Cardiac rehabilitation management system and method based on rehabilitation needs

Abstract

The present application discloses a cardiac rehabilitation management system and method based on rehabilitation needs, which relates to the field of intelligent management. It uses AI-based data analysis and coding technology to perform time series analysis and association coding on the time queue of displacement data, and at the same time performs variational modal decomposition and semantic feature extraction on the heart sound data, so as to automatically generate the recognition result based on the query optimization response representation between the semantic features of each heart sound data and the position multidimensional data association features. In this way, by combining displacement data and heart sound data and capturing the temporal dynamic association between the two, more comprehensive patient status information can be provided, reducing the limitations of a single data source. In addition, the system can dynamically adjust the threshold according to the specific circumstances of different patients, reduce false positives and false negatives caused by static thresholds, and thus improve the accuracy and personalization of detection.

Description

Technical Field

[0001] The present application relates to the field of intelligent management, and more specifically, to a cardiac rehabilitation management system and method based on rehabilitation needs. Background Art

[0002] With the increasing incidence of cardiovascular diseases, cardiac rehabilitation management has become an important medical issue. Heart sound data is one of the key indicators for evaluating heart function. By analyzing heart sound data, important information such as heart valve activity, heart rate and rhythm, and heart murmurs can be detected, so as to detect abnormal heart function early and improve the efficiency of early diagnosis and treatment.

[0003] In this regard, Chinese patent CN116759102B proposes an analysis and management system based on cardiac rehabilitation data, which uses adhesive patches, sound receiving units and inertial sensors to collect heart sound and motion data, and stores them in a sub-database of a specific heart disease category. The data analysis module integrates and analyzes these data, compares them with standard heart sound data, determines abnormalities, and alerts the user through vibration.

[0004] In the above patent, when the newly collected heart sound data arrives, the system first determines which characteristic motion parameter interval the data should belong to based on the current motion parameters (calculated by displacement and displacement speed), and then compares it with the standard heart sound data stored in the interval. If the difference between the new data and the standard data exceeds the preset threshold, the heart sound data is considered to be abnormal. However, this preset threshold may not be suitable for the specific conditions of all patients, because there are significant differences in the heart sound characteristics and motion responses of different patients. Static thresholds may lead to false positives or false negatives, especially when the individual differences of patients are large, and fixed thresholds are difficult to accurately distinguish between normal and abnormal situations. In addition, only by comparing the fit between the heart sound data and the standard data, other potentially important features, such as the temporal correlation between the displacement data and the heart sound data, are ignored. This single-dimensional analysis method cannot fully reflect the changes in the patient's cardiac function under different motion states, which may lead to inaccurate and incomplete abnormality detection. For example, some heart sound abnormalities may only appear in a specific motion stage, and a single heart sound comparison method may miss these key information.

[0005] Therefore, an optimized cardiac rehabilitation management program based on rehabilitation needs is desired. Summary of the invention

[0006] In order to solve the above technical problems, this application is proposed.

[0007] According to one aspect of the present application, a cardiac rehabilitation management system based on rehabilitation needs is provided, which includes: a data acquisition module, which is used to apply an adhesive sheet equipped with a sound receiving unit and a sensor unit to a predetermined sound receiving position of a monitored patient, and respectively collect heart sound data and a time queue of displacement data of the monitored patient through the sound receiving unit and the sensor unit, wherein the displacement data includes a displacement amount and a displacement speed; a storage module, which is used to store the time queue of the heart sound data and the displacement data in a database according to the category label of the heart disease of the monitored patient; an identification module, which is used to retrieve the time queue of the heart sound data and the displacement data from the database, and analyze the time queue of the heart sound data and the displacement data to obtain an identification result, wherein the identification result is used to indicate whether the heart function of the monitored patient is abnormal; the identification module includes:

[0008] A displacement data time series encoding unit, used for performing displacement time series encoding on the time queue of the displacement data to obtain a displacement amount time series associated feature vector and a displacement speed time series associated feature vector;

[0009] A displacement feature association unit, used for performing association coding on the displacement time series association feature vector and the displacement speed time series association feature vector to obtain a position multi-dimensional data association feature vector;

[0010] A heart sound data processing unit, used for performing variational modal decomposition and semantic feature extraction on the heart sound data to obtain a plurality of semantic feature vectors of intrinsic modal functions of the heart sound data;

[0011] The recognition result generating unit is used to perform multi-source data query encoding optimization on the semantic feature vectors of the intrinsic modal functions of the multiple heart sound data and the position multi-dimensional data associated feature vectors to obtain the position query optimized heart sound essential semantic response representation, and obtain the recognition result based on the position query optimized heart sound essential semantic response representation.

[0012] Furthermore, the adhesive patch also includes a vibration unit for sending a warning signal to the monitored patient in response to the identification result that the heart function of the monitored patient is abnormal.

[0013] Furthermore, the displacement data time series encoding unit is used to: pass the time queue of the displacement data through a displacement time series encoder based on a forward LSTM model to obtain the displacement amount time series associated feature vector and the displacement speed time series associated feature vector.

[0014] Furthermore, the heart sound data processing unit is used to:

[0015] performing variational modal decomposition on the heart sound data to obtain a plurality of intrinsic modal functions of the heart sound data;

[0016] The plurality of intrinsic modal functions of heart sound data are passed through an intrinsic modal function semantic feature extractor based on a one-dimensional convolutional neural network model to obtain semantic feature vectors of the plurality of intrinsic modal functions of heart sound data.

[0017] Furthermore, the recognition result generating unit includes:

[0018] A position heart sound multimodal query optimization unit, configured to perform multi-source data query encoding optimization on the plurality of heart sound data intrinsic modal function semantic feature vectors and the position multidimensional data association feature vector to obtain a position query optimized heart sound essential semantic response representation vector as the position query optimized heart sound essential semantic response representation;

[0019] The cardiac function recognition unit is used to pass the position query optimized cardiac sound essential semantic response representation vector through a classifier-based cardiac function identifier to obtain the recognition result.

[0020] Furthermore, the position heart sound multimodal query optimization unit includes:

[0021] A heart sound data essential representation subunit, configured to input the plurality of heart sound data intrinsic modal function semantic feature vectors into an essential feature capture network to obtain a heart sound data essential feature representation vector;

[0022] a heart sound data semantic contribution coefficient calculation subunit, configured to calculate a semantic contribution coefficient between each of the heart sound data intrinsic modal function semantic feature vectors in the plurality of heart sound data intrinsic modal function semantic feature vectors and the heart sound data essential feature representation vector to obtain a plurality of heart sound data semantic contribution coefficients;

[0023] The hierarchical feature modulation weighting subunit is used to perform hierarchical feature modulation weighting on the multiple intrinsic modal function semantic feature vectors of the heart sound data based on the multiple heart sound data semantic contribution coefficients to obtain the position query optimized heart sound essential semantic response representation vector.

[0024] Furthermore, the heart sound data essence representation subunit is used for:

[0025] Calculating a characteristic difference coefficient of each inherent modal function semantic feature vector of heart sound data relative to other inherent modal function semantic feature vectors of heart sound data among the plurality of inherent modal function semantic feature vectors of heart sound data to obtain a plurality of characteristic difference coefficients of heart sound data;

[0026] Normalizing the plurality of heart sound data characteristic difference coefficients to obtain a plurality of normalized heart sound data characteristic difference coefficients;

[0027] The plurality of normalized heart sound data feature difference coefficients are used as a weight sequence, and the position-weighted sum of the plurality of heart sound data intrinsic modal function semantic feature vectors is calculated to obtain the heart sound data essential feature representation vector.

[0028] Furthermore, the hierarchical feature modulation weighting subunit is used to:

[0029] Based on the multiple semantic contribution coefficients of the heart sound data, hierarchically modulate the multiple semantic feature vectors of the inherent modal function of the heart sound data to increase the semantic essential expression difference between each semantic feature vector of the inherent modal function of the heart sound data in the multiple semantic feature vectors of the inherent modal function of the heart sound data to obtain multiple modulated semantic feature vectors of the inherent modal function of the heart sound data;

[0030] Inputting the position multi-dimensional data association feature vector and each of the modulated heart sound data intrinsic modal function semantic feature vectors of the multiple modulated heart sound data intrinsic modal function semantic feature vectors into a cross-domain query encoding attention network to obtain multiple position heart sound cross-domain query attention coefficients;

[0031] Inputting the attention coefficients of the cross-domain query of the heart sounds at the multiple locations into a weight conversion network based on a Sigmoid activation function to obtain the attention weights of the cross-domain query of the heart sounds at the multiple locations;

[0032] The attention weights of the heart sound cross-domain query at the multiple locations are used as a sequence of weights, and the weighted sum of the semantic feature vectors of the intrinsic modal functions of the multiple heart sound data is calculated by position to obtain the representation vector of the position query optimized heart sound essential semantic response.

[0033] According to another aspect of the present application, a cardiac rehabilitation management method based on rehabilitation needs is provided, comprising:

[0034] A patch with a sound receiving unit and a sensor unit mounted thereon is applied to a predetermined sound receiving position of a monitored patient, and the time queue of heart sound data and displacement data of the monitored patient is collected by the sound receiving unit and the sensor unit respectively, wherein the displacement data includes displacement amount and displacement speed;

[0035] According to the heart disease category label of the monitored patient, the time queue of the heart sound data and the displacement data is stored in a database;

[0036] The time queues of the heart sound data and the displacement data are retrieved from the database, and the time queues of the heart sound data and the displacement data are analyzed to obtain a recognition result, wherein the recognition result is used to indicate whether the heart function of the monitored patient is abnormal.

[0037] Compared with the prior art, the present application provides a cardiac rehabilitation management system and method based on rehabilitation needs, which uses AI-based data analysis and coding technology to perform time series analysis and association coding on the time queue of the displacement data, and at the same time performs variational modal decomposition and semantic feature extraction on the heart sound data, so as to automatically generate the recognition result based on the query optimization response representation between the semantic features of each heart sound data and the position multidimensional data association features. In this way, by combining displacement data and heart sound data and capturing the temporal dynamic association between the two, more comprehensive patient status information can be provided, reducing the limitations of a single data source. In addition, the system can dynamically adjust the threshold according to the specific circumstances of different patients, reduce false positives and false negatives caused by static thresholds, and thus improve the accuracy and personalization of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0039] Figure 1 is a block diagram of a cardiac rehabilitation management system based on rehabilitation needs according to an embodiment of the present application;

[0040] Figure 2 is a data flow diagram of a cardiac rehabilitation management system based on rehabilitation needs according to an embodiment of the present application;

[0041] Figure 3 is a block diagram of an identification module in a cardiac rehabilitation management system based on rehabilitation needs according to an embodiment of the present application;

[0042] Figure 4 This is a flow chart of a cardiac rehabilitation management method based on rehabilitation needs according to an embodiment of the present application. DETAILED DESCRIPTION

[0043] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0044] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0045] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0046] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0047] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0048] In the technical solution of the present application, a cardiac rehabilitation management system based on rehabilitation needs is proposed. Figure 1 It is a block diagram of a cardiac rehabilitation management system based on rehabilitation needs according to an embodiment of the present application. Figure 2 FIG. 1 is a data flow diagram of a cardiac rehabilitation management system based on rehabilitation needs according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the cardiac rehabilitation management system 300 based on rehabilitation needs includes: a data acquisition module 310, which is used to stick the adhesive sheet equipped with a sound receiving unit and a sensor unit on the predetermined sound receiving position of the monitored patient, and respectively collect the time queue of the heart sound data and the displacement data of the monitored patient through the sound receiving unit and the sensor unit, wherein the displacement data includes the displacement amount and the displacement speed; a storage module 320, which is used to store the time queue of the heart sound data and the displacement data into a database according to the category label of the heart disease type of the monitored patient; an identification module 330, which is used to retrieve the time queue of the heart sound data and the displacement data from the database, and analyze the time queue of the heart sound data and the displacement data to obtain an identification result, wherein the identification result is used to indicate whether the heart function of the monitored patient is abnormal.

[0049] In particular, the data acquisition module 310 is used to stick an adhesive patch equipped with a sound receiving unit and a sensor unit on a predetermined sound receiving position of a monitored patient, and collect the time queue of the heart sound data and displacement data of the monitored patient through the sound receiving unit and the sensor unit respectively, wherein the displacement data includes displacement amount and displacement speed.

[0050] The core component of the data acquisition module is an adhesive patch, which is designed to be applied to the predetermined sound receiving part of the monitored patient. Usually, this part is the patient's chest, especially the area where the heart is located. There are two main acquisition units installed on the adhesive patch: the sound receiving unit and the sensor unit. The sound receiving unit is used to collect heart sound data. The sound receiving unit can be a highly sensitive microphone or other type of acoustic sensor that can capture sound signals generated by the closure of the heart valve, blood flow, etc. The design of the sound receiving unit should ensure that high-quality heart sound data can be captured, and usually requires noise immunity and high resolution. The sensor unit is used to collect displacement data. The sensor unit is usually one or more human motion sensors, such as an inertial sensor (IMU), which can measure the patient's displacement and displacement speed. The displacement refers to the total distance moved by the patient in a certain period of time, and the displacement speed refers to the average movement speed of the patient in a certain period of time. The design of the sensor unit should ensure the accuracy and stability of the data, and usually requires high precision and low power consumption.

[0051] First, the specific steps of data collection are as follows: Attach the patch to the patient's chest and ensure that the sound receiving unit and the sensor unit are correctly aligned with the heart area. The design of the patch should ensure comfort and stability for long-term wear. In order to ensure the accuracy and reliability of the data, medical staff need to carefully check the sticking position and sticking quality when attaching the patch. The sound receiving unit continuously collects the patient's heart sound data. This data includes information such as heart valve activity, heart rate and rhythm, and heart murmurs. Heart sound data is usually stored in the form of time series for subsequent analysis. The sound receiving unit needs to have high sensitivity and noise resistance to ensure that the collected heart sound data is clear and accurate. The sensor unit continuously collects the patient's displacement data, including displacement and displacement speed. Displacement refers to the total distance moved by the patient in a certain period of time, and displacement speed refers to the average movement speed of the patient in a certain period of time. These data are also stored in the form of time series. The sensor unit needs to have the characteristics of high precision and low power consumption to ensure long-term stable operation.

[0052] In particular, the storage module 320 is used to store the time queue of the heart sound data and the displacement data into a database according to the category label of the heart disease type of the monitored patient. In a specific example, each heart sound data and displacement data needs to be timestamped so that correlation analysis can be performed in the time dimension. The accuracy of the timestamp should be as high as possible to ensure that the timing relationship of the data is accurate. According to the patient's heart disease category label, the data is stored in the corresponding sub-database. Each sub-database specializes in storing the heart sound data and displacement data of patients of a certain heart disease category. The design of the sub-database should have efficient data management and retrieval capabilities to support the storage and access of large-scale data.

[0053] In the cardiac rehabilitation management system based on rehabilitation needs, the storage module is responsible for storing the time series of the collected heart sound data and displacement data in the database according to the category label of the heart disease of the monitored patient. The implementation details of this process are as follows: The data storage module is an important component of the system, responsible for managing and storing a large amount of heart sound data and displacement data. In order to efficiently manage and retrieve this data, the data storage module usually includes the following parts: database system, sub-database and data index. The database system is used to store the time series of heart sound data and displacement data. You can choose a relational database (such as MySQL, PostgreSQL) or a NoSQL database (such as MongoDB, Cassandra). Choosing a suitable database system can ensure efficient storage and retrieval of data.

[0054] The sub-database is the core part of the data storage module. Each sub-database is dedicated to storing the heart sound data and displacement data of patients with a certain type of heart disease. The design of the sub-database should have efficient data management and retrieval capabilities and support the storage and access of large-scale data. Data indexing is designed to improve the efficiency of data retrieval. Indexes are established for each sub-database to quickly retrieve and access stored data. The index design should be efficient and flexible, supporting multiple query methods, such as querying by fields such as timestamp, patient ID, and heart disease category label.

[0055] In the storage module, the specific steps of data storage include data acquisition and preprocessing, data classification, data storage and data indexing. Before data storage, it is first necessary to collect heart sound data and displacement data through the data acquisition module. The collected data needs to be preliminarily preprocessed to ensure the quality and availability of the data. Preprocessing includes filtering and calibration. Filtering is to process the heart sound data to remove environmental noise and other interference signals to ensure the purity of the heart sound signal. Common filtering methods include low-pass filtering, high-pass filtering and band-pass filtering, etc., and the appropriate filtering method can be selected according to the specific noise characteristics. Calibration processing is to process the displacement data to ensure that the sensor readings are accurate. This includes correction of the sensor's zero drift, temperature drift, etc. Calibration processing can be achieved by regularly calibrating the sensor or using a calibration algorithm.

[0056] In the storage module, the data needs to be classified according to the patient's heart disease category label. The heart disease category label is usually determined by the doctor when the patient first visits the doctor and recorded in the patient's electronic medical record. The specific steps of data classification are as follows:

[0057] First, obtain the heart disease category label from the patient's electronic medical record. The label can be in the form of a string, such as "coronary heart disease", "hypertensive heart disease", etc. The label can be obtained through the hospital's electronic medical record system or manually input.

[0058] Then, according to the heart disease category label, the time queues of heart sound data and displacement data are classified and stored in the corresponding sub-databases. Each sub-database specifically stores the data of patients with a certain heart disease category. Classified storage can ensure the organization and orderliness of the data, which is convenient for subsequent data processing and analysis.

[0059] The time queues of the classified heart sound data and displacement data are stored in the corresponding sub-databases. The specific steps are as follows:

[0060] First, each heart sound data and displacement data needs to be timestamped so that correlation analysis can be performed in the time dimension. The accuracy of the timestamp should be as high as possible to ensure the accuracy of the temporal relationship of the data. The timestamp can use the Unix timestamp or ISO 8601 standard format. Secondly, the time queues of the heart sound data and displacement data are formatted into a unified format for subsequent data processing and analysis. The formatted data can be in JSON, CSV, or other commonly used data formats. Formatting can ensure the consistency and readability of the data. Finally, the formatted data is stored in the corresponding sub-database. The data stored in each sub-database includes the time series of heart sound data and the time series of displacement data. Data storage can be achieved through the API or commands provided by the database management system to ensure the security and integrity of the data.

[0061] In order to improve data retrieval efficiency, it is necessary to create an index for each sub-database. The specific steps are as follows:

[0062] First, design a reasonable index based on the characteristics of the data and query requirements. The index can include fields such as heart disease category labels, timestamps, and patient IDs. The design of the index should take into account the frequency and complexity of queries to ensure the efficiency and flexibility of the index.

[0063] Then, create an index in the sub-database to quickly retrieve and access the stored data. The creation of an index can be achieved through the tools or commands provided by the database management system. When creating an index, you should choose an appropriate index type, such as B-tree index, hash index, etc.

[0064] Finally, regularly maintain the index to ensure its efficiency and accuracy. Index maintenance can be achieved by regularly rebuilding or optimizing the index. Maintaining the index can prevent index fragmentation and improve query performance.

[0065] In particular, the recognition module 330 is used to retrieve the time queue of the heart sound data and the displacement data from the database, and analyze the time queue of the heart sound data and the displacement data to obtain a recognition result, and the recognition result is used to indicate whether the heart function of the monitored patient is abnormal. In a specific example of the present application, Figure 3 As shown, the recognition module 330 includes: a displacement data time series encoding unit 331, which is used to perform displacement time series encoding on the time queue of the displacement data to obtain a displacement amount time series associated feature vector and a displacement speed time series associated feature vector; a displacement feature association unit 332, which is used to perform association encoding on the displacement amount time series associated feature vector and the displacement speed time series associated feature vector to obtain a position multi-dimensional data associated feature vector; a heart sound data processing unit 333, which is used to perform variational modal decomposition and semantic feature extraction on the heart sound data to obtain multiple heart sound data intrinsic modal function semantic feature vectors; a recognition result generation unit 334, which is used to perform multi-source data query encoding optimization on the multiple heart sound data intrinsic modal function semantic feature vectors and the position multi-dimensional data associated feature vector to obtain a position query optimized heart sound essential semantic response representation, and obtain the recognition result based on the position query optimized heart sound essential semantic response representation.

[0066] Specifically, the displacement data time series encoding unit 331 is used to perform displacement time series encoding on the time series of the displacement data to obtain a displacement amount time series correlation feature vector and a displacement speed time series correlation feature vector. Considering that the displacement amount and the displacement speed in the time series of the displacement data have a trend of changing over time, and there is a correlation between different time spans, and considering that the forward LSTM model is a special recurrent neural network (RNN), it is particularly good at processing and predicting time series data containing long time interval dependencies. Based on this, in the technical solution of the present application, the time series of the displacement data is passed through a displacement time series encoder based on the forward LSTM model to capture and extract the mutual dependence between different time spans, and obtain a displacement amount time series correlation feature vector and a displacement speed time series correlation feature vector. That is, by using the LSTM model as a displacement time series encoder, long-term dependency features reflecting the changing trends of the displacement amount and the displacement speed and their relationship can be effectively extracted from the time series of the displacement data.

[0067] Specifically, the displacement feature association unit 332 is used to associate and encode the displacement time series association feature vector and the displacement speed time series association feature vector to obtain a position multi-dimensional data association feature vector. Considering that the displacement and displacement speed have an associated influence relationship on the time scale, for example, a sudden fall of a patient will cause the displacement speed to drop sharply, while the displacement may remain unchanged in a short time. Therefore, in order to more accurately describe the patient's behavior pattern and health status, thereby enhancing the expression ability of the entire unique data feature set, in the technical solution of the present application, the displacement time series association feature vector and the displacement speed time series association feature vector are associated and encoded to obtain a position multi-dimensional data association feature vector. In particular, in a specific embodiment of the present application, the displacement time series association feature vector and the displacement speed time series association feature vector can be weighted and calculated by position to fuse the two parameters at each time point, thereby providing a more comprehensive description of the patient's activity pattern.

[0068] Specifically, the heart sound data processing unit 333 is used to perform variational mode decomposition and semantic feature extraction on the heart sound data to obtain multiple semantic feature vectors of the intrinsic mode functions of the heart sound data. That is, in an embodiment of the present application, first, the heart sound data is subjected to variational mode decomposition to obtain multiple intrinsic mode functions of the heart sound data. Considering that the heart sound data contains rich information, such as different frequency components such as heart, breathing and heart rate. Therefore, in order to more finely analyze and process the heart sound data and extract different frequency components and features therein, in the technical solution of the present application, the heart sound data is subjected to variational mode decomposition to obtain multiple intrinsic mode functions of the heart sound data. It should be understood that variational mode decomposition is a signal processing technology used to decompose a complex signal into multiple intrinsic mode functions. Compared with traditional empirical mode decomposition, variational mode decomposition has better stability and robustness, and can more accurately separate the components of the signal in the frequency domain. That is to say, variational mode decomposition can decompose complex heart sound signals into multiple intrinsic mode functions (IMFs), each IMF representing signal components in different frequency ranges, so that high-frequency and low-frequency components can be separated more clearly, which is helpful for analyzing different features in heart sound signals. Then, the multiple intrinsic mode functions of heart sound data are passed through an intrinsic mode function semantic feature extractor based on a one-dimensional convolutional neural network model to obtain the semantic feature vectors of the multiple intrinsic mode functions of heart sound data. Considering that the multiple intrinsic mode functions of heart sound data contain very rich and multi-level semantic information, such as physiological activity information such as heart activity, heart rate and myocardium, this information is of great significance for the evaluation of heart function and abnormality detection. Based on this, in the technical solution of the present application, the multiple intrinsic mode functions of heart sound data are passed through an intrinsic mode function semantic feature extractor based on a one-dimensional convolutional neural network model to capture local semantic features in the intrinsic mode function through a sliding window, and obtain multiple semantic feature vectors of the intrinsic mode function of heart sound data. That is, the one-dimensional convolutional neural network model can extract high-level semantic features from IMFs. These features are not just simple frequency or amplitude information, but deeper pattern and structure information, which is very useful for identifying abnormal patterns and physiological characteristics in heart sound signals.

[0069] Specifically, the recognition result generating unit 334 is used to perform multi-source data query encoding optimization on the multiple intrinsic modal function semantic feature vectors of the heart sound data and the position multi-dimensional data association feature vector to obtain the position query optimized heart sound essential semantic response representation, and obtain the recognition result based on the position query optimized heart sound essential semantic response representation. That is, in the embodiment of the present application, firstly, considering that the multiple intrinsic modal function semantic feature vectors of the heart sound data express the semantic information of the physiological and pathological characteristics of the patient's heart, the position multi-dimensional data association feature vector expresses the temporal characteristics of the patient's activity pattern and behavioral characteristics, and there is a mutual correlation relationship between the two features in the temporal dimension. In order to perform query optimization on the semantic features of the heart sound data based on the position to create a comprehensive, accurate and efficient feature representation, so that the representation can not only express the heart sound data features, but also reflect the position information, so as to more accurately describe the patient's health status and activity pattern, in the technical solution of the present application, the multiple intrinsic modal function semantic features of the heart sound data and the position multi-dimensional data association features are multi-source data query encoding optimization to obtain the position query optimized heart sound essential semantic response representation.

[0070] Specifically, the specific process of performing multi-source data query encoding optimization on the multiple semantic feature vectors of intrinsic modal functions of heart sound data and the position multidimensional data association feature vectors includes: first, inputting the multiple semantic feature vectors of intrinsic modal functions of heart sound data into an essential feature capture network to obtain an essential feature representation vector of heart sound data; that is, inputting the multiple semantic feature vectors of intrinsic modal functions of heart sound data into an essential feature capture network for extracting intrinsic attributes or core structures from data. The essential feature capture network is suitable for learning and extracting complex patterns from the multiple semantic feature vectors of intrinsic modal functions of heart sound data, and compressing them into a fixed-length essential feature representation vector of heart sound data. This vector retains the most important information, reduces noise, and provides a reference anchor point for subsequent processing.

[0071] Among them, the specific steps of inputting the multiple semantic feature vectors of the inherent modal functions of heart sound data into the essential feature capture network include: calculating the characteristic difference coefficient of each semantic feature vector of the inherent modal function of heart sound data relative to other semantic feature vectors of the inherent modal function of heart sound data in the multiple semantic feature vectors of the inherent modal function of heart sound data to obtain multiple characteristic difference coefficients of heart sound data; normalizing the multiple characteristic difference coefficients of heart sound data to obtain multiple normalized characteristic difference coefficients of heart sound data; using the multiple normalized characteristic difference coefficients of heart sound data as a sequence of weights, calculating the position-weighted sum of the multiple semantic feature vectors of the inherent modal function of heart sound data to obtain the essential feature representation vector of the heart sound data.

[0072] In the above embodiment, the multiple intrinsic modal function semantic feature vectors of the heart sound data are input into the intrinsic feature capture network to obtain the intrinsic feature representation vector of the heart sound data, including: inputting the multiple intrinsic modal function semantic feature vectors of the heart sound data into the intrinsic feature capture network to extract the intrinsic features using the following intrinsic feature extraction formula to obtain the intrinsic feature representation vector of the heart sound data; wherein the essential feature extraction formula is:

[0073] ;

[0074] ;

[0075] ;

[0076] in, is the semantic feature vector of the intrinsic modal function of the plurality of heart sound data, , , , and are respectively the first, second, and third semantic feature vectors of the intrinsic modal function of the plurality of heart sound data , and The semantic feature vector of the intrinsic mode function of the heart sound data, represents the norm of the eigenvector, is the number of vectors in the semantic feature vectors of the intrinsic modal functions of the plurality of heart sound data minus one, yes The corresponding heart sound data feature difference coefficient, Indicated by natural constant The exponential function value with base , is the number of vectors in the semantic feature vectors of the intrinsic modal functions of the plurality of heart sound data, It is the essential feature representation vector of heart sound data.

[0077] Next, the semantic contribution coefficient between each of the plurality of semantic feature vectors of intrinsic modal functions of heart sound data and the intrinsic feature representation vector of heart sound data is calculated to obtain a plurality of semantic contribution coefficients of heart sound data; by calculating the semantic contribution coefficient between each semantic feature vector of intrinsic modal functions of heart sound data and the intrinsic feature representation vector of heart sound data, the specific contribution degree of each semantic feature vector of intrinsic modal functions of heart sound data to the overall semantic expression can be determined. This quantification process helps identify which features are core and which may be redundant or irrelevant, thereby providing a basis for selectivity in subsequent steps.

[0078] In the above embodiment, calculating the semantic contribution coefficient between each of the multiple intrinsic modal function semantic feature vectors of heart sound data and the intrinsic feature representation vector of the heart sound data to obtain multiple semantic contribution coefficients of heart sound data includes: calculating the semantic contribution coefficient between each of the multiple intrinsic modal function semantic feature vectors of heart sound data and the intrinsic feature representation vector of the heart sound data using the following semantic contribution calculation formula to obtain multiple semantic contribution coefficients of heart sound data; wherein the semantic contribution calculation formula is:

[0079] ;

[0080] in, is the transpose of a vector, yes and The inverse matrix of the covariance matrix between yes The corresponding semantic contribution coefficient of heart sound data.

[0081] Then, based on the semantic contribution coefficients of the multiple heart sound data, the semantic feature vectors of the intrinsic modal functions of the multiple heart sound data are subjected to hierarchical feature modulation weighting to obtain the position query optimized heart sound essential semantic response representation vector. Here, through hierarchical modulation, the expressiveness of important features can be enhanced while suppressing the influence of minor features, so that the model can better focus on those features that are critical to the final decision, thereby improving robustness and generalization ability.

[0082] In the above embodiment, based on the multiple semantic contribution coefficients of the heart sound data, the multiple semantic feature vectors of the intrinsic modal functions of the heart sound data are subjected to hierarchical feature modulation weighting to obtain the position query optimized heart sound essential semantic response representation vector, including: based on the multiple semantic contribution coefficients of the heart sound data, the multiple semantic feature vectors of the intrinsic modal functions of the heart sound data are subjected to hierarchical feature modulation weighting according to the following hierarchical feature modulation weighting formula to obtain the position query optimized heart sound essential semantic response representation vector; wherein the hierarchical feature modulation weighting formula is:

[0083] ;

[0084] ;

[0085] ;

[0086] ;

[0087] in, is the preset threshold, For Perform mask modulation operation, for The semantic feature vector of the intrinsic modal function of the modulated heart sound data, is vector multiplication, is the position multidimensional data associated feature vector, is the bi-norm of the vector, for The corresponding position heart sound cross-domain query attention coefficient, yes function, for The corresponding position heart sound cross-domain query attention weight, and They are and The corresponding position heart sound cross-domain query attention weight, It is the semantic response representation vector of the location query optimized heart sound essence.

[0088] More specifically, the specific steps of performing hierarchical feature modulation weighting on the multiple intrinsic modal function semantic feature vectors of the heart sound data based on the multiple semantic contribution coefficients of the heart sound data include: performing hierarchical modulation on the multiple intrinsic modal function semantic feature vectors of the heart sound data based on the multiple semantic contribution coefficients of the heart sound data to widen the semantic essential expression difference between the individual intrinsic modal function semantic feature vectors of the multiple intrinsic modal function semantic feature vectors of the heart sound data to obtain multiple modulated intrinsic modal function semantic feature vectors of the heart sound data; associating the position multidimensional data feature vector with the multiple modulated feature vectors; Each modulated post-heart sound data intrinsic modal function semantic feature vector in the post-heart sound data intrinsic modal function semantic feature vector is input into the cross-domain query encoding attention network to obtain multiple position heart sound cross-domain query attention coefficients; the multiple position heart sound cross-domain query attention coefficients are input into the weight conversion network based on the Sigmoid activation function to obtain multiple position heart sound cross-domain query attention weights; using the multiple position heart sound cross-domain query attention weights as a sequence of weights, the position-weighted sum of the multiple heart sound data intrinsic modal function semantic feature vectors is calculated to obtain the position query optimized heart sound essential semantic response representation vector.

[0089] After obtaining the location query optimized heart sound essential semantic response representation, the location query optimized heart sound essential semantic response representation vector is further passed through a classifier-based cardiac function identifier to obtain the recognition result. That is, the location query optimized heart sound essential semantic response representation obtained by query optimization using the semantic features of the inherent modal functions of the multiple heart sound data and the location multidimensional data association features is classified and processed to automatically generate the recognition result. In this way, by combining displacement data and heart sound data and capturing the temporal dynamic association between the two, more comprehensive patient status information can be provided, reducing the limitations of a single data source. In addition, the system can dynamically adjust the threshold according to the specific circumstances of different patients, reduce false positives and false negatives caused by static thresholds, and thus improve the accuracy and personalization of detection.

[0090] Preferably, passing the location query optimized heart sound essential semantic response representation vector through a classifier-based heart function identifier to obtain the recognition result comprises:

[0091] The feature set of the location query optimized heart sound essence semantic response representation vector is subjected to feature clustering to obtain a feature set within the location query optimized heart sound essence semantic response class and a feature set outside the location query optimized heart sound essence semantic response class, namely:

[0092] ;

[0093] in, The first feature set of the location query optimization heart sound essential semantic response class eigenvalues, The first out-of-class feature set representing the location query optimization heart sound essential semantic response eigenvalues, Representation location query optimizes the intrinsic semantic response of heart sounds within the class feature set;

[0094] Calculate the ratio of the number of feature values ​​in the feature set of the location query optimized heart sound essential semantic response class to the number of feature values ​​in the feature set of the location query optimized heart sound essential semantic response representation vector ,Right now:

[0095] ;

[0096] in, represents the number of feature values ​​in the feature set of the location query optimization heart sound essential semantic response class, The number of eigenvalues ​​of the feature set representing the position query optimized heart sound essential semantic response representation vector;

[0097] Calculate the absolute value sum of all feature values ​​in the feature set of the location query optimization heart sound essential semantic response class The sum of the absolute values ​​of all feature values ​​in the feature set of the position query optimization heart sound essence semantic response representation vector The ratio of the powers is used to obtain the modulation weight of the location query optimized heart sound essential semantic response, that is:

[0098] ;

[0099] in, represents the modulation weight of the location query optimization heart sound essential semantic response, A feature set representing a vector representing a semantic response to the location query-optimized heart sound essence;

[0100] Calculate the sum of squares of all feature values ​​in the feature set of the location query optimization heart sound essential semantic response class The sum of the squares of all feature values ​​in the feature set of the vector representing the location query optimization heart sound essence semantic response is The ratio of the powers is used to obtain the location query optimization heart sound essential semantic response and reconciliation weight, that is:

[0101] ;

[0102] in, Indicates that the location query optimizes the intrinsic semantic response and reconciliation weight of the heart sound;

[0103] For each feature value in the feature set within the location query optimized heart sound essential semantic response class, calculate the product of the feature value and the location query optimized heart sound essential semantic response harmonic weight plus the location query optimized heart sound essential semantic response modulation weight to obtain the optimized feature value, that is:

[0104] ;

[0105] in, represents the first feature set within the class of optimizing the essential semantic response of heart sounds based on the position query Optimized eigenvalues;

[0106] For each feature value in the out-of-class feature set of the location query optimized heart sound essential semantic response, the product of the feature value and the location query optimized heart sound essential semantic response tuning weight is calculated to obtain the optimized feature value, that is:

[0107] ;

[0108] in, represents the first out-of-class feature set of the heart sound essential semantic response optimized based on the position query Optimized eigenvalues;

[0109] The optimized location query optimized heart sound essence semantic response representation vector composed of the optimized feature values ​​of the feature set within the location query optimized heart sound essence semantic response class and the optimized feature values ​​of the feature set outside the location query optimized heart sound essence semantic response class is passed through the classifier-based HART functional status detector to obtain the detection result.

[0110] Here, when the multiple intrinsic modal function semantic feature vectors of heart sound data and the position multidimensional data association feature vector respectively represent the one-dimensional local association semantic features of multiple variational modes of heart sound data and the multidimensional association coding features of displacement data, considering the differences in their essential characteristic modulation, it is expected to improve the semantically consistent aggregation expression effect of the position query optimized heart sound essential semantic response representation vector obtained by cross-domain query coding optimization.

[0111] Therefore, while performing feature clustering on the location query optimized essential semantic response representation vector of heart sounds, the interaction description of the key feature information of the location query optimized essential semantic response representation vector in the clustering process is carried out, and the geometric equivariant topology of the feature is constructed by low-rank harmonic modulation of the clustering features of the location query optimized essential semantic response representation vector of heart sounds and the equivariance of the feature as a whole, so as to obtain the graphical distribution translation and rotation symmetry of the clustering features of the location query optimized essential semantic response representation vector of heart sounds relative to the feature as a whole, thereby introducing geometric message passing into the feature expression of the location query optimized essential semantic response representation vector of heart sounds, and realizing the cluster mapping symmetry of the location query optimized essential semantic response representation vector through manipulation of irreducible low-rank coefficients, thereby improving the clustering-based feature representation consistency of the location query optimized essential semantic response representation vector of heart sounds, and thus improving the accuracy of the recognition result obtained by the location query optimized essential semantic response representation vector of heart sounds through a classifier-based cardiac function identifier.

[0112] As described above, the cardiac rehabilitation management system 300 based on rehabilitation needs according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a cardiac rehabilitation management algorithm based on rehabilitation needs. In a possible implementation, the cardiac rehabilitation management system 300 based on rehabilitation needs according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the cardiac rehabilitation management system 300 based on rehabilitation needs can be a software module in the operating system of the wireless terminal, or can be an application developed for the wireless terminal; of course, the cardiac rehabilitation management system 300 based on rehabilitation needs can also be one of the many hardware modules of the wireless terminal.

[0113] Alternatively, in another example, the rehabilitation needs-based cardiac rehabilitation management system 300 and the wireless terminal may also be separate devices, and the rehabilitation needs-based cardiac rehabilitation management system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0114] Furthermore, a cardiac rehabilitation management method based on rehabilitation needs is also provided.

[0115] Figure 4 FIG. 1 is a flow chart of a cardiac rehabilitation management method based on rehabilitation needs according to an embodiment of the present application. Figure 4 As shown, according to the embodiment of the present application, the cardiac rehabilitation management method based on rehabilitation needs includes the following steps: S1, applying an adhesive sheet equipped with a sound receiving unit and a sensor unit to a predetermined sound receiving position of a monitored patient, and respectively collecting the time queue of heart sound data and displacement data of the monitored patient through the sound receiving unit and the sensor unit, wherein the displacement data includes a displacement amount and a displacement speed; S2, storing the time queue of the heart sound data and the displacement data into a database according to the category label of the heart disease type of the monitored patient; S3, retrieving the time queue of the heart sound data and the displacement data from the database, and analyzing the time queue of the heart sound data and the displacement data to obtain a recognition result, wherein the recognition result is used to indicate whether the heart function of the monitored patient is abnormal.

[0116] In summary, the cardiac rehabilitation management method based on rehabilitation needs according to the embodiment of the present application is explained, which uses AI-based data analysis and coding technology to perform time series analysis and association coding on the time queue of the displacement data, and at the same time performs variational modal decomposition and semantic feature extraction on the heart sound data, so as to automatically generate the recognition result based on the query optimization response representation between the semantic features of each heart sound data and the position multidimensional data association features. In this way, by combining the displacement data and the heart sound data and capturing the temporal dynamic association between the two, more comprehensive patient status information can be provided, reducing the limitations of a single data source. In addition, the system can dynamically adjust the threshold according to the specific circumstances of different patients, reduce the false positives and false negatives caused by static thresholds, and thus improve the accuracy and personalization of detection.

[0117] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A cardiac rehabilitation management system based on rehabilitation needs, characterized in that: include: A data acquisition module, used to apply an adhesive sheet equipped with a sound receiving unit and a sensor unit to a predetermined sound receiving position of a monitored patient, and collect heart sound data and a time queue of displacement data of the monitored patient through the sound receiving unit and the sensor unit, respectively, wherein the displacement data includes a displacement amount and a displacement speed; a storage module, used to store the time queue of the heart sound data and the displacement data in a database according to the category label of the heart disease of the monitored patient; an identification module, used to retrieve the time queue of the heart sound data and the displacement data from the database, and analyze the time queue of the heart sound data and the displacement data to obtain an identification result, wherein the identification result is used to indicate whether the heart function of the monitored patient is abnormal; the identification module includes: A displacement data time series encoding unit, used for performing displacement time series encoding on the time queue of the displacement data to obtain a displacement amount time series associated feature vector and a displacement speed time series associated feature vector; A displacement feature association unit, used for performing association coding on the displacement time series association feature vector and the displacement speed time series association feature vector to obtain a position multi-dimensional data association feature vector; A heart sound data processing unit, used for performing variational modal decomposition and semantic feature extraction on the heart sound data to obtain a plurality of semantic feature vectors of intrinsic modal functions of the heart sound data; a recognition result generating unit, configured to perform multi-source data query encoding optimization on the plurality of intrinsic modal function semantic feature vectors of the heart sound data and the position multi-dimensional data associated feature vector to obtain a position query optimized heart sound essential semantic response representation, and obtain the recognition result based on the position query optimized heart sound essential semantic response representation; Wherein, the recognition result generating unit includes: A position heart sound multimodal query optimization unit, configured to perform multi-source data query encoding optimization on the plurality of heart sound data intrinsic modal function semantic feature vectors and the position multidimensional data association feature vector to obtain a position query optimized heart sound essential semantic response representation vector as the position query optimized heart sound essential semantic response representation; A cardiac function recognition unit, configured to pass the location query optimized cardiac sound essential semantic response representation vector through a classifier-based cardiac function recognizer to obtain the recognition result; Wherein, the position heart sound multimodal query optimization unit includes: A heart sound data essential representation subunit, configured to input the plurality of heart sound data intrinsic modal function semantic feature vectors into an essential feature capture network to obtain a heart sound data essential feature representation vector; a heart sound data semantic contribution coefficient calculation subunit, configured to calculate a semantic contribution coefficient between each of the heart sound data intrinsic modal function semantic feature vectors in the plurality of heart sound data intrinsic modal function semantic feature vectors and the heart sound data essential feature representation vector to obtain a plurality of heart sound data semantic contribution coefficients; The hierarchical feature modulation weighting subunit is used to perform hierarchical feature modulation weighting on the multiple intrinsic modal function semantic feature vectors of the heart sound data based on the multiple heart sound data semantic contribution coefficients to obtain the position query optimized heart sound essential semantic response representation vector.

2. The cardiac rehabilitation management system based on rehabilitation needs according to claim 1, characterized in that: The adhesive patch further comprises a vibration unit for sending an early warning signal to the monitored patient in response to the recognition result that the heart function of the monitored patient is abnormal.

3. The cardiac rehabilitation management system based on rehabilitation needs according to claim 2, characterized in that: The displacement data time series encoding unit is used to: pass the time queue of the displacement data through a displacement time series encoder based on a forward LSTM model to obtain the displacement amount time series associated feature vector and the displacement speed time series associated feature vector.

4. The cardiac rehabilitation management system based on rehabilitation needs according to claim 3, characterized in that: The heart sound data processing unit is used for: performing variational modal decomposition on the heart sound data to obtain a plurality of intrinsic modal functions of the heart sound data; The plurality of intrinsic modal functions of heart sound data are passed through an intrinsic modal function semantic feature extractor based on a one-dimensional convolutional neural network model to obtain semantic feature vectors of the plurality of intrinsic modal functions of heart sound data.

5. The cardiac rehabilitation management system based on rehabilitation needs according to claim 4, characterized in that: The heart sound data intrinsic representation subunit is used for: Calculating a characteristic difference coefficient of each inherent modal function semantic feature vector of heart sound data relative to other inherent modal function semantic feature vectors of heart sound data among the plurality of inherent modal function semantic feature vectors of heart sound data to obtain a plurality of characteristic difference coefficients of heart sound data; Normalizing the plurality of heart sound data characteristic difference coefficients to obtain a plurality of normalized heart sound data characteristic difference coefficients; The plurality of normalized heart sound data feature difference coefficients are used as a weight sequence, and the position-weighted sum of the plurality of heart sound data intrinsic modal function semantic feature vectors is calculated to obtain the heart sound data essential feature representation vector.

6. The cardiac rehabilitation management system based on rehabilitation needs according to claim 5, characterized in that: The hierarchical feature modulation weighting subunit is used to: Based on the multiple semantic contribution coefficients of the heart sound data, hierarchically modulate the multiple semantic feature vectors of the inherent modal function of the heart sound data to increase the semantic essential expression difference between each semantic feature vector of the inherent modal function of the heart sound data in the multiple semantic feature vectors of the inherent modal function of the heart sound data to obtain multiple modulated semantic feature vectors of the inherent modal function of the heart sound data; Inputting the position multi-dimensional data association feature vector and each of the modulated heart sound data intrinsic modal function semantic feature vectors of the multiple modulated heart sound data intrinsic modal function semantic feature vectors into a cross-domain query encoding attention network to obtain multiple position heart sound cross-domain query attention coefficients; Inputting the attention coefficients of the cross-domain query of the heart sounds at multiple locations into a weight conversion network based on a Sigmoid activation function to obtain the attention weights of the cross-domain query of the heart sounds at multiple locations; The attention weights of the heart sound cross-domain query at the multiple locations are used as a sequence of weights, and the weighted sum of the semantic feature vectors of the intrinsic modal functions of the multiple heart sound data is calculated by position to obtain the representation vector of the position query optimized heart sound essential semantic response.

7. A cardiac rehabilitation management method based on rehabilitation needs, using the cardiac rehabilitation management system based on rehabilitation needs according to claim 1, characterized in that: include: A patch with a sound receiving unit and a sensor unit mounted thereon is applied to a predetermined sound receiving position of a monitored patient, and the time queue of heart sound data and displacement data of the monitored patient is collected by the sound receiving unit and the sensor unit respectively, wherein the displacement data includes displacement amount and displacement speed; According to the heart disease category label of the monitored patient, the time queue of the heart sound data and the displacement data is stored in a database; The time queues of the heart sound data and the displacement data are retrieved from the database, and the time queues of the heart sound data and the displacement data are analyzed to obtain a recognition result, wherein the recognition result is used to indicate whether the heart function of the monitored patient is abnormal.

Images