Coronary heart disease accurate prediction method based on multi-source heterogeneous data integration

Abstract

The invention provides a multi-source heterogeneous data integrated coronary heart disease accurate prediction method, which comprises the following steps: acquiring various physiological signals of a patient in real time, the physiological signals comprising ST segment change characteristics and basic cycle function parameters in electrocardiosignals, and obtaining a multi-source signal data set; according to the joint data set, analyzing instantaneous fluctuation characteristics of a vascular resistance index in a postprandial hyperlipemia window period, extracting a vascular resistance fluctuation amplitude from the instantaneous fluctuation characteristics, and if it is detected that the fluctuation amplitude exceeds a preset threshold range, marking the window as a high-risk time window; if the score value exceeds a preset threshold value, triggering a coronary heart disease early warning signal according to the comprehensive risk score value in combination with a feature mode of a coronary heart disease hidden period in historical data; and storing the current high-risk time window characteristics and the blood sugar and blood fat curve change trend through a triggered coronary heart disease early warning signal to obtain structured risk archive data.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and particularly relates to a multi-source heterogeneous data integrated coronary heart disease precise prediction method. BACKGROUND

[0002] Coronary heart disease, as an important type of cardiovascular disease, poses a serious threat to human health. Early prediction of the incubation period is of great significance, which can help patients gain valuable intervention time and reduce the risk of sudden events. Research in this field is not only the frontier of the intersection of medicine and engineering technology, but also the key to improving public health. However, current methods for predicting the incubation period of coronary heart disease rely too much on detection data at a single time point, ignoring the complex changes in human physiological state under different circumstances, especially within a specific physiological window period. The dynamic response mechanism inside the body is not fully captured. In addition, existing methods often lack in-depth analysis of the interaction between different signals when integrating multi-source data, resulting in poor adaptability of the prediction model to individual differences and difficulty in meeting real needs in complex environments. Therefore, how to accurately depict the dynamic relationship between body circulation function and blood flow reserve in a specific state, such as a postprandial special period. Changes in this relationship directly affect the balance of vascular resistance and myocardial blood supply, and failure to effectively capture this instantaneous fluctuation feature makes it difficult to reveal abnormal signals hidden beneath the surface. Further, the complexity of this dynamic relationship presents another challenge, which is how to reasonably allocate attention to different time periods in multi-source physiological data, ensuring that key signals are not drowned out, while avoiding interference from irrelevant information. Therefore, how to integrate multi-source physiological signals and dynamically adjust the importance of different time windows to accurately capture the instantaneous changes in the relationship between circulation function and blood flow reserve has become a key problem that needs to be solved. SUMMARY

[0003] The present application provides a multi-source heterogeneous data integrated coronary heart disease precise prediction method, mainly comprising:

[0004] The physiological signal of the patient is collected, and the physiological signal is processed to obtain a multi-source signal data set. Feature extraction is performed on the multi-source signal data set to obtain a target time period feature in the postprandial hyperlipidemia window period, label the abnormal fluctuations of the ST segment change feature and the basic circulation function parameter, and determine the myocardial perfusion pressure abnormal interval. The myocardial perfusion pressure abnormal interval data is obtained, combined with the blood glucose and blood lipid data, and a joint data set is formed by synchronous matching. The joint data set is analyzed, the fluctuation amplitude of the vascular resistance index in the postprandial hyperlipidemia window period is extracted, and a high-risk time window with a fluctuation amplitude exceeding a preset range is marked. The correlation coefficient matrix of the coronary flow reserve index in the high-risk time window data and the basic circulation function parameter is calculated, and the weight distribution of the multi-source signal data set is adjusted. The correlation between the vascular resistance index fluctuation and the blood glucose and blood lipid data is analyzed to obtain an endothelial function abnormality index. The endothelial function abnormality index and the multi-source signal data set are weighted, and a comprehensive risk score value is output. According to the comprehensive risk score value and the historical coronary heart disease latent period feature mode, a coronary heart disease early warning signal is output. The features of the high-risk time window and the blood glucose and blood lipid data are stored, and a structured risk profile data is generated.

[0005] Further, the physiological signal of the patient is collected by the wearable device, and the physiological signal is processed to obtain a multi-source signal data set, including:

[0006] The electrocardiogram signal is collected, the voltage difference value and the slope change value of the ST segment change feature are determined, the blood pressure value, the blood oxygen saturation percentage and the heart rate value of the basic circulation function parameter are collected, and the original data stream containing the time stamp is generated. The electrocardiogram signal in the original data stream is filtered to remove baseline drift and power frequency interference, the blood pressure value is removed to remove pulse noise, and the blood oxygen saturation percentage is smoothed to remove instantaneous fluctuations and remove outliers to obtain a denoising data sequence. The ST segment voltage difference value, the blood pressure value, the heart rate value and the blood oxygen saturation percentage of the denoising data sequence are normalized to generate the multi-source signal data set with uniform dimensions.

[0007] Further, the feature extraction of the multi-source signal data set to obtain the target time period feature in the postprandial hyperlipidemia window period, label the abnormal fluctuations of the ST segment change feature and the basic circulation function parameter, and determine the myocardial perfusion pressure abnormal interval, including:

[0008] The multi-source signal data set is segmented, the voltage offset value of the ST segment change feature in the postprandial time window is calculated, the ST segment elevation or depression event is recorded, the peak ratio and time span of T wave morphology change are extracted, the mean value of R-R interval and the square root of the difference value are calculated to obtain the heart rate variability; the ST segment elevation amplitude, T wave morphology change and heart rate variability are dimensionally reduced, the principal components are retained, and the target time period feature is generated; according to the target time period feature, the abnormal value of the basic circulation function parameter is located, and a time sequence sequence containing the ST segment event and the abnormal marker is generated; the product of the ST segment offset value and the blood pressure difference in the time sequence sequence is calculated, and the interval continuously lower than the threshold value is determined as the myocardial perfusion pressure abnormal interval.

[0009] Further, the myocardial perfusion pressure abnormal interval data is acquired, combined with blood glucose and blood lipid data, and synchronized to form a joint data set, including:

[0010] The ST segment offset value sequence and the blood pressure change value sequence of the myocardial perfusion pressure abnormal interval are extracted, and the concentration value sequence of the blood glucose and blood lipid data is acquired; the concentration value sequence of the blood glucose and blood lipid data is interpolated to generate an equidistantly sampled time sequence; the time sequence is aligned with the timestamp of the myocardial perfusion pressure abnormal interval data, and resampling is performed to generate a synchronized metabolic parameter sequence; the ST segment offset value sequence, the blood pressure change value sequence, the blood glucose concentration value sequence and the blood lipid concentration value sequence are combined to form the joint data set.

[0011] Further, the joint data set is analyzed, the fluctuation amplitude of the vascular resistance index in the postprandial hyperlipidemia window period is extracted, and the high-risk time window with the fluctuation amplitude exceeding the preset range is marked, including:

[0012] The blood pressure change value sequence is extracted from the joint data set, the pulse pressure difference and the mean arterial pressure are calculated, the cardiac output is estimated in combination with the heart rate sequence, and the vascular resistance index sequence is generated; the difference between the maximum value and the minimum value of the vascular resistance index sequence in the window is calculated to determine the fluctuation amplitude; the blood lipid concentration value is extracted from the joint data set, and the window with the fluctuation amplitude exceeding the preset range and the blood lipid concentration value exceeding the threshold value is marked as the high-risk time window.

[0013] Further, the correlation coefficient matrix of the coronary flow reserve index in the high-risk time window data and the basic circulation function parameter is calculated, and the weight distribution of the multi-source signal data set is adjusted, including:

[0014] The product of the mean arterial pressure and the heart rate ratio in the high-risk time window data is calculated to generate the coronary flow reserve index; the correlation coefficient of the coronary flow reserve index and the basic circulation function parameter is calculated to construct the correlation coefficient matrix; the absolute values of the coefficients of the correlation coefficient matrix are normalized, the coefficient weight lower than the threshold value is adjusted, and the weight distribution is generated.

[0015] Further, the analysis of the correlation between the blood glucose and blood lipid data and the fluctuation of the vascular resistance index obtains an endothelial dysfunction index, the endothelial dysfunction index is weighted with the multi-source signal data set, and a comprehensive risk score value is output, including:

[0016] The correlation coefficient of the vascular resistance index and the blood glucose and blood lipid data in a window is calculated, an endothelial injury value is determined, the endothelial dysfunction index is generated, the ST segment offset value, the blood pressure change value, the heart rate variability, the blood oxygen saturation, the blood glucose concentration value and the blood lipid concentration value of the multi-source signal data set are weighted and summed, the endothelial dysfunction index is combined to generate a weighted score value, the weighted score value is mapped to a preset interval according to the mean value of the non-diagonal elements of the correlation coefficient matrix, and the comprehensive risk score value is output.

[0017] Further, the output of the coronary heart disease early warning signal according to the comprehensive risk score value and the historical coronary heart disease occult period characteristic mode includes:

[0018] The ST segment offset value difference ratio, the heart rate variability low value duration ratio and the blood pressure threshold value ratio are extracted from historical data to generate a characteristic mode set, the ST segment offset value sequence, the heart rate variability sequence and the blood pressure sequence are extracted from real-time data, the absolute value difference of the characteristic mode set is calculated to generate a mode matching mark, and the coronary heart disease early warning signal is output according to the mode matching mark and the continuous state of the comprehensive risk score value.

[0019] Further, the storage of the characteristics of the high-risk time window and the blood glucose and blood lipid data generates structured risk profile data, including:

[0020] The ST segment offset maximum value, the blood pressure fluctuation standard deviation and the heart rate variability minimum value of the high-risk time window are extracted, the change rate of the blood glucose and blood lipid data is calculated, a to-be-stored data set is generated, the risk record entry is generated by arranging the patient identifier, the early warning time, the comprehensive risk score value, the ST segment offset maximum value, the blood pressure fluctuation standard deviation, the heart rate variability minimum value, the blood glucose change rate and the blood lipid change rate, and the risk record entry is serialized, a check value and a time stamp are added, and the structured risk profile data is generated.

[0021] The technical scheme provided by the embodiment of the application can include the following beneficial effects:

[0022] This invention discloses a method for accurate prediction of coronary artery disease (CAD) based on multi-source heterogeneous data integration. Addressing the comprehensive risk assessment of abnormal myocardial perfusion pressure and vascular endothelial function during the postprandial hyperlipidemia window, the method constructs a joint dataset by denoising, standardizing, extracting features, and reducing the dimensionality of multi-source signals, combined with time-aligned blood glucose and lipid profiles. It analyzes the instantaneous fluctuation characteristics of the vascular resistance index, extracts high-risk time windows, calculates the correlation between coronary flow reserve and circulatory function parameters, adjusts signal weights, and generates endothelial function abnormality indicators and a comprehensive risk score. When the score exceeds a threshold, an early warning signal is triggered based on historical CAD latent period characteristics, and structured risk profile data is stored. This invention significantly improves the accuracy and real-time performance of early CAD detection through deep fusion and dynamic analysis of multi-source signals and metabolic parameters, providing efficient support for personalized cardiovascular risk management. Attached Figure Description

[0023] Fig. 1 This is a flowchart of a method for accurate prediction of coronary heart disease by integrating multi-source heterogeneous data according to the present invention.

[0024] Fig. 2 This is a schematic diagram of a method for accurate prediction of coronary heart disease by integrating multi-source heterogeneous data according to the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and thoroughly described below with reference to the accompanying drawings. The described embodiments are merely some embodiments of the present invention.

[0026] like Figs. 1-2 This embodiment of a method for accurate prediction of coronary heart disease by integrating multi-source heterogeneous data may specifically include:

[0027] S101. Real-time acquisition of multiple physiological signals from patients, including ST segment change characteristics in electrocardiogram signals and basic circulatory function parameters, to obtain a multi-source signal dataset.

[0028] The first preset sampling rate is synchronized with the acquisition of the patient's chest lead ECG signal through the multi-channel sensor array built-in the wearable device, the J point position in the first preset time window after the R wave peak value in the ECG waveform is determined to determine the starting point of the ST segment, the baseline regression point in the second preset time window before the T wave starting point is determined to determine the termination point of the ST segment, the voltage difference between the starting point and the termination point is calculated as the ST segment offset, the least square fitting straight line slope of the segment curve is calculated as the slope change value, and the systolic and diastolic pressure values of the cuff blood pressure sensor, the blood oxygen saturation percentage of the photoelectric plethysmogram sensor and the pulse frequency count value are acquired at the second preset sampling rate, and the multi-source physiological signal original data stream containing the time stamp mark is obtained. The ECG signal in the multi-source physiological signal original data stream is filtered by a Butterworth band-pass filter to remove baseline drift below the first cutoff frequency and power frequency interference above the second cutoff frequency, the blood pressure signal is filtered by a median filter to remove pulse noise, the blood oxygen signal is smoothed by a moving average method to remove instantaneous fluctuations, and the abnormal values are removed according to the preset physiological threshold range of each signal, and the denoised multi-source physiological signal sequence is obtained. According to the denoised multi-source physiological signal sequence, the normalized ST segment feature is obtained by dividing the ST segment offset of the ECG signal by the preset reference voltage value, the standardized blood pressure feature is obtained by dividing the blood pressure value by the preset blood pressure range value after subtracting the preset reference blood pressure value, the standardized heart rate feature is obtained by dividing the heart rate by the preset heart rate range value after subtracting the preset minimum heart rate value, the standardized blood oxygen feature is obtained by dividing the blood oxygen saturation by the preset blood oxygen range value after subtracting the preset minimum blood oxygen value, and the unified dimension multi-source signal data set is formed according to the time stamp alignment of each standardized feature.

[0029] Specifically, the wearable physiological signal monitoring device realizes real-time acquisition of physiological parameters of patients through a multi-channel sensor array. The extraction of the ST segment change feature of the ECG signal is an important indicator for diagnosing myocardial ischemia.

[0030] Specifically, the recognition of the R wave peak value uses the difference threshold method, when the voltage difference of adjacent sampling points exceeds the preset threshold and continues to rise, it is determined as the R wave rising branch, and when the voltage reaches the maximum value and then starts to decrease, the R wave peak value position is determined. In the time window of 40-80 milliseconds extended from the R wave peak value, the J point, i.e. the starting point of the ST segment, is determined by finding the inflection point of the voltage regression to the baseline level. Within 20-40 milliseconds before the T wave starting point, the baseline level is determined by calculating the average value of a plurality of consecutive sampling points, so as to mark the termination point of the ST segment. The calculation of the ST segment offset involves the measurement of the voltage difference between the starting point and the termination point.

[0031] In one possible implementation, the voltage values of all sampling points in the ST segment relative to the baseline are first calculated, and then the least square method is used to fit these points to form a straight line, and the slope of the straight line is the slope change value. Under normal circumstances, the ST segment should be close to horizontal, and when there is a significant upward or downward slope, it indicates that there may be abnormal myocardial blood supply. This quantitative analysis method is more objective and accurate than simply observing the waveform. The synchronous collection of basic circulation function parameters such as blood pressure, heart rate and blood oxygen saturation provides a data basis for comprehensive evaluation.

[0032] It should be noted that the sampling frequency settings of different physiological parameters take into account the signal characteristics and clinical needs. The electrocardiogram signal changes rapidly and requires a higher sampling rate to accurately capture the waveform details; while blood pressure and other parameters change relatively slowly, a lower sampling rate can meet the monitoring needs, while reducing the data storage and transmission burden. Denoising is a key link to ensure signal quality. The Butterworth band-pass filter has the characteristics of flat frequency response in the passband, which can effectively filter out interference while preserving the main frequency components of the electrocardiogram signal. Baseline drift is mainly caused by respiratory movement and poor electrode contact, and the frequency is usually below 0.5Hz; power frequency interference comes from the power grid, with a frequency of 50Hz or 60Hz. Median filtering has good suppression effect on pulse noise in blood pressure signal, it replaces the current value with the median value in the neighborhood, which can remove abnormal spikes without excessive smoothing of the signal. Standardization processing makes physiological parameters of different dimensions comparable.

[0033] For example, the electrocardiogram ST segment offset is usually measured in millivolts, while blood pressure is measured in millimeters of mercury, and direct comparison is meaningless. Through normalization mapping, the ST segment offset is converted to a dimensionless range of 0-1, and blood pressure, heart rate and other parameters are also subjected to corresponding linear transformation.

[0034] S102, feature extraction and dimensionality reduction are performed on the multi-source signal data set to obtain the target time period characteristics in the postprandial hyperlipidemia window period, and the ST segment change in the electrocardiogram signal and the abnormal fluctuation of the circulatory function parameter are time series labeled to determine the myocardial perfusion pressure abnormal interval.

[0035] The multi-source signal data set is segmented according to the meal time point, the voltage offset value of the ST segment relative to the baseline is calculated in the interval from the first preset time to the second preset time after the meal, if the offset value exceeds the preset positive threshold, it is recorded as an ST segment elevation event, if the offset value is lower than the preset negative threshold, it is recorded as an ST segment depression event, the ratio of the T wave peak value to the baseline and the time span of the T wave from the start to the peak are extracted, the time interval of adjacent R wave peaks is identified from the electrocardiogram signal as the R-R interval, the mean value of all R-R intervals and the square sum of the difference value of each R-R interval and the mean value are calculated, and then the heart rate variability index is obtained by dividing the number of R-R intervals and taking the square root. The principal component analysis method is used for dimension reduction processing of the ST segment offset value, the T wave shape parameter and the heart rate variability index, and the principal components with an accumulated contribution rate exceeding a preset contribution rate threshold are retained to obtain the target time period feature vector in the postprandial hyperlipidemia window period. According to the time when the ST segment elevation or depression event occurs in the target time period feature vector, the blood pressure value, heart rate value and blood oxygen saturation value at the corresponding time in the multi-source signal data set are located, the average value of each parameter in the preset time window before the meal is calculated as the individual baseline value, if the absolute value of the difference between the current blood pressure value and the blood pressure baseline value exceeds the preset blood pressure fluctuation threshold, the absolute value of the difference between the heart rate value and the heart rate baseline value exceeds the preset heart rate fluctuation threshold, or the blood oxygen saturation is lower than the preset minimum threshold, a circulatory dysfunction marker is added at the time to form a time sequence labeling sequence containing the time when the ST segment changes, the event type and the circulatory dysfunction marker. For each time in the time sequence labeling sequence, the overlapping interval of the ST segment change event and the circulatory dysfunction marker on the time axis is identified, the difference between the systolic pressure and the systolic pressure baseline value in the overlapping interval is calculated as the blood pressure change value, the product of the absolute value of the ST segment offset value and the blood pressure change value is taken as the perfusion pressure index, and if the number of continuous time points with the perfusion pressure index continuously lower than the preset perfusion threshold exceeds the sampling point number corresponding to the preset continuous time, the time interval corresponding to the continuous time is determined as the myocardial perfusion pressure abnormal interval.

[0036] Specifically, accurate identification of the postprandial hyperlipidemia window period is of great significance for cardiovascular event warning.

[0037] In a possible implementation, the marking of the meal time point is realized through a manual marking button on the patient's wearing device or through the identification of eating actions by an acceleration sensor. The 2-6 hours after the meal is a key period of lipid metabolism, and the increase in the concentration of triglycerides in the blood will affect the blood viscosity and then affect the myocardial blood supply. The monitoring of the ST segment change is a sensitive indicator for capturing such microcirculation changes. The determination of ST segment elevation and depression involves accurate identification of the baseline.

[0038] Specifically, the baseline is determined by the average voltage level of the PR segment, which is the isoelectric line between the end of atrial depolarization and the beginning of ventricular depolarization. An ST segment elevation is recorded when the voltage value at 60-80 milliseconds after the J point of the ST segment is higher than the baseline by more than 0.1 mV, and an ST segment depression is recorded when the voltage value is lower than the baseline by more than 0.05 mV. This quantitative standard makes the identification of abnormal events more objective and reliable. Heart rate variability, as an important indicator of autonomic nervous function, is calculated based on the statistical properties of R-R intervals. The R-wave peak is identified by the highest point of the QRS complex, and the time interval between two adjacent R-wave peaks is the R-R interval.

[0039] For example, if 10 consecutive R-R intervals are measured as 850, 860, 845, 855, 865, 840, 870, 850, 855, and 860 milliseconds, the average value is first calculated as 855 milliseconds, then the sum of the squares of the differences between each interval and the average value is calculated, and finally the standard deviation is obtained by taking the square root, which is about 9.3 milliseconds. This is the time-domain heart rate variability indicator. The application of principal component analysis in feature dimension reduction is based on the correlation of the data.

[0040] It should be noted that there is a physiological correlation between the ST segment shift value, the T wave morphology parameter, and the heart rate variability indicator. By constructing a covariance matrix and calculating the eigenvalues and eigenvectors, the main direction of data change can be found. When the cumulative contribution rate of the current three principal components reaches more than 85%, these three principal components can replace the original multiple features, retaining the main information while reducing the data dimension. The establishment of individual baseline values takes into account the individual differences in physiological parameters.

[0041] In one embodiment, blood pressure and heart rate are measured every 5 minutes within 30 minutes before a meal, and the average of 6 measurements is taken as the baseline value for the individual. This method better reflects the individual's basic state than using a fixed standard value, improving the accuracy of abnormal detection. The perfusion pressure indicator reflects the myocardial blood supply state. When the ST segment appears to be depressed and the blood pressure drops, it indicates that both myocardial blood supply and systemic circulation are problematic, and the product of the two can comprehensively reflect the degree of deterioration of myocardial perfusion.

[0042] For example, if the ST segment is depressed by 0.2 mV and the systolic blood pressure is decreased by 20 mmHg compared to the baseline, the perfusion pressure indicator is 4, and if this indicator remains below the preset threshold for more than 5 minutes, it indicates that the myocardium is in a state of persistent ischemia and needs to be intervened in a timely manner.

[0043] S103, acquire myocardial perfusion pressure abnormal interval data, combine the blood glucose and blood lipid change curve drawn based on the real-time monitored blood glucose and blood lipid data, and perform synchronous matching through time sequence alignment method to obtain a combined data set of fused physiological signals and metabolic parameters.

[0044] The myocardial perfusion pressure abnormal interval data is acquired, the starting timestamp, the ending timestamp, the ST segment offset value sequence and the blood pressure change value sequence in each abnormal interval are extracted, the blood glucose concentration value of each preset monitoring interval in the corresponding time period is read from the continuous blood glucose monitor, the blood lipid concentration value of each preset detection interval is read from the portable blood lipid detection device, the time markers of each measurement value are arranged in time sequence according to time, and the intermediate values are calculated between adjacent measurement points according to a target sampling rate by a linear interpolation method, so as to obtain the blood glucose time sequence and the blood lipid time sequence sampled at equal intervals. For the blood glucose time sequence and the blood lipid time sequence, the timestamp information is compared with the timestamp of the myocardial perfusion pressure abnormal interval, the time deviation value of the two is calculated, if the deviation value exceeds the preset synchronization threshold, the time markers of the blood glucose and blood lipid data are adjusted by the timestamp interpolation method, so as to be aligned with the timestamp of the myocardial perfusion pressure abnormal interval data, after alignment, the blood glucose and blood lipid data are resampled according to the sampling rate of the myocardial perfusion pressure data, and the time-synchronized metabolic parameter sequence is obtained. According to the time-synchronized metabolic parameter sequence, the ST segment offset value, the blood pressure change value in the myocardial perfusion pressure abnormal interval and the blood glucose concentration value and the blood lipid concentration value at the corresponding moment are extracted at each common time point, a four-dimensional data vector is constructed in the order of the ST segment offset value, the blood pressure change value, the blood glucose concentration value and the blood lipid concentration value, all four-dimensional data vectors are arranged in time sequence, and a combined data set of the fused physiological signal and metabolic parameter is formed.

[0045] Specifically, the acquisition of the myocardial perfusion pressure abnormal interval data involves comprehensive determination of multiple physiological parameters.

[0046] Specifically, when the ST segment is continuously depressed by more than 0.1 mV and the systolic pressure is decreased by more than 15 mmHg, this time period is marked as an abnormal interval. Each abnormal interval contains complete starting and ending timestamps, and the ST segment offset value and the blood pressure change value of all sampling points in the interval. These data are stored in the form of time sequence, providing a basis for subsequent time sequence analysis. The continuous blood glucose monitor realizes real-time monitoring through a subcutaneously implanted glucose sensor.

[0047] In one possible implementation, the sensor measures the glucose concentration of interstitial fluid every 5 minutes, and converts it into a blood glucose concentration value through a calibration algorithm. The portable blood lipid detection device uses a reflectance photometry method to perform rapid detection on the test paper through fingertip blood sampling, and usually measures once every 30 minutes to 1 hour. The difference in sampling frequency is the reason for the subsequent time alignment. The linear interpolation method plays a key role in generating equally spaced sampling data.

[0048] For example, the blood glucose is measured as 6.5 mmol / L at 10:00 and 6.8 mmol / L at 10:05. If the value at 10:03 is needed, it is calculated as 6.68 mmol / L by linear interpolation. This method assumes that the change between adjacent measurement points is linear, which has a good approximation effect in a short time interval although there is a certain error. The necessity of time synchronization is due to the clock deviation and sampling frequency difference of different devices.

[0049] It should be noted that even if each device is set to the standard time, due to the precision limitation of the internal clock, there will still be a deviation of seconds or even minutes after a long time of running. The time stamp comparison is realized by calculating the time difference of the same physiological event recorded by each device.

[0050] For example, the event of starting a meal is recorded as 10:00:00 on the electrocardiogram monitoring device and 10:00:45 on the blood glucose monitoring device, and there is a time deviation of 45 seconds. The resampling process ensures that the multi-source data has the same time resolution. If the sampling rate of the myocardial perfusion pressure data is 1 Hz, and the interpolated blood glucose data still maintains an interval of 5 minutes, further interpolation processing is needed for the blood glucose data to generate 299 interpolation points between every two adjacent 5-minute data points, so as to reach a sampling rate of 1 Hz. Although this processing cannot increase the amount of information, it creates the necessary conditions for data fusion. The construction of the four-dimensional data vector realizes the unified representation of heterogeneous data.

[0051] In one embodiment, the vector at a certain time may be [0.15, 18, 7.2, 2.8], representing an ST segment depression of 0.15 mV, a blood pressure drop of 18 mmHg, a blood glucose of 7.2 mmol / L, and a triglyceride of 2.8 mmol / L, respectively. This vector representation not only facilitates subsequent data storage and transmission, but more importantly, lays the foundation for multi-parameter joint analysis.

[0052] S104, according to the joint data set, analyze the instantaneous fluctuation characteristics of the vascular resistance index in the postprandial hyperlipidemia window period, extract the vascular resistance fluctuation amplitude, and if the fluctuation amplitude is detected to be out of the preset threshold range, mark it as a high-risk time window.

[0053] According to the blood pressure change value sequence in the joint data set, the systolic pressure and diastolic pressure values are extracted, the difference between the systolic pressure and the diastolic pressure is calculated to obtain the pulse pressure difference, the diastolic pressure is added to the pulse pressure difference multiplied by a preset coefficient to obtain the mean arterial pressure, at the same time, from the heart rate sequence and the pulse pressure difference sequence, the stroke volume is estimated by dividing the pulse pressure difference by a preset arterial stiffness parameter, the product of the stroke volume and the heart rate is obtained to obtain the cardiac output estimate value, the mean arterial pressure is divided by the cardiac output estimate value to obtain the vascular resistance index, and the vascular resistance index time sequence in the postprandial hyperlipidemia window period is formed. For the vascular resistance index time sequence, a sliding window method is used to process the data in each preset length time window, the difference between the maximum value and the minimum value of the vascular resistance index in the window is calculated as the fluctuation amplitude, the frequency corresponding to the maximum amplitude is identified as the main fluctuation frequency by performing fast Fourier transform on the frequency spectrum, and the number of continuous sampling points whose fluctuation amplitude exceeds the preset multiple of the mean value in the window is counted as the abnormal duration, and the instantaneous fluctuation characteristic data containing the fluctuation amplitude, the main fluctuation frequency and the abnormal duration of each time window are obtained. The fluctuation amplitude sequence is extracted from the instantaneous fluctuation characteristic data, and the blood lipid concentration value at the corresponding time is extracted from the joint data set, if the fluctuation amplitude of a time window exceeds the upper limit of the preset normal fluctuation range, and the blood lipid concentration value at the time corresponding to the window exceeds the preset hyperlipidemia determination threshold, the time window is marked as a high-risk time window.

[0054] Specifically, the calculation of the vascular resistance index involves the basic principles of hemodynamics.

[0055] Specifically, the vascular resistance reflects the resistance encountered by blood flowing in the blood vessels, similar to the concept of resistance in an electrical circuit. The calculation of the mean arterial pressure uses an empirical formula, which is the diastolic pressure plus one-third of the pulse pressure difference. This coefficient 0.33 comes from the physiological characteristics that the systolic period accounts for about one-third and the diastolic period accounts for about two-thirds in a cardiac cycle.

[0056] For example, when the systolic pressure is 130 mmHg and the diastolic pressure is 80 mmHg, the pulse pressure difference is 50 mmHg, and the mean arterial pressure is about 97 mmHg. The estimation of stroke volume is based on the principle of arterial compliance.

[0057] In one possible implementation, the greater the pulse pressure difference, the more blood pumped out with each heart beat. The arterial stiffness parameter reflects the elastic characteristics of the blood vessel wall. Young and healthy people have a smaller arterial stiffness parameter and good blood vessel elasticity. Patients with hyperlipidemia have increased arterial stiffness due to lipid deposition in the blood vessel wall. By dividing the pulse pressure difference by the arterial stiffness parameter, the stroke volume can be estimated, and then multiplied by the heart rate to obtain the cardiac output. Although this indirect measurement method is not as accurate as cardiac catheterization, it has the advantages of non-invasiveness and continuous monitoring. The sliding window method is widely used in time series analysis. The choice of window length needs to balance the time resolution and statistical stability. A shorter window can capture rapid changes but is easily affected by noise; a longer window has stable statistical properties but may miss transient abnormalities.

[0058] For example, selecting a window length of 5 minutes and sliding every 30 seconds can ensure a certain statistical significance while timely detecting abnormal fluctuations. Fast Fourier transform converts the time domain signal to the frequency domain for analysis.

[0059] It should be noted that the periodic fluctuations in vascular resistance can be caused by various physiological rhythms, including chest pressure changes caused by respiration, and blood vessel contraction and relaxation responses regulated by the autonomic nervous system. Through frequency spectrum analysis, the dominant frequency component can be identified.

[0060] For example, fluctuations around 0.25 Hz may correspond to the respiratory rate, and fluctuations around 0.1 Hz may reflect blood vessel movement. When there are abnormally high frequency fluctuations, it often indicates that the vascular regulation function is disordered. The abnormal determination of fluctuation amplitude uses a relative standard rather than an absolute standard.

[0061] In one embodiment, the mean value of the vascular resistance index in the window is first calculated, and then 1.5 times the mean value is set as the abnormality determination threshold. This method takes into account individual differences and different baseline states. The statistics of the duration help to distinguish between occasional fluctuations and sustained abnormalities. Transient fluctuations may be normal physiological responses, while sustained large fluctuations indicate pathological conditions. The double determination criteria of the high-risk time window improve the accuracy of the warning. Excessive blood lipid concentration is an important factor in the formation of atherosclerosis, and abnormal fluctuations in vascular resistance reflect immediate changes in blood vessel function. When both occur simultaneously, it indicates that the patient is in a high-risk state.

[0062] For example, 2-4 hours after a meal, the blood lipid concentration reaches a peak, and if the fluctuation amplitude of the vascular resistance index also increases significantly at this time, it indicates that hyperlipidemia is affecting blood vessel function and needs to be closely monitored and intervention measures need to be taken.

[0063] S105, calculate the correlation index of the change of coronary flow reserve and the circulation function by the marked high-risk time window data, obtain the correlation coefficient matrix between the coronary flow reserve index and the circulation function parameters, and redistribute the weight of the multi-source signals according to the matrix to obtain the adjusted signal weight distribution.

[0064] Through the marked high-risk time window data, the ratio of the average arterial pressure in the window to the average arterial pressure in the preset time period before the meal and the ratio of the heart rate in the window to the average heart rate in the preset time period before the meal are extracted, and the product of the two is taken as the coronary flow reserve index; the myocardial oxygen consumption index is calculated according to the product of the systolic pressure and the heart rate; the vascular compliance index is obtained by dividing the difference between the current pulse pressure difference and the pulse pressure difference before the meal by the difference between the current stroke volume and the stroke volume before the meal, forming the correlation index data including the coronary flow reserve index, the myocardial oxygen consumption index and the vascular compliance index. According to the correlation index data and the multi-source signal data in the high-risk time window, the Pearson correlation coefficients of the coronary flow reserve index with the blood pressure change value, the heart rate change value and the blood oxygen saturation, the Pearson correlation coefficients of the myocardial oxygen consumption index with the ST segment deviation value and the heart rate variability, and the Pearson correlation coefficients of the vascular compliance index with the blood lipid concentration and the blood glucose concentration are calculated, and the correlation coefficient values are arranged according to the row-column correspondence to construct the correlation coefficient matrix between the coronary flow reserve index and the circulation function parameters. For each correlation coefficient in the row of the coronary flow reserve index in the correlation coefficient matrix, the absolute value of each coefficient is calculated, normalized by dividing by the sum of all absolute values as the initial weight of the corresponding circulation function parameter, and if the absolute value of the correlation coefficient of a certain parameter is lower than a preset correlation threshold, the weight of the parameter is multiplied by a preset attenuation factor, and all weights are normalized again to make the total weight meet the preset normalization condition to obtain the adjusted signal weight distribution.

[0065] Specifically, the coronary flow reserve index reflects the ability of the heart to increase blood supply under stress.

[0066] In one possible implementation, the blood pressure and heart rate are measured every 5 minutes within 30 minutes before the meal, and the average value is taken as the reference value of the basic state. During the postprandial hyperlipidemia, the increase in blood viscosity leads to an increase in coronary blood flow resistance, and the heart needs to increase arterial pressure and heart rate to maintain sufficient myocardial perfusion. When the postprandial mean arterial pressure is 1.2 times the basic value and the heart rate is 1.1 times the basic value, the coronary flow reserve index is 1.32, indicating that the heart is using 32% of its reserve capacity to cope with the challenge of hyperlipidemia. The myocardial oxygen consumption index is estimated by the product of the systolic pressure and the heart rate, and this simplified formula is based on the linear relationship between myocardial work and oxygen consumption. Systolic pressure reflects the afterload overcome by the heart during ejection, and heart rate determines the number of work times per unit time.

[0067] For example, when the systolic pressure is 140 mmHg and the heart rate is 80 beats per minute, the myocardial oxygen consumption index is 11200. Although this value is a relative value, it can reflect the dynamic trend of myocardial oxygen consumption and provide a quantitative index for evaluating cardiac burden. The calculation of vascular compliance index involves the evaluation of vascular elasticity characteristics.

[0068] It should be noted that the change in pulse pressure difference reflects the change in arterial elasticity, while the change in stroke volume reflects the adjustment of cardiac pumping function. The ratio of the two eliminates the influence of cardiac function changes and more purely reflects the compliance of the blood vessels. When the pulse pressure difference increases from 50 mmHg to 60 mmHg, and the stroke volume only increases from 70 ml to 75 ml, it indicates that the vascular compliance decreases, and a larger pressure change is needed to accommodate the same blood volume. The calculation of the Pearson correlation coefficient quantifies the degree of linear correlation between different physiological parameters.

[0069] Specifically, the correlation coefficient ranges from -1 to 1, with positive values indicating positive correlation and negative values indicating negative correlation. The closer the absolute value is to 1, the stronger the correlation.

[0070] For example, the correlation coefficient between the coronary flow reserve index and the blood pressure change value is 0.75, indicating a strong positive correlation between the two; and the correlation coefficient with blood oxygen saturation is -0.45, indicating that the higher the degree of coronary flow reserve activation, the more likely it is that blood oxygen saturation will decrease. The construction of the correlation coefficient matrix uses a systematic arrangement. The rows of the matrix represent different physiological indicators, the columns represent the circulatory function parameters, and each element is the correlation coefficient of the corresponding row and column variables. This matrix representation not only facilitates an overall understanding of the relationships between parameters, but also provides a quantitative basis for subsequent weight allocation. The dynamic adjustment mechanism of weight allocation ensures the adaptability of the monitoring system.

[0071] In one embodiment, the correlation threshold is set to 0.3, and when the absolute value of the correlation coefficient of a parameter is less than this value, it indicates that the parameter has a weak influence on the coronary flow reserve, and its weight is reduced by multiplying a decay factor of 0.5. Normalization ensures that the sum of all weights is 1, so that the weighted comprehensive index has a unified dimension.

[0072] S106, obtain an endothelial dysfunction index by analyzing the correlation between the vascular resistance index fluctuation and the blood glucose and blood lipid change curve, perform feature weighting on the endothelial dysfunction index according to the adjusted signal weight distribution, input the multi-source signal data set and the correlation coefficient matrix, and output a comprehensive risk score value.

[0073] By analyzing the blood vessel resistance index fluctuation sequence and the blood glucose and blood lipid change curve, the Pearson correlation coefficient of the two sequences in each time window is calculated as the cross-correlation coefficient by using a sliding window. If the proportion of the window number whose cross-correlation coefficient exceeds the preset positive correlation threshold value in the total window number exceeds the preset synchronization threshold value, the slope ratio of the blood vessel resistance index to the blood glucose concentration and the slope ratio of the blood vessel resistance index to the blood lipid concentration are calculated, and the larger value of the two ratio values is taken as the endothelial injury value of the window. The mean value of the endothelial injury values in all windows is the endothelial dysfunction index. According to the endothelial dysfunction index and the adjusted signal weight distribution, six parameter values of the ST segment offset value, the blood pressure change value, the heart rate variability, the blood oxygen saturation, the blood glucose concentration and the blood lipid concentration at the current moment are extracted from the multi-source signal data set, multiplied by the corresponding signal weight respectively, and summed to obtain a basic score. Then the basic score is multiplied by the endothelial dysfunction index to obtain a weighted score value. For the weighted score value and the correlation coefficient matrix, the average value of the absolute values of all non-diagonal elements in the matrix is taken as the parameter coupling strength. The weighted score value is multiplied by the parameter coupling strength and then mapped to a preset score interval. If the result exceeds the preset high-risk score value, the output comprehensive risk score value is the preset high score value. If it is lower than the preset low-risk score value, the output comprehensive risk score value is the preset low score value. Otherwise, the comprehensive risk score value is calculated according to the linear mapping relationship.

[0074] Specifically, the quantitative evaluation of vascular endothelial dysfunction is a key link of cardiovascular risk early warning.

[0075] Specifically, the vascular endothelial cells release inflammatory factors under hyperlipidemia, leading to increased vascular resistance. When blood glucose and blood lipid increase, the blood viscosity increases, and the vascular endothelial cells need to secrete more nitric oxide to maintain vasodilation, but the high-fat environment inhibits this process, forming a vicious cycle. By analyzing the synchronous changes of vascular resistance and metabolic parameters, the endothelial function status can be indirectly evaluated. The calculation of the Pearson correlation coefficient in the sliding window needs to consider the time-varying characteristics of the physiological signals.

[0076] In one possible implementation, a window length of 5 minutes is selected, and the window slides once every minute. In each window, the covariance of the blood vessel resistance index sequence and the blood glucose concentration sequence is calculated, and the correlation coefficient is obtained by dividing the product of the standard deviations of the two sequences. When the correlation coefficient exceeds 0.7, it indicates that there is a strong positive correlation between the two. If more than 80% of the windows show strong positive correlation within a 2-hour monitoring period, it indicates that the vascular function and the metabolic state are highly coupled. The slope ratio calculation reflects the sensitivity of the vascular response.

[0077] It should be noted that under normal circumstances, the blood vessels have automatic regulation ability and can maintain stable blood flow within a certain range. However, when the endothelial function is damaged, this regulation ability decreases.

[0078] For example, if blood glucose increases from 6 mmol / L to 8 mmol / L, the resistance index of normal blood vessels may only increase from 1.0 to 1.1, and the slope ratio is 0.05; while in endothelial dysfunction, the resistance index may increase from 1.0 to 1.3, and the slope ratio reaches 0.15, indicating that the blood vessels are excessively sensitive to metabolic changes. The multi-parameter weighted scoring mechanism comprehensively considers the contribution of each physiological index.

[0079] For example, the ST segment offset value is 0.2 mV, the weight is 0.25, the blood pressure change value is 20 mmHg, the weight is 0.20, the heart rate variability is 10 ms, the weight is 0.15, the blood oxygen saturation is 95%, the weight is 0.15, the blood glucose is 8 mmol / L, the weight is 0.15, and the blood lipid is 3 mmol / L, the weight is 0.10. The basic score is calculated as the sum of the product of each parameter after normalization and the weight. When the endothelial dysfunction index is 1.5, it indicates that the vascular function is deteriorated by 50%, and the basic score needs to be multiplied by 1.5 to obtain the weighted score. The introduction of parameter coupling strength reflects the overall coordination of the physiological system.

[0080] In one embodiment, the correlation coefficient matrix is a 6x6 matrix containing the correlation coefficients between each pair of six physiological parameters. The non-diagonal elements represent the mutual influence between different parameters. When the average coupling strength is 0.6, it indicates that there is a moderate degree of mutual association between parameters. This coupling reflects the normal regulation of the physiological system, and may also indicate the compensatory response in pathological state. The mapping mechanism of risk score ensures the interpretability of the output. The score interval is set to 0-100, where 0-30 is low risk, 30-70 is medium risk, and 70-100 is high risk. Through linear mapping, the weighted score value is converted to this standard interval, so that the risk levels of different patients are comparable.

[0081] S107, if the score value exceeds the preset threshold, an early warning signal of coronary heart disease is triggered according to the comprehensive risk score value and in combination with the characteristic mode of the occult stage of coronary heart disease in historical data.

[0082] If the comprehensive risk score value exceeds the preset threshold value, physiological signal data in a preset time period before the onset is extracted from the pre-established historical records of coronary heart disease patients, the difference value of ST segment offset value between adjacent time windows is calculated, the proportion of the number of windows whose absolute value of the difference value exceeds the preset change threshold value to the total number of windows is counted, and if the proportion exceeds the preset proportion threshold value, it is defined as the ST segment dynamic change mode; the longest duration of consecutive values below the preset normal lower limit in the heart rate variability index sequence is calculated as the proportion of the total monitoring time, which is taken as the heart rate variability reduction mode parameter; the blood pressure sequence is segmented by a preset time length, the standard deviation of the blood pressure value in each segment is calculated, and the proportion of the segmented segments whose standard deviation exceeds the preset fluctuation threshold value to the total number of segments is counted as the blood pressure fluctuation abnormality mode parameter, forming a feature mode set containing three mode parameters. According to the feature mode set, the ST segment offset value sequence, the heart rate variability sequence and the blood pressure sequence in the recent preset time length are extracted from the real-time monitoring data of the current patient, the absolute value of the difference between the current ST segment adjacent window difference value exceeding the threshold value and the historical ST segment dynamic change mode is calculated, the absolute value of the difference between the current heart rate variability low value duration ratio and the historical heart rate variability reduction mode parameter is calculated, and the absolute value of the difference between the current blood pressure segment standard deviation exceeding the threshold value and the historical blood pressure fluctuation abnormality mode parameter is calculated. If at least two of the three absolute difference values are less than the preset similarity threshold value, a mode matching success marker is generated. For the mode matching success marker and the continuous state of the comprehensive risk score value, if there is a mode matching success marker and the comprehensive risk score value exceeds the preset threshold value at consecutive preset sampling points, an early warning signal of coronary heart disease is output.

[0083] Specifically, the extraction of the feature mode in the occult period of coronary heart disease is based on the retrospective analysis of a large number of clinical cases.

[0084] In one possible implementation, the continuous monitoring data of the patients diagnosed with coronary heart disease is traced back to 3-6 months before the onset from their historical records. The selection of this time period is of great significance because coronary artery stenosis is a gradual process, and there are usually pathological changes for several months before the appearance of obvious symptoms. By analyzing the common characteristics of these patients, a reference template for early warning can be established. The ST segment dynamic change mode reflects the intermittent nature of myocardial ischemia.

[0085] Specifically, in the early stage of coronary stenosis, the blood vessels still have some compensatory capacity, and ischemia often occurs during activity or emotional excitement and is relieved after rest. This ischemia-reperfusion cycle causes the ST segment to exhibit characteristic fluctuations.

[0086] For example, the ST segment offset value is calculated every 10 minutes. If the offset value changes from 0.05 mV to 0.15 mV at adjacent time points, the difference is 0.1 mV, which exceeds the threshold of 0.08 mV. If more than 60% of the time windows within 2 hours show this super-threshold change, a typical dynamic change pattern is formed. The heart rate variability reduction pattern reveals the impairment of autonomic nervous function.

[0087] It should be noted that the heart rate of a normal person is not absolutely constant, but there are slight rhythmic changes, which are regulated by both sympathetic and parasympathetic nerves. When coronary artery disease affects the nerve supply to the heart, this fine regulation ability decreases. If the heart rate variability indicator is continuously lower than 5 ms for more than 70% of the monitoring time, it indicates that the autonomic nervous regulation function is severely impaired, which is an important warning signal of coronary heart disease. The abnormal blood pressure fluctuation pattern reflects the disorder of vascular regulation function. Under normal circumstances, blood pressure is maintained relatively stable through baroreceptor reflex. However, in the occult stage of coronary heart disease, arteriosclerosis leads to decreased vascular elasticity and reduced baroreceptor sensitivity. The 24-hour blood pressure data is segmented every hour, and the standard deviation in each segment is calculated. When the systolic pressure fluctuates sharply between 120-150 mmHg in a segment, and the standard deviation exceeds 15 mmHg, the segment is marked as abnormal. If more than 8 of the 24 segments are abnormal, an abnormal blood pressure fluctuation pattern is formed. The similarity calculation of pattern matching uses the parameter difference method instead of the traditional sequence alignment. The advantage of this method is to reduce the influence of individual differences and focus on the change rule rather than the absolute value.

[0088] For example, the current ST segment dynamic change ratio of a patient is 55%, the historical pattern is 60%, the difference is only 5%, which is less than the similarity threshold of 10%; the heart rate variability low value duration is 65%, the historical pattern is 70%, the difference is 5%; the abnormal blood pressure fluctuation ratio is 30%, the historical pattern is 33%, the difference is 3%. All three indicators meet the similarity condition, triggering the early warning. The continuity judgment mechanism avoids false positives caused by occasional events.

[0089] In one embodiment, the comprehensive risk score is required to exceed the threshold for 30 consecutive sampling points to trigger the early warning. This is equivalent to a high-risk state for 30 consecutive minutes, excluding the interference of transient stress reactions.

[0090] S108, store the current high-risk time window features and blood glucose and blood lipid curve change trends through the triggered early warning signal of coronary heart disease, to obtain structured risk profile data.

[0091] Through the triggered early warning signal of coronary heart disease, a preset data acquisition process is activated, the start timestamp and the end timestamp of a high-risk time window, the maximum ST segment offset, the standard deviation of blood pressure fluctuation, and the minimum heart rate variability in the window are extracted from the real-time monitoring data cache, and the blood glucose concentration sequence and the blood lipid concentration sequence in the corresponding time period are obtained, the blood glucose change rate and the blood lipid change rate are obtained by calculating the difference between the last value and the first value of the sequence divided by the time interval, and the data set to be stored is formed by combining the time mark, the physiological parameter characteristic value, and the metabolic change rate. According to the data set to be stored, the data is arranged in the order of patient identification, early warning trigger time, comprehensive risk score value, maximum ST segment offset, standard deviation of blood pressure fluctuation, minimum heart rate variability, blood glucose change rate, and blood lipid change rate, and a data record is constructed in a key-value pair mode, each data item is configured with a corresponding data type identifier and unit identifier, and a risk record entry with a unified format is generated. For the risk record entry, the cyclic redundancy check value of all numerical fields is calculated as the data integrity identifier, the current system timestamp is added as the record creation time, the risk record entry is serialized into a binary format and written into a storage device, and structured risk profile data is obtained.

[0092] Specifically, the data acquisition process after the early warning signal is triggered reflects the rapid response capability of the system.

[0093] In a possible implementation, when the comprehensive risk score exceeds the threshold value and triggers an early warning, the system immediately starts a data saving process. Real-time monitoring data is usually stored in a ring buffer, which allows continuous writing of new data while retaining historical data for a recent period of time. When the early warning is triggered, the system locks the current buffer position and extracts the complete data of the high-risk time window by backtracking. The calculation of the change rate uses a simplified difference method, which has practical significance.

[0094] Specifically, the blood glucose rises from 6.5 mmol / L at the beginning of the window to 8.2 mmol / L at the end, with a time span of 30 minutes and a change rate of 0.057 mmol / L per minute. Although this linear approximation ignores the details of the fluctuations in between, it can quickly reflect the overall trend. For clinical decision-making, it is more important to understand whether the blood glucose is rising rapidly or changing slowly than the precise value at each moment. The design of the data cache takes into account the real-time requirements of medical devices.

[0095] It should be noted that physiological signal monitoring devices usually use double buffering or multi-buffering mechanism, one buffer is used for receiving new data, and the other is used for processing and transmission. When the early warning is triggered, the system needs to extract historical data without interrupting real-time monitoring. Through the management of the buffer pointer, the data segment corresponding to the high-risk time window can be accurately located. The choice of the key-value pair data structure is based on the heterogeneous characteristics of medical data.

[0096] For example, "patient identification" corresponds to an ID number of string type, "ST segment offset max" corresponds to a floating point number and is marked with unit of millivolt, and "warning trigger time" corresponds to a timestamp format. Such structure not only ensures self-descriptiveness of data, but also facilitates subsequent query and analysis. Each key-value pair contains semantic information of data, avoiding ambiguity that may be caused by pure numerical sequence. Calculation of cyclic redundancy check value ensures data integrity.

[0097] In an embodiment, all numerical fields are sequentially connected into a byte stream, and a 32-bit check code is calculated through polynomial division. When data is read from a storage device, the check code is recalculated and compared with the stored value, so that errors in transmission or storage process can be detected. This is particularly important for medical data, because any data damage may affect clinical judgment. Binary serialization improves storage efficiency and reading speed. Compared with text format, binary format can compress a double-precision floating point number from possible 20 characters to 8 bytes. For a risk profile containing a large number of physiological parameters, such compression can significantly reduce storage space requirement. At the same time, binary format has faster parsing speed, which is beneficial to subsequent batch data analysis and historical trend mining. Structured storage lays a foundation for subsequent data utilization. Unified data format enables risk profiles of different times and different patients to be compared horizontally. Doctors can quickly retrieve all high-risk events in a specific time period and analyze their common characteristics; researchers can conduct statistical analysis based on a large amount of structured data to discover new risk factors.

[0098] The preferred embodiments of the present application are described above with reference to the drawings, but the patent scope of the present application is not limited to the preferred embodiments. Any equivalent structure or equivalent process transformation based on the content of the specification and drawings, or direct or indirect application in other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A method for precise prediction of coronary heart disease by multi-source heterogeneous data integration, characterized in that, The method comprises: Collecting physiological signals of a patient, processing the physiological signals to obtain a multi-source signal data set, extracting features from the multi-source signal data set to obtain target time period features in a postprandial hyperlipidemia window period, labeling abnormal fluctuations of ST segment change features and the basic circulatory function parameters, and determining a myocardial perfusion pressure abnormal interval; obtaining the myocardial perfusion pressure abnormal interval data, combining blood glucose and blood lipid data, and synchronously matching to form a joint data set; analyzing the joint data set, extracting the fluctuation amplitude of the vascular resistance index in the postprandial hyperlipidemia window period, and marking a high-risk time window whose fluctuation amplitude exceeds a preset range; calculating the correlation coefficient matrix of the coronary flow reserve index in the high-risk time window data and the basic circulatory function parameters, adjusting the weight distribution of the multi-source signal data set; analyzing the correlation between the vascular resistance index fluctuation and the blood glucose and blood lipid data to obtain an endothelial function abnormality index, weighting the endothelial function abnormality index and the multi-source signal data set, and outputting a comprehensive risk score value; outputting a coronary heart disease early warning signal according to the comprehensive risk score value and historical coronary heart disease occult period characteristic patterns; storing the features of the high-risk time window and the blood glucose and blood lipid data, and generating structured risk profile data. 2.The method of claim 1, wherein, The collecting physiological signals of a patient by a wearable device, processing the physiological signals to obtain a multi-source signal data set comprises: Collecting electrocardiogram signals, determining voltage difference and slope change values of ST segment change features, collecting blood pressure values, blood oxygen saturation percentages, and heart rate values of basic circulatory function parameters, and generating original data streams containing time stamps; filtering baseline drift and power frequency interference from the electrocardiogram signals in the original data streams, removing pulse noise from the blood pressure values, smoothing transient fluctuations of the blood oxygen saturation percentages, and removing abnormal values to obtain denoised data sequences; normalizing the ST segment voltage difference, blood pressure values, heart rate values, and blood oxygen saturation percentages of the denoised data sequences to generate the multi-source signal data set with uniform dimensions. 3.The method of claim 1, wherein, The extracting features from the multi-source signal data set to obtain target time period features in a postprandial hyperlipidemia window period, labeling abnormal fluctuations of ST segment change features and the basic circulatory function parameters, and determining a myocardial perfusion pressure abnormal interval comprises: Segmenting the multi-source signal data set, calculating voltage offset values of ST segment change features in a postprandial time window, recording ST segment elevation or depression events, extracting peak ratios and time spans of T wave morphology changes, calculating R-R interval mean and square root of difference sum to obtain heart rate variability; reducing the dimensions of ST segment elevation amplitude, T wave morphology changes, and heart rate variability, retaining principal components, and generating target time period features; positioning abnormal values of basic circulatory function parameters according to the target time period features, and generating time sequence sequences containing ST segment events and abnormal markers; calculating the product of ST segment offset values and blood pressure difference values in the time sequence sequences, and determining an interval that is continuously below a threshold value as the myocardial perfusion pressure abnormal interval. 4.The method of claim 1, wherein, The obtaining the myocardial perfusion pressure abnormal interval data, combining blood glucose and blood lipid data, and synchronously matching to form a joint data set comprises: Extract the ST segment offset value sequence and blood pressure change value sequence of the myocardial perfusion pressure abnormal interval, obtain the concentration value sequence of blood glucose and blood lipid data, interpolate the concentration value sequence of blood glucose and blood lipid data to generate an equidistantly sampled time sequence, align the time sequence with the time stamp of the myocardial perfusion pressure abnormal interval data, and resample to generate a synchronized metabolic parameter sequence, and combine the ST segment offset value sequence, the blood pressure change value sequence, the blood glucose concentration value sequence, and the blood lipid concentration value sequence to form the joint data set.

5. The method of claim 1, wherein, The analysis of the joint data set extracts the fluctuation amplitude of the vascular resistance index in the postprandial hyperlipidemia window period, and marks the high-risk time window whose fluctuation amplitude exceeds the preset range, including: Extract the blood pressure change value sequence from the joint data set, calculate the pulse pressure difference and mean arterial pressure, estimate the cardiac output combined with the heart rate sequence, generate the vascular resistance index sequence, calculate the difference between the maximum and minimum values of the vascular resistance index sequence in the window to determine the fluctuation amplitude, and extract the blood lipid concentration value from the joint data set. The window whose fluctuation amplitude exceeds the preset range and whose blood lipid concentration value exceeds the threshold is marked as the high-risk time window. 6.The method of claim 1, wherein, The calculation of the correlation coefficient matrix of the coronary flow reserve index in the high-risk time window data and the basic circulation function parameter adjusts the weight distribution of the multi-source signal data set, including: Calculate the product of the mean arterial pressure and the heart rate ratio in the high-risk time window data to generate the coronary flow reserve index, calculate the correlation coefficient of the coronary flow reserve index and the basic circulation function parameter to construct the correlation coefficient matrix, and normalize the absolute value of the coefficients of the correlation coefficient matrix to adjust the coefficient weight below the threshold to generate the weight distribution.

7. The method of claim 1, wherein, The analysis of the correlation between the vascular resistance index fluctuation and the blood glucose and blood lipid data obtains the endothelial function abnormality index, and the weighted endothelial function abnormality index and the multi-source signal data set output the comprehensive risk score value, including: Calculate the correlation coefficient of the vascular resistance index and the blood glucose and blood lipid data in the window to determine the endothelial damage value to generate the endothelial function abnormality index, and the weighted sum of the ST segment offset value, blood pressure change value, heart rate variability, blood oxygen saturation, blood glucose concentration value, and blood lipid concentration value of the multi-source signal data set is combined with the endothelial function abnormality index to generate a weighted score value. According to the mean value of the off-diagonal elements of the correlation coefficient matrix, the weighted score value is mapped to the preset interval to output the comprehensive risk score value. 8.The method of claim 1, wherein, According to the comprehensive risk score value and the historical coronary heart disease occult period characteristic mode, the coronary heart disease early warning signal is output, including: Extract the ST segment offset value difference ratio, heart rate variability low value duration ratio, and blood pressure standard deviation threshold value from historical data to generate a feature mode set, extract the ST segment offset value sequence, heart rate variability sequence, and blood pressure sequence from real-time data, calculate the absolute value of the difference from the feature mode set to generate a pattern matching label, and output the coronary heart disease early warning signal according to the pattern matching label and the sustained state of the comprehensive risk score value. 9.The method of claim 1, wherein, The storage of the high-risk time window features and the blood glucose and blood lipid data generates structured risk profile data, including: Extracting the ST segment offset maximum value, blood pressure fluctuation standard deviation, and heart rate variability minimum value of the high-risk time window, calculating the change rate of the blood glucose and blood lipid data, generating a data set to be stored; arranging the risk record entries according to the patient identification, early warning time, comprehensive risk score value, ST segment offset maximum value, blood pressure fluctuation standard deviation, heart rate variability minimum value, blood glucose change rate, and blood lipid change rate; serializing the risk record entries, adding a check value and a timestamp, and generating the structured risk profile data.

Images