Electrocardiogram analysis method and system based on general representation learning
By employing a general representation learning-based electrocardiogram (ECG) analysis method, and utilizing category alignment and category-level ensemble modules, the generalization ability and accuracy of ECG classification models are improved. This addresses the challenges of patient variability and data complexity in ECG diagnosis, enabling more efficient personalized diagnosis.
Patent Information
- Application Number
- CN202511481239.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-16
- Publication Date
- 2026-01-06
AI Technical Summary
Current electrocardiogram (ECG) diagnosis relies on manual interpretation, which is labor-intensive and susceptible to interference. The variability of artificial intelligence models among different patients leads to a decline in generalization ability. Existing solutions have problems such as data dependence, stability, and privacy risks.
An electrocardiogram (ECG) analysis method based on general representation learning is adopted. The general features of ECG data are extracted through the category alignment module, and multi-class fusion is performed using the category-level ensemble module. The ECG classification model is trained with a large-scale dataset. Category alignment and decision-level ensemble strategies are designed to improve the model's generalization ability and accuracy.
It improves the accuracy and robustness of electrocardiogram classification and diagnosis, solves the analytical challenges brought about by patient variability, and enhances the model's performance in personalized diagnosis.
Smart Images
Figure CN121265083A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of medical technology, and in particular to an electrocardiogram analysis method and system based on general representation learning. Background Technology
[0002] An electrocardiogram (ECG) reflects the electrophysiological activity of the heart and is considered the "gold standard" for diagnosing arrhythmias. With its advantages of being simple to perform, non-invasive, and cost-effective, ECG has become a fundamental clinical examination tool for assessing heart health. Currently, ECG diagnosis heavily relies on manual interpretation by physicians, and the large number of ECG examinations performed in hospitals places a significant workload on doctors. Furthermore, the ECG acquisition process is susceptible to various interferences, further increasing the difficulty of diagnosis. In recent years, artificial intelligence (AI) has shown great potential in improving the efficiency and accuracy of ECG interpretation. Various AI algorithms, especially deep neural networks (DNNs), have been applied to ECG analysis.
[0003] Although artificial intelligence algorithms have been used to assist in electrocardiogram (ECG) analysis, significant differences in ECGs among different patients lead to decreased generalization ability of models in practical applications. Developing a deep neural network model that is both accurate and has strong generalization ability for clinical diagnosis remains a significant challenge. From a technical perspective, several solutions exist to address patient-to-patient variability in ECG classification; these solutions can be categorized into adaptation techniques, generative methods, and meta-learning. The performance of adaptation techniques is highly dependent on the availability of large amounts of labeled data. Furthermore, class imbalance in the training data significantly weakens the effectiveness and generalization ability of adaptation techniques. Generative techniques also face challenges, such as unstable training processes, potential privacy risks associated with generating high-quality personalized ECG signals, and limited generation results when training data lacks diversity and comprehensiveness. Meta-learning methods may encounter scalability issues when applied to larger-scale or more complex clinical scenarios. Summary of the Invention
[0004] The embodiments of this application aim to provide an electrocardiogram (ECG) analysis method and system based on general representation learning, which can improve the accuracy and generalization ability of ECG classification and diagnosis.
[0005] The technical solution of this application is implemented as follows: In a first aspect, embodiments of this application provide an electrocardiogram analysis method based on general representation learning, the method comprising: Obtain the raw electrocardiogram (ECG) data of the target patient; wherein, the target patient is the patient for whom ECG data diagnosis is currently required; The raw electrocardiogram data is preprocessed to obtain electrocardiogram data; The electrocardiogram (ECG) data is classified and diagnosed using a pre-defined ECG classification model to obtain the ECG diagnosis result of the target patient. The ECG classification model includes a category alignment module and a category-level integration module. The category alignment module is used to extract effective general features and multiple categories from the ECG data. The category-level integration module is used to fuse the multiple categories.
[0006] In the above scheme, the step of preprocessing the original electrocardiogram (ECG) data to obtain ECG data includes: The original electrocardiogram (ECG) data is filtered using a bandpass filter to obtain filtered original ECG data; wherein the operating frequency of the bandpass filter is 0.5 to 45 Hz. The filtered raw electrocardiogram data is segmented to obtain electrocardiogram segments of fixed length; The electrocardiogram (ECG) image segments are Z-score normalized to obtain the ECG data.
[0007] In the above scheme, before classifying and diagnosing the electrocardiogram data using a pre-determined electrocardiogram classification model to obtain the electrocardiogram diagnosis result of the target patient, the method further includes: Acquire electrocardiogram (ECG) data from different patients; and establish an ECG dataset based on the ECG data; Determine the label information corresponding to each electrocardiogram data in the electrocardiogram dataset; The initial electrocardiogram (ECG) classification model is trained using the ECG dataset and the label information to determine the ECG classification model.
[0008] In the above scheme, the step of training an initial electrocardiogram (ECG) classification model using the ECG dataset and the label information to determine the ECG classification model includes: The electrocardiogram dataset is preprocessed to obtain a preprocessed electrocardiogram dataset; Using the initial category alignment modulus in the initial electrocardiogram classification model and the label information, a similarity-difference operation is performed on the preprocessed electrocardiogram dataset to obtain multiple branch prediction results. The initial category-level fusion module in the initial electrocardiogram classification model is used to fuse the prediction results of the multiple branches to obtain the classification diagnosis result. Based on the classification and diagnosis results, a loss value is calculated, and based on the loss value, the electrocardiogram classification model is determined.
[0009] In the above scheme, the similarity-difference operation is performed on the preprocessed electrocardiogram dataset using the initial category alignment modulus in the initial electrocardiogram classification model and the label information to obtain multiple branch prediction results, including: The similarity between data of the same category in the preprocessed electrocardiogram dataset is calculated using the initial category alignment model in the initial electrocardiogram classification model and the label information. The differences between different categories of data in the preprocessed electrocardiogram dataset are calculated using the initial category alignment model in the initial electrocardiogram classification model and the label information. Based on the similarity and the difference, the prediction results of the multiple branches are determined.
[0010] In the above scheme, the step of fusing the multiple branch prediction results through the initial category-level ensemble module in the initial electrocardiogram classification model to obtain the classification diagnosis result includes: The initial category-level ensemble module in the initial electrocardiogram classification model is used to perform category-level fusion of the meta-learner matrix and the prediction results of the multiple branches to obtain the calculation result. The calculation result is then subjected to a diagonal value operation to obtain the initial classification and diagnosis result; Based on the preset cross-entropy criterion, the initial classification diagnosis results are fused to obtain the classification diagnosis results.
[0011] In the above scheme, determining the electrocardiogram classification model based on the loss value includes: If the loss value is less than the preset loss value, then the electrocardiogram classification model is determined; If the loss value is greater than or equal to the preset loss value, then the initial ECG classification model is trained again using the ECG dataset and the label information until the updated loss value is less than the preset loss value, or the number of updates reaches a preset number of cycles, then the ECG classification model is determined.
[0012] Secondly, embodiments of this application provide an electrocardiogram (ECG) analysis system based on general representation learning, comprising: an acquisition unit, a preprocessing unit, and a classification and diagnosis unit; wherein, The acquisition unit is used to acquire the original electrocardiogram (ECG) data of the target patient; wherein, the target patient is the patient who currently needs to undergo ECG data diagnosis. The preprocessing unit is used to preprocess the raw electrocardiogram data to obtain electrocardiogram data; The classification and diagnosis unit is used to classify and diagnose the electrocardiogram (ECG) data using a pre-determined ECG classification model to obtain the ECG diagnosis result of the target patient. The ECG classification model includes a category alignment module and a category-level integration module. The category alignment module is used to extract effective general features and multiple categories from the ECG data. The category-level integration module is used to fuse the multiple categories.
[0013] Thirdly, embodiments of this application provide an electrocardiogram (ECG) analysis device based on universal representation learning, the ECG analysis device based on universal representation learning comprising: a processor and a memory; wherein, The memory is used to store computer programs; The processor is configured to call and run the computer program from the memory to perform the method as described in the first aspect.
[0014] This application provides an electrocardiogram (ECG) analysis method and system based on general representation learning. The ECG analysis method based on general representation learning includes: acquiring raw ECG data of a target patient; wherein the target patient is a patient currently requiring ECG data diagnosis; preprocessing the raw ECG data to obtain ECG data; classifying and diagnosing the ECG data using a pre-determined ECG classification model to obtain the ECG diagnosis result for the target patient; wherein the ECG classification model includes a category alignment module and a category-level integration module; the category alignment module is used to extract effective general features and multiple categories from the ECG data; the category-level integration module is used to fuse the multiple categories. Using the above scheme, the ECG classification model includes a category alignment module and a category-level integration module. By employing the category alignment strategy corresponding to the category alignment module, more generalizable feature representations are obtained from large-scale ECG datasets, thereby improving the model's performance in personalized diagnosis. At the decision level, a category-based ensemble strategy corresponding to the category-level ensemble module is adopted to further optimize the classification performance of the multi-branch architecture, thereby improving the classification accuracy and robustness of the model, and thus improving the accuracy of electrocardiogram classification diagnosis. Attached Figure Description
[0015] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application. Obviously, the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.
[0016] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.
[0017] Figure 1 This application provides an optional flowchart illustrating an electrocardiogram analysis method based on general representation learning. Figure 2 This application provides a schematic diagram of the structure of an electrocardiogram analysis system based on general representation learning. Figure 3 This application provides a schematic diagram of the structure of an electrocardiogram analysis device based on general representation learning. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the specific technical solutions of this application will be further described in detail below with reference to the accompanying drawings of the embodiments of this application. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application.
[0019] 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 application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.
[0020] In the following description, references to "some embodiments," "this embodiment," "this application embodiment," and examples, etc., describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same subset or different subset of all possible embodiments and may be combined with each other without conflict.
[0021] If the application documents contain similar descriptions such as "first / second", the following explanation shall be added: In the following description, the terms "first / second / third" are used only to distinguish similar objects and do not represent a specific order of objects. It is understood that "first / second / third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.
[0022] This application provides an electrocardiogram analysis method based on general representation learning. Figure 1 This application provides an optional flowchart illustrating an electrocardiogram analysis method based on general representation learning, which will be combined with... Figure 1 The steps shown are explained.
[0023] S101. Obtain the original electrocardiogram (ECG) data of the target patient; wherein, the target patient is the patient for whom ECG data diagnosis is currently required.
[0024] In some embodiments of this application, the raw electrocardiogram data includes heart rate, P wave, PR interval, QRS complex, ST segment, and T wave data.
[0025] In some embodiments of this application, the electrocardiogram analysis method based on general representation learning is applicable to intelligent electrocardiogram analysis scenarios.
[0026] In some embodiments of this application, the execution entity of the electrocardiogram analysis method based on general representation learning is an electrocardiogram analysis device based on general representation learning.
[0027] S102. Perform data preprocessing on the raw electrocardiogram data to obtain electrocardiogram data.
[0028] In some embodiments of this application, the original electrocardiogram (ECG) data is filtered using a bandpass filter to obtain filtered original ECG data; wherein the operating frequency of the bandpass filter is 0.5 to 45 Hz; the filtered original ECG data is segmented to obtain ECG image segments of fixed length; the ECG image segments are then Z-score normalized to obtain ECG data.
[0029] S103. Using a pre-determined electrocardiogram (ECG) classification model, ECG data is classified and diagnosed to obtain the ECG diagnosis results for the target patient. The ECG classification model includes a category alignment module and a category-level integration module. The category alignment module is used to extract effective general features and multiple categories from the ECG data. The category-level integration module is used to fuse multiple categories.
[0030] Understandably, the ECG classification model includes a class alignment module and a class-level ensemble module. The class alignment module employs a class alignment strategy to obtain more generalizable feature representations from large-scale ECG datasets, thereby improving the model's performance in personalized diagnosis. At the decision layer, the class-level ensemble module uses a class-based ensemble strategy to further optimize the classification performance of the multi-branch architecture, thus improving the model's classification accuracy and robustness, and ultimately enhancing the accuracy of ECG classification and diagnosis.
[0031] In some embodiments of this application, S201-S203 are executed before S103, as follows: S201. Obtain electrocardiogram (ECG) data from different patients; and establish an ECG dataset based on the ECG data.
[0032] S202. Determine the label information corresponding to each electrocardiogram data in the electrocardiogram dataset.
[0033] S203. Using the ECG dataset and label information, train the initial ECG classification model to determine the ECG classification model.
[0034] In some embodiments of this application, the electrocardiogram (ECG) dataset is preprocessed to obtain a preprocessed ECG dataset; similarity and difference operations are performed on the preprocessed ECG dataset using the initial category alignment modulus and label information in the initial ECG classification model to obtain multiple branch prediction results; the multiple branch prediction results are fused using the initial category-level fusion module in the initial ECG classification model to obtain a classification diagnosis result; based on the classification diagnosis result, a loss value is calculated, and based on the loss value, an ECG classification model is determined.
[0035] In some embodiments of this application, the similarity between data of the same category in the preprocessed electrocardiogram dataset is calculated using the initial category alignment modulus and label information in the initial electrocardiogram classification model; the difference between data of different categories in the preprocessed electrocardiogram dataset is calculated using the initial category alignment modulus and label information in the initial electrocardiogram classification model; and multiple branch prediction results are determined based on similarity and difference.
[0036] In some embodiments of this application, the initial category-level fusion module in the initial electrocardiogram classification model performs category-level fusion on the meta-learner matrix and multiple branch prediction results to obtain the calculation result; the calculation result is then subjected to diagonal value operation to obtain the initial classification diagnosis result; and the initial classification diagnosis result is fused based on the preset cross-entropy criterion to obtain the final classification diagnosis result.
[0037] In some embodiments of this application, if the loss value is less than a preset loss value, an electrocardiogram (ECG) classification model is determined; if the loss value is greater than or equal to the preset loss value, the initial ECG classification model is trained again using the ECG dataset and label information until the updated loss value is less than the preset loss value, or the number of updates reaches a preset number of cycles, and then the ECG classification model is determined.
[0038] For example, the category alignment strategy corresponding to the category alignment module in the electrocardiogram classification model is as follows: Distinguishing between different categories and individuals of ECG signals is extremely challenging due to inter-patient variability and noise interference. To enable the model to achieve effective general feature representation, this method proposes a supervised class alignment strategy. Specifically, using label information, (1) the similarity between multiple samples of the same class in each batch of model output is calculated; (2) the difference between samples of different classes is calculated. In any training batch, the loss function of the model is calculated using the following formula. This guides the model training, as shown in the following formula:
[0039] Where i∈I≡{1...2N} is the index of any sample in any batch, and (i) ≡ {p∈A(i): yp=yi} represents the set of indices of all positive samples in the multi-view batch that are different from sample i but belong to the same category. |P(i)| represents the number of samples in this set. fj=Proj(Enc(xj)) represents the high-dimensional representation of the j-th sample in this batch, denoted by "". "" indicates the inner product (dot product), and τ is a constant temperature parameter.
[0040] The designed model includes three parallel branches to enhance its deep representation capabilities. The category-level ensemble strategy corresponding to the category-level ensemble module in the ECG classification model is as follows: To fuse the three predictions at the category level, this method introduces a meta-learner matrix M, multiplies it by the three outputs, and then retains the diagonal value of the result as the final prediction. The specific formula is as follows:
[0041] in, (i∈{1,2,3}) represents the prediction result of the i-th branch. This is an operation to take the diagonal value.
[0042] For the optimization of the entire model, a multi-objective function is used, i.e., L = ,in, The binary cross-entropy loss function for multi-class classification is calculated as follows:
[0043] in, For the predicted results, For the label results, For category The weight, For batch size, This represents the number of categories.
[0044] This application uses two large-scale 12-lead electrocardiogram (ECG) datasets to validate the effectiveness of the algorithm: the Chapman-Shaoxing ECG database (Chapman-Shaoxing DB) and the 2018 China Physiological Signal Challenge database (CPSC2018 DB). Many records in these two datasets contain multiple labels; therefore, the validation task is multi-label classification. First, both datasets are randomly shuffled, and then each dataset is randomly divided into training, validation, and test sets in an 8:1:1 ratio.
[0045] The analysis results are as follows: First, consider whether to use a category-level integration strategy: This application designs a deep learning network architecture with multiple branches, and determines the fusion method of the prediction results of the three branches by setting a meta-learner matrix. Subsequently, comparative experiments were conducted to compare the prediction performance of models using and not using the category-level ensemble strategy. After the same training process, the test performance results are shown in Table 1.
[0046] Table 1 Effects of using category-level integration strategy
[0047] Second, consider whether to use a category alignment strategy: Electrocardiograms (ECGs) vary among patients, leading to decreased generalization ability of models in real-world scenarios. To address this challenge, this paper proposes a novel class alignment strategy and integrates it into a deep learning-based multi-label ECG classification method. Comparative experiments were then conducted to compare the effects of using and not using this alignment strategy during model training. The test results on the two datasets after the same training process are shown in Table 2.
[0048] Table 2. Effects of using category alignment strategy
[0049] Understandably, unlike methods that address inter-patient variability through adaptation, generation, or meta-learning strategies, this application advocates learning more generalized feature representations from large-scale ECG datasets to address the analytical challenges posed by inter-patient data variability. Therefore, to address the inter-patient variability problem in ECG signals, this application proposes a novel deep learning (DL) method that integrates class alignment and decision-level ensemble strategies within a multi-branch architecture. The main innovations include: 1) employing a class alignment strategy to obtain more generalized feature representations from large-scale ECG datasets, thereby improving model performance in personalized diagnosis; 2) designing a class-based ensemble strategy at the decision level to further optimize the classification performance of the multi-branch architecture, thereby improving the model's classification accuracy and robustness.
[0050] Based on the electrocardiogram (ECG) analysis method based on general representation learning described in the above embodiments, this application also provides an ECG analysis system based on general representation learning, such as... Figure 2 As shown, Figure 2 This application provides a schematic diagram of the structure of an electrocardiogram (ECG) analysis system based on general representation learning. The ECG analysis system 2 based on general representation learning includes: an acquisition unit 201, a preprocessing unit 202, and a classification and diagnosis unit 203; wherein, The acquisition unit 201 is used to acquire the original electrocardiogram data of the target patient; wherein, the target patient is the patient who currently needs to undergo electrocardiogram data diagnosis; The preprocessing unit 202 is used to preprocess the original electrocardiogram data to obtain electrocardiogram data. The classification and diagnosis unit 203 is used to classify and diagnose the electrocardiogram (ECG) data using a pre-determined ECG classification model to obtain the ECG diagnosis result of the target patient; wherein, the ECG classification model includes a category alignment module and a category-level integration module; the category alignment module is used to extract effective general features and multiple categories from the ECG data; the category-level integration module is used to fuse the multiple categories.
[0051] In some embodiments of this application, the acquisition unit 201 is further configured to filter the original electrocardiogram (ECG) data using a bandpass filter to obtain filtered original ECG data; wherein the operating frequency of the bandpass filter is 0.5 to 45 Hz; the filtered original ECG data is segmented to obtain ECG image segments of fixed length; and the ECG image segments are Z-score normalized to obtain the ECG data.
[0052] In some embodiments of this application, the acquisition unit 201 is further configured to acquire electrocardiogram data of different patients before classifying and diagnosing the electrocardiogram data using a pre-determined electrocardiogram classification model to obtain the electrocardiogram diagnosis result of the target patient; and to establish an electrocardiogram dataset based on the electrocardiogram data; to determine the label information corresponding to each electrocardiogram data in the electrocardiogram dataset; and to train an initial electrocardiogram classification model using the electrocardiogram dataset and the label information to determine the electrocardiogram classification model.
[0053] In some embodiments of this application, the preprocessing unit 202 is further configured to perform data preprocessing on the electrocardiogram dataset to obtain a preprocessed electrocardiogram dataset; The classification and diagnosis unit 203 is further configured to perform similarity and difference operations on the preprocessed electrocardiogram dataset using the initial category alignment module in the initial electrocardiogram classification model and the label information to obtain multiple branch prediction results; fuse the multiple branch prediction results using the initial category-level integration module in the initial electrocardiogram classification model to obtain a classification and diagnosis result; calculate a loss value based on the classification and diagnosis result, and determine the electrocardiogram classification model based on the loss value.
[0054] In some embodiments of this application, the classification and diagnosis unit 203 is further configured to calculate the similarity between data of the same category in the preprocessed electrocardiogram dataset using the initial category alignment modulus in the initial electrocardiogram classification model and the label information; calculate the difference between data of different categories in the preprocessed electrocardiogram dataset using the initial category alignment modulus in the initial electrocardiogram classification model and the label information; and determine the multiple branch prediction results based on the similarity and the difference.
[0055] In some embodiments of this application, the classification diagnosis unit 203 is further configured to perform category-level fusion of the meta-learner matrix and the multiple branch prediction results through the initial category-level integration module in the initial electrocardiogram classification model to obtain a calculation result; perform diagonal value operation on the calculation result to obtain an initial classification diagnosis result; and fuse the initial classification diagnosis result based on a preset cross-entropy criterion to obtain the classification diagnosis result.
[0056] In some embodiments of this application, the classification and diagnosis unit 203 is further configured to determine the electrocardiogram classification model if the loss value is less than a preset loss value; and if the loss value is greater than or equal to the preset loss value, to continue training the initial electrocardiogram classification model using the electrocardiogram dataset and the label information until the updated loss value is less than the preset loss value, or the number of updates reaches a preset number of cycles, and then determine the electrocardiogram classification model.
[0057] Based on the electrocardiogram (ECG) analysis method based on general representation learning described in the above embodiments, this application also provides an ECG analysis device based on general representation learning, such as... Figure 3 As shown, Figure 3 This is a schematic diagram of the structure of an electrocardiogram (ECG) analysis device based on general representation learning, provided in an embodiment of this application. The ECG analysis device 3 based on general representation learning includes a processor 301 and a memory 302. The memory 302 is used to store computer programs; the processor 301 is used to call and run the computer programs from the memory to execute the ECG analysis method based on general representation learning as described in the above embodiment.
[0058] In the embodiments of this application, the processor 301 described above can be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that for different devices, the electronic device used to implement the above processor function can also be other types, and the embodiments of this application do not specifically limit it.
[0059] This application provides a computer-readable storage medium storing a computer program for implementing the electrocardiogram analysis method based on general representation learning as described in any of the above embodiments when executed by a processor.
[0060] For example, the program instructions corresponding to an electrocardiogram analysis method based on general representation learning in this embodiment can be stored on storage media such as optical discs, hard disks, and USB flash drives. When the program instructions corresponding to an electrocardiogram analysis method based on general representation learning in the storage media are read or executed by an electronic device, the electrocardiogram analysis method based on general representation learning as described in any of the above embodiments can be implemented.
[0061] Furthermore, in the embodiments of this application, the functional modules can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional module.
[0062] If the integrated unit is implemented as a software functional module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the method of this embodiment. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0063] It should be understood that the phrases "one embodiment," "an embodiment," or "some embodiments" mentioned throughout the specification mean that a specific feature, structure, or characteristic related to an embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment," "in one embodiment," or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely for descriptive purposes and do not represent the superiority or inferiority of the embodiments. The descriptions of the various embodiments above tend to emphasize the differences between the various embodiments; their similarities or commonalities can be referred to mutually, and for the sake of brevity, they will not be repeated here.
[0064] The modules described above as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules. They may be located in one place or distributed across multiple network units. Some or all of the modules may be selected to achieve the purpose of this embodiment according to actual needs.
[0065] In addition, each functional module in the various embodiments of this application can be integrated into one processing unit, or each module can be a separate unit, or two or more modules can be integrated into one unit; the integrated modules can be implemented in hardware or in the form of hardware plus software functional units.
[0066] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.
[0067] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.
[0068] The features disclosed in the several product embodiments provided in this application can be arbitrarily combined without conflict to obtain new product embodiments.
[0069] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.
[0070] The above description is merely an embodiment of this application, but the protection scope of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. Therefore, the protection scope of this application should be determined by the protection scope of the claims.
Claims
1. A method for electrocardiogram analysis based on general representation learning, characterized in that, The method comprises: obtaining original electrocardiogram data of a target patient; wherein the target patient is a patient currently in need of electrocardiogram data diagnosis; performing data preprocessing on the original electrocardiogram data to obtain electrocardiogram data; performing classification diagnosis on the electrocardiogram data through a pre-determined electrocardiogram classification model to obtain an electrocardiogram diagnosis result of the target patient; wherein the electrocardiogram classification model comprises a class alignment module and a class-level integration module; the class alignment module is used to extract effective general features and multiple classes of the electrocardiogram data; and the class-level integration module is used to fuse the multiple classes.
2. The method of claim 1, wherein, The data preprocessing on the original electrocardiogram data to obtain electrocardiogram data comprises: filtering the original electrocardiogram data through a band-pass filter to obtain filtered original electrocardiogram data; wherein the working frequency of the band-pass filter is 0.5 to 45 Hz; segmenting the filtered original electrocardiogram data to obtain electrocardiogram image segments of a fixed length; performing Z-score standardization processing on the electrocardiogram image segments to obtain the electrocardiogram data.
3. The method of claim 1, wherein, Before the classification diagnosis on the electrocardiogram data through the pre-determined electrocardiogram classification model to obtain the electrocardiogram diagnosis result of the target patient, the method further comprises: obtaining electrocardiogram data of different patients; and establishing an electrocardiogram data set according to the electrocardiogram data; determining label information corresponding to each electrocardiogram data in the electrocardiogram data set; training an initial electrocardiogram classification model through the electrocardiogram data set and the label information to determine the electrocardiogram classification model.
4. The method of claim 3, wherein, The training of the initial electrocardiogram classification model through the electrocardiogram data set and the label information to determine the electrocardiogram classification model comprises: performing data preprocessing on the electrocardiogram data set to obtain a pre-processed electrocardiogram data set; performing similarity and difference operations on the pre-processed electrocardiogram data set through an initial class alignment module in the initial electrocardiogram classification model and the label information to obtain multiple branch prediction results; fusing the multiple branch prediction results through an initial class-level integration module in the initial electrocardiogram classification model to obtain a classification diagnosis result; calculating a loss value based on the classification diagnosis result, and determining the electrocardiogram classification model based on the loss value.
5. The method of claim 4, wherein, The similarity and difference operations on the pre-processed electrocardiogram data set through the initial class alignment module in the initial electrocardiogram classification model and the label information to obtain the multiple branch prediction results comprise: calculating the similarity between data of the same class in the pre-processed electrocardiogram data set through the initial class alignment module in the initial electrocardiogram classification model and the label information; calculating the difference between data of different classes in the pre-processed electrocardiogram data set through the initial class alignment module in the initial electrocardiogram classification model and the label information; determining the multiple branch prediction results based on the similarity and the difference.
6. The method of claim 4, wherein, The classification diagnosis result is obtained by fusing the multiple branch prediction results through an initial class level integration module in the initial electrocardiogram classification model. The operation result is obtained by fusing the meta-learner matrix and the multiple branch prediction results through an initial class level integration module in the initial electrocardiogram classification model. The initial classification diagnosis result is obtained by performing a diagonal value operation on the operation result. The initial classification diagnosis result is fused based on a preset cross-entropy criterion to obtain the classification diagnosis result.
7. The method of claim 4, wherein, The electrocardiogram classification model is determined based on the loss value. If the loss value is less than a preset loss value, the electrocardiogram classification model is determined. If the loss value is greater than or equal to the preset loss value, The initial electrocardiogram classification model is continuously trained based on the electrocardiogram data set and the label information until the updated loss value is less than the preset loss value or the number of updates reaches a preset period, and then the electrocardiogram classification model is determined.
8. A general representation learning based electrocardiogram analysis system, characterized by, It comprises: An acquisition unit, a preprocessing unit and a classification diagnosis unit; wherein The acquisition unit is configured to acquire original electrocardiogram data of a target patient; wherein the target patient is a patient currently in need of electrocardiogram data diagnosis; The preprocessing unit is configured to perform data preprocessing on the original electrocardiogram data to obtain electrocardiogram data; The classification diagnosis unit is configured to perform classification diagnosis on the electrocardiogram data through a pre-determined electrocardiogram classification model to obtain an electrocardiogram diagnosis result of the target patient; wherein the electrocardiogram classification model comprises a class alignment module and a class level integration module; the class alignment module is configured to extract effective general features and multiple classes of the electrocardiogram data; and the class level integration module is configured to fuse the multiple classes.
9. An electrocardiogram analysis device based on general representation learning, characterized by, It comprises: A processor and a memory, wherein The memory is configured to store a computer program; The processor is configured to call and run the computer program from the memory to execute the method of any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, Executable instructions are stored for causing the processor to execute when the method of any one of claims 1 to 7 is implemented.