Intelligent rehabilitation monitoring method and device for heart failure patients

Abstract

This application discloses an intelligent rehabilitation monitoring method and device for heart failure patients. The invention relates to the field of rehabilitation data processing technology. The method includes: obtaining a time-series data set of monitoring items based on multiple acquisition time scales, including time-series datasets corresponding to each monitoring item at multiple acquisition time scales; extracting health feature values ​​from the time-series datasets of the monitoring items to generate feature value sequences corresponding to different acquisition time scales; constructing a multivariate time-series sequence at multiple time scales, and quantifying the physiological system elasticity coefficient value of heart failure patients using a preset event elasticity extraction mechanism; inputting the multivariate time-series sequence and elasticity coefficient values ​​as monitoring features into a heart failure rehabilitation trajectory prediction model, outputting a health index sequence within a future intervention window, and identifying elasticity loss thresholds as early intervention points. This improves the accuracy and foresight of monitoring the rehabilitation health trends of heart failure patients.

Description

Technical Field

[0001] This invention relates to the field of rehabilitation data processing technology, and in particular to an intelligent rehabilitation monitoring method and device for heart failure patients. Background Technology

[0002] Heart failure is a complex clinical syndrome characterized by high morbidity, high readmission rates, and high mortality. With the accelerating aging of the population, the number of heart failure patients continues to rise, placing a heavy burden on the healthcare system. The rehabilitation management of heart failure patients is a long and complex process, requiring continuous monitoring of physiological indicators and timely adjustments to treatment plans.

[0003] Existing technologies for health monitoring in patient rehabilitation training mainly rely on single sensors or simple multi-sensor fusion methods. This usually results in insufficient completeness and reliability of monitoring data, and most of them focus on data collection and status assessment at the current moment, lacking the ability to predict the evolution trend of heart failure pathological state. Acute decompensation in heart failure patients often has a gradual development process, and the transition from the stable period to the pre-decompensation period is usually accompanied by specific physiological pattern changes. Summary of the Invention

[0004] This application provides an intelligent rehabilitation monitoring method and device for heart failure patients, which improves the accuracy and foresight of monitoring the rehabilitation health trends of heart failure patients and reduces the risk of acute attacks.

[0005] This application provides an intelligent rehabilitation monitoring method for heart failure patients, including:

[0006] S101, based on the preset multiple acquisition time scales, obtain the time series data of the heart failure patient corresponding to each acquisition time scale in the preset time period for each acquisition time scale, and form a time series data set of the monitoring item, including the time series dataset corresponding to each monitoring item in the multiple acquisition time scales.

[0007] S102, extract health feature values ​​from the time series dataset of each monitoring item to generate feature value sequences corresponding to different collection time scales. The feature value sequences are composed of the health feature values ​​of each collection point at the corresponding collection time scale.

[0008] S103, based on the feature value sequence of all monitoring items at each acquisition time scale, construct a multivariate time series sequence with multiple time scales, including the multivariate time series sequence matrix corresponding to each time scale;

[0009] S104, based on multi-timescale multivariate time series, uses a preset event elasticity extraction mechanism to quantify the physiological system elasticity coefficient value of heart failure patients.

[0010] S105 uses multi-timescale multivariate time series and elasticity coefficient values ​​as monitoring features to input into the pre-trained heart failure rehabilitation trajectory prediction model, and outputs a health index sequence within the future intervention window;

[0011] S106 identifies the critical point of elasticity loss in the health index sequence and uses it as an early intervention point, which is then synchronized with monitoring personnel.

[0012] Preferably, the time-series data set of the monitoring items is represented as follows: P represents the total number of monitored items. This is the time-series dataset for the P-th monitoring item, which includes time-series data corresponding to multiple preset acquisition time scales. K represents the number of preset data acquisition time scales. This refers to the time series data corresponding to the Kth collection time scale of this monitoring item.

[0013] Preferably, the preset multiple acquisition time scales are set to three acquisition time scales, namely high-frequency scales. Mid-frequency scale Low-frequency scale The time series dataset of each monitoring item in the monitoring item time series data set is divided into three collection time scales: high-frequency physiological parameters collected at the first preset time interval, medium-frequency trend parameters collected at the second preset time interval, and low-frequency cumulative parameters collected at the third preset time interval, wherein the first preset time interval is shorter than the second preset time interval, and the second preset time interval is shorter than the third preset time interval.

[0014] Preferably, the extraction of health feature values ​​from the time-series dataset for each monitoring item includes:

[0015] All time-series data are preprocessed, including data cleaning and normalization.

[0016] For time-series data of each monitoring item at different collection time scales, health feature values ​​are extracted from the data at each collection point to generate a feature value sequence at each collection time scale.

[0017] Preferably, the multi-timescale multivariate time series includes a multivariate time series sequence matrix corresponding to the high-frequency scale. Multivariate time series sequence matrix corresponding to the mid-frequency scale Multivariate time series sequence matrix corresponding to low-frequency scale .

[0018] Preferably, S103 specifically includes:

[0019] For each time scale The multivariate time-series sequence matrix is ​​constructed as follows: for the time scale... , build The dimensional sequence matrix serves as the multivariate time series sequence matrix corresponding to this time scale:

[0020]

[0021] in, Time scale The length of the feature value sequence is given below, where P is the number of monitoring items. For monitoring items At any moment Health characteristic values, ∈[ , ].

[0022] Preferably, the preset event elastic extraction mechanism specifically includes:

[0023] S201, based on the feature value sequence corresponding to the mid-frequency scale, identify the perturbation events in it, and extract the data segment corresponding to each perturbation event in the feature value sequence of the high-frequency scale;

[0024] S202, based on the data segment corresponding to each perturbation event in the extracted high-frequency scale feature value sequence, calculate its health feature value recovery speed and quantify the perturbation resistance ability of heart failure patients;

[0025] S203, Calculate the elasticity coefficient of the physiological system based on the recovery rate of the health characteristic values ​​corresponding to the disturbance event.

[0026] Preferably, the criteria for determining the disturbance event are set as follows:

[0027] For each monitoring item's feature value sequence, identify all consecutive intervals where the health feature value exceeds the preset health feature threshold interval as disturbance events.

[0028] Preferably, the recovery rate of the health characteristic value is calculated according to the following formula:

[0029]

[0030] in, Let be the recovery rate for the i-th disturbance event. This represents the peak value of the health characteristic value of the disturbance event in the corresponding data segment. These are the stable characteristic values ​​following the disturbance event in the corresponding data segment. and They are respectively and The corresponding moment;

[0031] The elasticity coefficient of the physiological system is calculated according to the following formula:

[0032]

[0033] Where E is the elasticity coefficient and M is the total number of disturbance events. Let i be the recovery rate for the i-th disturbance event. This represents the maximum recovery rate for this heart failure patient.

[0034] This application also provides an intelligent rehabilitation monitoring device for heart failure patients, including: a multi-source sensor, a feature extraction module, and a prediction module;

[0035] The multi-source sensor is used to obtain the time-series data of the heart failure patient corresponding to each acquisition time scale within a preset time period based on preset multiple acquisition time scales, forming a time-series data set of the monitoring item, including the time-series dataset corresponding to each monitoring item at multiple acquisition time scales.

[0036] The feature extraction module is used to extract health feature values ​​from the time series dataset of each monitoring item, generating feature value sequences corresponding to different collection time scales. The feature value sequence is composed of the health feature values ​​of each collection point at the corresponding collection time scale. Based on the feature value sequences of all monitoring items at each collection time scale, a multivariate time series with multiple time scales is constructed, including a multivariate time series sequence matrix corresponding to each time scale.

[0037] The prediction module is based on multi-timescale multivariate time series and uses a preset event elasticity extraction mechanism to quantify the physiological system elasticity coefficient value of heart failure patients. The multivariate time series and elasticity coefficient value are used as monitoring features and input into the pre-trained heart failure rehabilitation trajectory prediction model to output the health index sequence within the future intervention window. The elasticity loss threshold of the health index sequence is identified as an early intervention point and synchronized to the monitoring personnel.

[0038] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0039] By collecting patients' physiological data at different time scales using multiple sensors, constructing multivariate time series after preprocessing and feature extraction, quantifying the elasticity coefficient of the physiological system, and then using a prediction model to output a future health index series, finally identifying the critical point of elasticity loss and generating intervention suggestions, its role is to comprehensively and accurately monitor the health status of heart failure patients, predict health change trends in advance, promptly identify patients' health problems and provide professional intervention guidance, and improve the rehabilitation effect of heart failure patients.

[0040] Based on multi-timescale multivariate time series and utilizing a pre-defined event elasticity extraction mechanism, this study quantifies the physiological system elasticity coefficient of heart failure patients. By identifying mid-frequency scale perturbation events, extracting corresponding high-frequency scale data segments, and calculating the recovery speed of health characteristic values, the physiological system elasticity coefficient is quantified. This approach accurately assesses the perturbation resistance of heart failure patients, providing crucial information for determining their health status. Specifically, the method of identifying perturbation events based on mid-frequency scale characteristic value sequences, extracting corresponding high-frequency scale data segments, and quantifying the patient's perturbation resistance by calculating the recovery speed of health characteristic values, thereby obtaining the physiological system elasticity coefficient, overcomes the limitations of traditional single-timescale analysis. It provides a more comprehensive and accurate assessment of the physiological system elasticity of heart failure patients, offering a unique and effective basis for determining their health status and developing intervention strategies. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating an intelligent rehabilitation monitoring method for heart failure patients according to an embodiment of the present invention.

[0042] Figure 2 This is a schematic diagram of the structure of an intelligent rehabilitation monitoring device for heart failure patients according to an embodiment of the present invention. Detailed Implementation

[0043] To facilitate understanding of the present invention, a more complete description of this application will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the present invention.

[0044] It should be noted that the terms "vertical," "horizontal," "up," "down," "left," "right," and similar expressions used in this article are for illustrative purposes only and do not represent the only possible implementation.

[0045] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0046] Example 1: Figure 1 This is a flowchart illustrating an intelligent rehabilitation monitoring method for heart failure patients according to an embodiment of the present invention.

[0047] like Figure 1 As shown, an intelligent rehabilitation monitoring method for heart failure patients includes the following steps:

[0048] S101, based on preset multiple acquisition time scales, obtains the time series data of the heart failure patient corresponding to each acquisition time scale in the preset time period for each acquisition time scale, and assembles all the time series data into a time series data set of the monitoring item, including the time series dataset corresponding to each monitoring item in the multiple acquisition time scales.

[0049] Specifically, by deploying multiple sources of sensors (any number of which are electrocardiogram sensors, pulse oximeters, respiration sensors, accelerometers, and bioimpedance sensors), a time-series dataset of monitoring items for heart failure patients at multiple acquisition time scales is obtained over a preset time period. P represents the total number of monitored items. Let be the time-series dataset for the P-th monitoring item. The monitoring items include any number of heart rate, blood oxygen saturation, respiratory rate, exercise tolerance, and edema index. The time-series dataset for each monitoring item includes time-series data corresponding to multiple preset acquisition time scales. That is, the time-series data corresponding to each acquisition time scale constitutes the time-series dataset for that monitoring item. K represents the number of preset data acquisition time scales. This refers to the time series data corresponding to the Kth collection time scale of the monitoring item, that is, the time series data collected within the preset time period according to the collection time scale and based on the corresponding time interval.

[0050] It should be noted that the preset time period can be set to the nearest day, or it can be dynamically adjusted according to the actual situation.

[0051] In this embodiment of the invention, the preset multiple acquisition time scales are set to three acquisition time scales, namely, a high-frequency scale. Mid-frequency scale Low-frequency scale The time-series dataset for each monitoring item in the monitoring item time-series data set is divided into three collection time scales: high-frequency physiological parameters collected at a first preset time interval, mid-frequency trend parameters collected at a second preset time interval, and low-frequency cumulative parameters collected at a third preset time interval. The first preset time interval is shorter than the second preset time interval, the second preset time interval is shorter than the third preset time interval, and the third preset time interval is much shorter than the preset time period. That is, within the preset time period, based on different collection time scales, time-series data for different monitoring items are periodically collected at corresponding preset time intervals, forming the monitoring item time-series data set.

[0052] For example, among the three acquisition time scales, the first preset time interval can be set to 1 minute, the second preset time interval can be set to 1 hour, and the third preset time interval can be set to 6 hours. These can be dynamically adjusted according to actual monitoring conditions and needs to ensure the capture of complete physiological information from instantaneous fluctuations to long-term trends. It should be noted that the difference between the parameters acquired at different time scales lies only in the difference in acquisition frequency or acquisition period; therefore, they are defined as high-frequency physiological parameters, mid-frequency trend parameters, and low-frequency cumulative parameters to distinguish different acquisition frequencies.

[0053] S102, extract health feature values ​​from the time series dataset of each monitoring item to generate feature value sequences corresponding to different collection time scales. The feature value sequences are composed of the health feature values ​​of each collection point at the corresponding collection time scale.

[0054] Specifically, health feature values ​​are extracted from the time-series dataset for each monitoring item, including:

[0055] All time-series data are preprocessed, including data cleaning and normalization.

[0056] For time-series data of each monitoring item at different collection time scales, health feature values ​​are extracted from the data at each collection point to generate a feature value sequence at each collection time scale.

[0057] For any monitoring item's time series dataset At any acquisition time t, the feature extraction function preset for this monitoring item is used. A scalar characteristic value is calculated from its raw data and used as the health characteristic value of the monitored item at that moment. , This represents the raw data value of the monitored item at time t.

[0058] It should be noted that, for this monitoring item, the health characteristic value of the data value at a certain collection time (which can be understood as the data value currently monitored by this monitoring item reflecting the health status of the heart failure patient) can be evaluated using data within a historical time window (which can be set according to the monitoring item, for example, set to 5 minutes), such as the rate of change, fluctuation, and change cycle. The feature extraction methods of relevant existing technologies can be referred to, and this invention will not elaborate on this.

[0059] For example:

[0060] For heart rate monitoring, the standard deviation of heart rate variability is extracted as a health characteristic value, reflecting the state of autonomic nervous system function. The peak value of the R wave is detected from continuous electrocardiogram signals within a historical time window, and the time interval between adjacent R waves (RR interval) is calculated. The standard deviation is then calculated based on the RR interval sequence.

[0061]

[0062] in, The RR interval of the i-th sinus beat is a series of consecutive heartbeat intervals extracted within the historical time window (smaller than a preset time period; experts can set the size of the historical time window according to the characteristics of the monitoring items) at the acquisition time t. It represents the time interval between two consecutive R waves. Let N be the arithmetic mean of all RR intervals within this historical time window, and let N be the total number of RR intervals within this historical time window. (Right now () represents the health characteristic value of the heart rate monitoring item, reflecting the overall degree of heart rate variability.

[0063] For blood oxygen monitoring, the fluctuation range is extracted as a health characteristic value to assess oxygenation stability. Within a historical time window, the range of blood oxygen saturation values ​​is calculated.

[0064]

[0065] in, This represents the blood oxygen saturation measurement at time t. Set a preset historical time window (e.g., the past 5 minutes). , These respectively represent the preset historical time window. The maximum and minimum values ​​of internal blood oxygen saturation. (Right now The range value within this time window represents the health characteristic value of the blood oxygen monitoring item, reflecting the stability of blood oxygen levels.

[0066] For respiratory monitoring items, the dominant period is extracted as a health characteristic value using the autocorrelation function to quantify respiratory rhythm. Time-series data is then analyzed using the autocorrelation function to identify the most significant periodic component.

[0067]

[0068]

[0069] in, It is the autocorrelation function. This represents the i-th data point within the historical time window (the respiratory rate value at the i-th time point). This represents the average respiratory rate within that historical time window. The length of this historical time window, For lag parameters, Search for a preset respiratory cycle range (e.g., a respiratory rate of 8-30 breaths / minute). (Right now The lag time corresponding to the maximum absolute value of the autocorrelation function is the dominant respiratory cycle, which is used as a health characteristic value.

[0070] For the activity endurance item (it should be noted that this monitoring item is only included when heart failure patients are monitored in a standing dynamic state), gait cycle and walking speed are extracted as health characteristics:

[0071] The gait cycle during walking is detected by an accelerometer, and the average time interval between consecutive steps is calculated.

[0072] , Let Y be the ground contact time at step y (determined by detecting the peak value of the acceleration signal), and Y be the total number of steps detected within the historical time window. The average gait cycle reflects the stability of walking rhythm;

[0073] By combining accelerometer and gyroscope data, the average walking speed is calculated through step count and displacement assessment.

[0074] Y represents the total number of steps within the historical time window, and SL represents the average step length, estimated using height and gait characteristics, for example... , Total walking time Total walking distance, Average walking speed reflects activity endurance level and serves as a health characteristic.

[0075] For the edema index, the rate of change of bioimpedance is calculated as a health characteristic value. The relative rate of change of impedance value at sampling time t is calculated by bioimpedance spectroscopy measurement.

[0076] , This is the most recent bioimpedance value from the corresponding acquisition time t (usually the impedance value at a frequency of 50kHz). This is the bioimpedance value from the previous measurement. This represents the difference in time between two bioimpedance measurements. This represents the rate of change of bioimpedance; a negative value indicates a decrease in impedance, suggesting an increase in tissue fluid volume.

[0077] It should be noted that the feature values ​​of all monitoring items are calculated separately according to the preset collection time scale, forming a multi-time scale feature value sequence, which provides comprehensive input features for subsequent health index prediction.

[0078] S103, based on the eigenvalue sequences of all monitored items at each time scale, constructs a multivariate time series sequence with multiple time scales, including a multivariate time series sequence matrix corresponding to high-frequency scales. Multivariate time series sequence matrix corresponding to the mid-frequency scale Multivariate time series sequence matrix corresponding to low-frequency scale .

[0079] In some embodiments, S103 specifically includes:

[0080] Each eigenvalue sequence is organized into a structured matrix for each time scale. , build Multidimensional time series sequence matrix ,in The length of the feature value sequence at this time scale is given by P, and the number of monitoring items is given by P. Each row of the matrix corresponds to a collection point (collection time point), and each column corresponds to the feature value sequence of a monitoring item, forming a complete multivariate time series representation.

[0081] Specifically, the multivariate time series sequence matrix is ​​constructed as follows:

[0082] For time scale , build The dimensional sequence matrix serves as the multivariate time series sequence matrix corresponding to this time scale:

[0083]

[0084] in, Time scale The length of the feature value sequence (i.e., the total number of acquisition times in the sequence), where P is the number of monitoring items. For monitoring items At any moment The health characteristic value of i, i in 1 to Between ranges, ∈[ , ].

[0085] S104, based on multi-timescale multivariate time series, uses a preset event elasticity extraction mechanism to quantify the physiological system elasticity coefficient value of heart failure patients.

[0086] In some embodiments, the preset event elastic extraction mechanism specifically includes:

[0087] S201, based on the feature value sequence corresponding to the mid-frequency scale, identifies the perturbation events in it, and extracts the data segment corresponding to each perturbation event in the feature value sequence of the high-frequency scale.

[0088] Specifically, the criteria for judging disturbance events are set as follows: for the feature value sequence of each monitoring item on the mid-frequency scale, all continuous intervals in which the health feature value exceeds the preset health feature threshold interval (it should be noted that for different monitoring items, there are corresponding health feature threshold intervals set in advance to measure the abnormality of the corresponding monitoring item, which are determined based on historical data and expert experience) are identified as disturbance events.

[0089] Specifically, the data segment extraction method is as follows: For each disturbance event, a data segment consisting of a first window before the disturbance event, the period of the disturbance event, and a second window after the disturbance event is extracted from the feature value sequence at the high-frequency scale. The first and second windows are set according to the actual monitoring situation. The first window is smaller than the second window, and the first window needs to be larger than a first preset time interval corresponding to the high-frequency scale. For example, the first window is set to 1 hour, and the second window is set to 2 hours.

[0090] S202, based on the data segment corresponding to each perturbation event in the extracted high-frequency scale feature value sequence, calculates the recovery speed of its healthy feature values, quantifying the perturbation resistance of heart failure patients:

[0091]

[0092] in, Let be the recovery rate for the i-th disturbance event. This represents the peak characteristic value of the disturbance event in the corresponding data segment. This refers to the stable characteristic value after the disturbance event in the corresponding data segment (i.e., the health characteristic value corresponding to the two consecutive health characteristic values ​​monitored after the disturbance event do not exceed the preset health characteristic threshold range). and They are respectively and The corresponding moment.

[0093] S203. Based on the recovery speed of the health characteristic values ​​corresponding to the disturbance event, calculate the elasticity coefficient value of the physiological system. The larger the value, the better the elasticity.

[0094] Specifically, the elasticity coefficient of the physiological system is calculated using the following formula:

[0095]

[0096] Where E is the elasticity coefficient and M is the total number of disturbance events. Let i be the recovery rate for the i-th disturbance event. The maximum recovery rate corresponding to this heart failure patient (determined based on historical data).

[0097] S105 takes the multivariate time series and elasticity coefficient values ​​at multiple time scales as monitoring features and inputs them into the pre-trained heart failure rehabilitation trajectory prediction model, outputting a health index sequence within the future intervention window.

[0098] In some embodiments, the pre-trained heart failure rehabilitation trajectory prediction model adopts a Transformer architecture, with input being a multivariate time series at multiple time scales. The elasticity coefficient value E is used to learn multi-scale temporal dependencies through a self-attention mechanism, and the output is a health index sequence within a future intervention window (which can be set according to the actual monitoring situation, for example, the intervention window can be set to 1 hour).

[0099] Specifically, the method for obtaining the heart failure rehabilitation trajectory prediction model is as follows:

[0100] A1. Obtain a large set of time-series monitoring data of heart failure patients within a preset time period in history, and obtain historical monitoring characteristics and corresponding health index sequences within the actual intervention window.

[0101] It should be noted that, based on each historical monitoring feature, the corresponding health index sequence within the actual intervention window can be based on expert evaluation and annotation. That is, the health index of each collection point is evaluated on the health feature values ​​of different monitoring items within the high-frequency collection time scale of the future time window (monitoring the predetermined intervention window) of the heart failure patient (quantified between 0 and 1, the larger the value, the better the recovery status of the heart failure patient). Alternatively, the health feature values ​​of multiple monitoring items can be uniformly processed (for health indexes with the same evaluation direction, for example, the larger the health feature value, the higher the health indicator, indicating the same direction), and a simple weighted fusion can be performed to determine the weights. The weights are set based on the clinical guidelines for heart failure, which will not be elaborated on in this invention.

[0102] A2. Label historical monitoring features using health index sequences to obtain a training dataset. Train the pre-selected neural network structure, continuously optimize the model, and generate the final heart failure rehabilitation trajectory prediction model.

[0103] S106 identifies the critical point of elasticity loss in the health index sequence as an early intervention point. Combined with the health characteristic values ​​of each monitoring item of the heart failure patient at the current moment, a health status report is generated and synchronized to the monitoring personnel to generate professional intervention suggestions, i.e., early prevention.

[0104] In some embodiments, the elasticity loss threshold is identified as follows:

[0105] A critical point is identified based on a health index sequence over a time series when the following conditions are met:

[0106] Each health index in the health index sequence is traversed sequentially, and the rate of change of the health index is calculated. When the rate of change is lower than the preset negative health change threshold (set based on historical data and expert experience to measure the health change situation; the negative health change threshold represents the maximum bottom line of the downward trend of the health index) and the physiological system elasticity coefficient is lower than the preset elasticity threshold (set based on expert experience and physical assessment data of heart failure patients to represent the degree of the heart failure patient's resistance to interference), the time point is considered to be the critical point of loss of elasticity.

[0107] In summary, by using multi-source sensors and based on preset multiple acquisition time scales, time-series data of various monitoring items of heart failure patients within a preset time period can be acquired, forming a time-series data set of monitoring items. By setting multiple acquisition time scales (high frequency, medium frequency, and low frequency), the complete physiological information of heart failure patients from instantaneous fluctuations to long-term trends can be comprehensively captured, providing rich data for subsequent analysis.

[0108] Health feature values ​​are extracted from the time series dataset of each monitoring item to generate feature value sequences corresponding to different collection time scales. Data preprocessing ensures data quality. Health feature values ​​are extracted from data at different collection time scales to accurately reflect the health status of each monitoring item at different time scales, providing a foundation for the subsequent construction of multivariate time series.

[0109] Based on the feature value sequences of all monitoring items at each time scale, a multivariate time series sequence with multiple time scales is constructed. The feature value sequences are organized into a structured matrix to form a complete multivariate time series representation, which facilitates the subsequent quantification of the elasticity coefficient of the physiological system using a preset mechanism and input into the heart failure rehabilitation trajectory prediction model.

[0110] Based on multi-timescale multivariate time series and utilizing a pre-defined event elasticity extraction mechanism, this study quantifies the physiological system elasticity coefficient of heart failure patients. By identifying mid-frequency scale perturbation events, extracting corresponding high-frequency scale data segments, and calculating the recovery speed of health characteristic values, the physiological system elasticity coefficient is quantified. This approach accurately assesses the perturbation resistance of heart failure patients, providing crucial information for determining their health status. Specifically, the method of identifying perturbation events based on mid-frequency scale characteristic value sequences, extracting corresponding high-frequency scale data segments, and quantifying the patient's perturbation resistance by calculating the recovery speed of health characteristic values, thereby obtaining the physiological system elasticity coefficient, overcomes the limitations of traditional single-timescale analysis. It provides a more comprehensive and accurate assessment of the physiological system elasticity of heart failure patients, offering a unique and effective basis for determining their health status and developing intervention strategies.

[0111] By inputting multi-timescale multivariate time series and elasticity coefficient values ​​as monitoring features into a pre-trained heart failure rehabilitation trajectory prediction model, the model outputs a health index sequence within the future intervention window. The model learns multi-scale temporal dependencies and can accurately predict the health index sequence of heart failure patients within the future intervention window, providing a reference for early intervention. The model also identifies elasticity loss thresholds in the health index sequence as early intervention points. By identifying elasticity loss thresholds under specific conditions, it can promptly detect the time points when a patient's health status may deteriorate. Combined with current health characteristic values, reports and intervention suggestions are generated, which helps to achieve early prevention.

[0112] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages:

[0113] By constructing a multi-timescale data acquisition system, extracting clinically significant health characteristic values, calculating the elasticity coefficient of the physiological system, and predicting health trajectories based on deep learning models, this approach effectively solves key problems in traditional heart failure monitoring, such as delayed early warning, high false alarm rates, and lack of forward-looking predictive capabilities. It achieves a shift from passive monitoring to proactive prediction. By quantifying the physiological system elasticity index, it enables early identification of heart failure decompensation risk, providing a valuable time window for clinical intervention. Simultaneously, multi-scale time-series analysis combined with predictive models significantly improves the accuracy and forward-looking nature of health status assessment, ultimately achieving the goals of reducing the risk of acute attacks and optimizing rehabilitation management outcomes.

[0114] By collecting patients' physiological data at different time scales using multiple sensors, and constructing a multivariate time series after preprocessing and feature extraction, the elasticity coefficient of the physiological system is quantified. Then, a predictive model is used to output a future health index sequence. Finally, the critical point of elasticity loss is identified and intervention suggestions are generated. Its role is to comprehensively and accurately monitor the health status of heart failure patients, predict health change trends in advance, promptly identify patients' health problems, provide professional intervention guidance, and improve the rehabilitation effect of heart failure patients.

[0115] Furthermore, this embodiment of the invention also provides an intelligent rehabilitation monitoring device for patients with heart failure.

[0116] Figure 2 This is a schematic diagram of the structure of an intelligent rehabilitation monitoring device for heart failure patients according to an embodiment of the present invention.

[0117] like Figure 2 As shown, an intelligent rehabilitation monitoring device for heart failure patients includes: a multi-source sensor, a feature extraction module, and a prediction module;

[0118] Multi-source sensors are used to obtain time-series data of heart failure patients corresponding to each acquisition time scale within a preset time period based on preset multiple acquisition time scales, forming a time-series data set of the monitoring items, including the time-series dataset corresponding to each monitoring item at multiple acquisition time scales;

[0119] The feature extraction module is used to extract health feature values ​​from the time series dataset of each monitoring item, generating feature value sequences corresponding to different collection time scales. The feature value sequence consists of the health feature values ​​of each collection point at the corresponding collection time scale. Based on the feature value sequences of all monitoring items at each collection time scale, a multivariate time series with multiple time scales is constructed, including the multivariate time series sequence matrix corresponding to each time scale.

[0120] The prediction module is based on multi-timescale multivariate time series and uses a preset event elasticity extraction mechanism to quantify the physiological system elasticity coefficient value of heart failure patients. The multivariate time series and elasticity coefficient value are used as monitoring features and input into the pre-trained heart failure rehabilitation trajectory prediction model to output the health index sequence within the future intervention window. The elasticity loss threshold of the health index sequence is identified as an early intervention point and synchronized to the monitoring personnel.

[0121] It should be noted that other specific implementation details of the embodiments of the present invention can refer to the above-described intelligent rehabilitation monitoring method for heart failure patients.

[0122] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An intelligent rehabilitation monitoring device for heart failure patients, characterized in that, The device includes: a multi-source sensor, a feature extraction module, and a prediction module; The multi-source sensor is used to obtain the time-series data of the heart failure patient corresponding to each acquisition time scale within a preset time period based on preset multiple acquisition time scales, forming a time-series data set of the monitoring item, including the time-series dataset corresponding to each monitoring item at multiple acquisition time scales. The feature extraction module is used to extract health feature values ​​from the time series dataset of each monitoring item, generating feature value sequences corresponding to different collection time scales. The feature value sequence is composed of the health feature values ​​of each collection point at the corresponding collection time scale. Based on the feature value sequences of all monitoring items at each collection time scale, a multivariate time series with multiple time scales is constructed, including a multivariate time series sequence matrix corresponding to each time scale. The prediction module is based on multi-timescale multivariate time series and uses a preset event elasticity extraction mechanism to quantify the physiological system elasticity coefficient value of heart failure patients. The multivariate time series and elasticity coefficient value are used as monitoring features and input into the pre-trained heart failure rehabilitation trajectory prediction model to output the health index sequence within the future intervention window. The elasticity loss threshold of the health index sequence is identified as an early intervention point and synchronized to the monitoring personnel. The device employs the following method: S101, based on preset multiple acquisition time scales, obtain the time-series data of the heart failure patient corresponding to each acquisition time scale within a preset time period for each monitoring item, forming a time-series data set for the monitoring item, including the time-series dataset corresponding to each monitoring item at multiple acquisition time scales; the preset multiple acquisition time scales are set to three acquisition time scales, namely high-frequency scales. Mid-frequency scale Low-frequency scale ; S102, extract health feature values ​​from the time series dataset of each monitoring item to generate feature value sequences corresponding to different collection time scales. The feature value sequences are composed of the health feature values ​​of each collection point at the corresponding collection time scale. S103, based on the feature value sequence of all monitoring items at each acquisition time scale, construct a multivariate time series sequence with multiple time scales, including the multivariate time series sequence matrix corresponding to each time scale; S104, Based on a multi-timescale multivariate time series, a preset event elasticity extraction mechanism is used to quantify the physiological system elasticity coefficient value of heart failure patients; the preset event elasticity extraction mechanism specifically includes: S201, based on the feature value sequence corresponding to the mid-frequency scale, identifying disturbance events therein, and extracting the data segment corresponding to each disturbance event in the feature value sequence of the high-frequency scale; the judgment condition for the disturbance event is set as follows: for the feature value sequence of each monitoring item, all continuous intervals where the health feature value exceeds the preset health feature threshold interval are identified as disturbance events; S202, based on the data segment corresponding to each disturbance event in the extracted high-frequency scale feature value sequence, the recovery speed of its health feature value is calculated to quantify the anti-disturbance ability of heart failure patients; S203, based on the recovery speed of the health feature value corresponding to the disturbance event, the physiological system elasticity coefficient value is calculated; S105 uses multi-timescale multivariate time series and elasticity coefficient values ​​as monitoring features to input into the pre-trained heart failure rehabilitation trajectory prediction model, and outputs a health index sequence within the future intervention window; S106 identifies the critical point of elasticity loss in the health index sequence and uses it as an early intervention point, which is then synchronized with monitoring personnel.

2. The intelligent rehabilitation monitoring device for heart failure patients as described in claim 1, characterized in that, The time-series data set of the monitoring item is represented as follows: P represents the total number of monitored items. This is the time-series dataset for the P-th monitoring item, which includes time-series data corresponding to multiple preset acquisition time scales. K represents the number of preset data acquisition time scales. This refers to the time series data corresponding to the Kth collection time scale of this monitoring item.

3. The intelligent rehabilitation monitoring device for heart failure patients as described in claim 2, characterized in that, The time series dataset of each monitoring item in the monitoring item time series data set is divided into three collection time scales: high-frequency physiological parameters collected at a first preset time interval, mid-frequency trend parameters collected at a second preset time interval, and low-frequency cumulative parameters collected at a third preset time interval, wherein the first preset time interval is less than the second preset time interval, and the second preset time interval is less than the third preset time interval.

4. The intelligent rehabilitation monitoring device for heart failure patients as described in claim 3, characterized in that, The extraction of health feature values ​​from the time-series dataset for each monitoring item includes: All time-series data are preprocessed, including data cleaning and normalization. For time-series data of each monitoring item at different collection time scales, health feature values ​​are extracted from the data at each collection point to generate a feature value sequence at each collection time scale.

5. The intelligent rehabilitation monitoring device for heart failure patients as described in claim 4, characterized in that, The multi-timescale multivariate time series includes a multivariate time series sequence matrix corresponding to the high-frequency scale. Multivariate time series sequence matrix corresponding to the mid-frequency scale Multivariate time series sequence matrix corresponding to low-frequency scale .

6. The intelligent rehabilitation monitoring device for heart failure patients as described in claim 5, characterized in that, S103 specifically includes: For each time scale The multivariate time-series sequence matrix is ​​constructed as follows: for the time scale... , build The dimensional sequence matrix serves as the multivariate time series sequence matrix corresponding to this time scale: ； in, Time scale The length of the feature value sequence is given below, where P is the number of monitoring items. For monitoring items At any moment Health characteristic values, ∈[ , ].

7. The intelligent rehabilitation monitoring device for heart failure patients as described in claim 6, characterized in that, The recovery rate of the health characteristic value is calculated according to the following formula: ； in, Let be the recovery rate for the i-th disturbance event. This represents the peak value of the health characteristic value of the disturbance event in the corresponding data segment. These are the stable characteristic values ​​following the disturbance event in the corresponding data segment. and They are respectively and The corresponding moment; The elasticity coefficient of the physiological system is calculated according to the following formula: ; Where E is the elasticity coefficient and M is the total number of disturbance events. Let i be the recovery rate for the i-th disturbance event. This represents the maximum recovery rate for this heart failure patient.

Images