate threshold are located, sets the target time point as the vector cutting point, calls the corresponding electrocardiogram cycle signals within the sorted time period, performs signal segmentation operations according to the cutting point position, records each segment of the signal as an independent sub-segment and numbers it to obtain a set of drawdown-marked electrocardiogram fluctuation sub-segments; 5. The physiological signal-driven diabetes metabolic risk modeling and analysis system according to claim 1, wherein: The distribution border construction module includes: A number tracing sub-module obtains the electrocardiogram sub-segment numbers recorded in the set of drawdown-marked electrocardiogram fluctuation sub-segments, and based on the time index interval corresponding to the numbers, calls the skin temperature fluctuation record data and the blood oxygen cycle record data, and extracts the skin temperature amplitude change interval and the blood oxygen fluctuation cycle that overlap with the time period of the electrocardiogram sub-segment number respectively to obtain the corresponding skin temperature and blood oxygen segments; An intersection extraction sub-module constructs a skin temperature skewed distribution map and a blood oxygen skewed distribution map respectively according to the skin temperature amplitude change data and the blood oxygen cycle change data in the corresponding skin temperature and blood oxygen segments, obtains the position of the intersection point of the two distribution maps, calculates the density change slope of the intersection point area and judges the overlapping range area between regions, as well as the offset direction and distance between the two distribution peaks to obtain the parameter distribution intersection characteristics; A structure establishment sub-module calls the intersection point density slope, the region overlapping area and the peak offset direction in the parameter distribution intersection characteristics, determines the boundary points of the asynchronous response interval between the skin temperature and blood oxygen distributions according to the slope change trend and the regional position drift amplitude, constructs a parameter intersection relationship boundary structure on the unified time axis, and establishes an asynchronous physiological signal intersection boundary map.

6. The physiological signal-driven diabetes metabolic risk modeling and analysis system according to claim 5, wherein: For constructing the skin temperature skewed distribution map and the blood oxygen skewed distribution map, the formula is used: ； Calculate the skewness of the distribution of skin temperature and blood oxygen data , and refer to the skewness of the distribution of skin temperature and blood oxygen data to construct a skin temperature skewed distribution map and a blood oxygen skewed distribution map; wherein, is the total number of data points, is each data point in the skin temperature or blood oxygen data set, is the data point index, is the mean of the data set, is the standard deviation of the data set.

7. The physiological signal-driven diabetes metabolic risk modeling and analysis system according to claim 5, characterized in that: For calculating the density change slope of the intersection area , the formula is adopted: ； Among them, used to reflect the change rate of the skin temperature and blood oxygen distribution map at the intersection position, is the time derivative, is the skin temperature distribution density, is the blood oxygen distribution density, represents the moment on the time axis, represents the time point when the intersection of the skin temperature and blood oxygen distribution map occurs.

8. The physiological signal-driven diabetes metabolic risk modeling and analysis system according to claim 1, wherein: The risk modeling module includes: A factor construction sub-module obtains the skin temperature and blood oxygen intersection boundary structure delimited in the asynchronous physiological signal intersection boundary map, calls the electrocardiogram sub-segment numbers in the associated set of drawdown-marked electrocardiogram fluctuation sub-segments, extracts the corresponding amplitude drawdown rate parameters, the skin temperature amplitude change rate and the blood oxygen cycle compression duration according to the numbers, combines the three parameters into the same structure to generate a physiological signal association set; The target calibration submodule is based on each group of signals in the physiological signal association set, according to the time period range corresponding to the signal crossing boundary, and sets the corresponding crossing area mark, adds the mark as a target factor to the signal combination, establishes a mapping attribute between the target factor and the signal feature, and obtains a cross-mark target factor set; The pathway classification submodule constructs the corresponding relationship between the target factor and the sample set according to the target factors and corresponding signal features recorded in the cross-labeled target factor set, and establishes a metabolic risk classification model by classifying and segmenting the path arrangement forms of all target factor combinations and numerical structures within the sample set.

9. The physiological signal-driven diabetes metabolic risk modeling and analysis system according to claim 1, wherein: It also includes a regulation deconstruction module, which extracts the classification results of the sample path in the metabolic risk classification model, redefines the boundary interval adjustment logic of the model output category, performs asymmetric structural splitting on the classification boundary, and generates a trend risk boundary adjustment result; The trend risk boundary adjustment result specifically refers to the risk edge area sample number, trend variation feature group, trend starting rhythm, number of amplitude turning points, signal start sequence, continuous intensity comparison, and updated classification boundary structure.

10. The physiological signal-driven diabetes metabolic risk modeling and analysis system according to claim 9, wherein: The regulation and deconstruction module includes: The edge recognition submodule, based on the classification results of all sample paths in the metabolic risk classification model, screens the sample numbers determined to be risk edge areas, extracts the amplitude retracement rate mutation points in the ECG bands corresponding to the numbers, and simultaneously obtains the skin temperature fluctuation rate and blood oxygen cycle change direction in the time interval before and after the bands, and establishes the risk segment positioning results; The trend construction submodule calls the signal fragments in the risk segment positioning results, reconstructs the complete skin temperature, blood oxygen and ECG trend sequence under each sample number, calculates the starting rhythm and amplitude turning number of each trend, and obtains the trend variation feature indicator set; The boundary adjustment submodule determines the classification boundary interval into which the feature combination falls based on the comparison between the signal start-up sequence and the continuous intensity parameter in the trend variation feature indicator set, redefines the boundary demarcation rules for the category corresponding to each group of combinations, performs asymmetric structural splitting of the original classification boundary, and generates a trend risk boundary adjustment result.

Images