A method, system, device and medium for optimizing performance of an electrocardiosignal classification model

Abstract

The application discloses a kind of electrocardiosignal classification model performance optimization method, system, device and medium, wherein the method comprises the following steps: the electrocardiosignal classification task is divided into multiple subtasks;Different subtasks are constructed and trained expert model;The decision of all expert model outputs is fused using a gating model to obtain an overall solution.The application can effectively deal with large-scale and variable electrocardiogram data by utilizing the expertise of different expert models and the dynamic weight distribution of the gating mechanism.This flexibility and robustness make it a powerful tool for arrhythmia monitoring and diagnosis in clinical and telemedicine settings.The application can be widely applied in electrocardiosignal classification field.

Description

Technical Field

[0001] This invention relates to the field of electrocardiogram (ECG) signal classification, and more particularly to a method, system, device, and medium for optimizing the performance of an ECG signal classification model. Background Technology

[0002] Electrocardiography (ECG) is a crucial tool for detecting heart disease, and early and accurate diagnosis of cardiac arrhythmias (CA) is essential for reducing mortality and healthcare costs. Experienced cardiologists can diagnose arrhythmias by analyzing typical ECG changes, such as P waves, QRS complexes, ST segments, and T waves. However, manually analyzing these subtle changes is not only time-consuming but also susceptible to subjective factors. Therefore, developing automated ECG analysis methods to improve detection speed and repeatability has become a research hotspot in this field. ECG signal analysis based on deep learning technology has significantly improved diagnostic accuracy by effectively combining joint ECG feature extraction and classification through end-to-end learning. However, non-ensemble methods based on a single classifier are insufficient to represent the diverse pathological ECG features that appear in different shapes and morphologies (multi-scale) in various CAs. Summary of the Invention

[0003] In order to at least partially solve one of the technical problems existing in the prior art, the purpose of this invention is to provide a method, system, device and medium for performance optimization of electrocardiogram signal classification model of hybrid expert system.

[0004] The first technical solution adopted in this invention is:

[0005] A method for optimizing the performance of an electrocardiogram (ECG) signal classification model includes the following steps:

[0006] The ECG signal classification task was divided into multiple sub-tasks;

[0007] Build and train expert models for different sub-tasks;

[0008] A gating model is used to fuse the decisions from all expert model outputs to obtain an overall solution.

[0009] Furthermore, the construction and training of expert models for different sub-tasks includes:

[0010] Obtain an electrocardiogram (ECG) database, which includes data on arrhythmias and normal ECG data, and records them with tags;

[0011] Based on the subtask partitioning results, the ECG database is divided into multiple different data subsets according to the labels, and different expert models are trained using different data subsets.

[0012] Furthermore, the subtask-based partitioning results divide the ECG database into multiple different data subsets according to labels, and different expert models are trained using different data subsets, including:

[0013] For the four types of arrhythmias in the ECG database, the following four strategies are used to construct data subsets:

[0014] 1) Divide the dataset into two categories (C42) for any two disease types and generate 6 expert models with strong classification ability for the two types of diseases;

[0015] 2) Divide the dataset (C43) into any three disease classes and generate four expert models with strong classification capabilities for two disease classes;

[0016] 3) Divide the dataset into all disease types and train an expert model (ALLin) using the randomness of neural network weight initialization;

[0017] 4) Based on the three disease expert models, add an emergency expert model (EC) trained with all types of data to obtain a specialized expert model proficient in several disease types and an all-round emergency expert model with slightly less ability.

[0018] Furthermore, the gating model is trained in the following manner:

[0019] The gating model is trained together with the expert model so that the gating model can learn under what circumstances to trust a particular expert model.

[0020] Furthermore, the gating model processes all expert models using a top-k strategy, selects the k best-performing expert models from among all expert models, and combines the outputs of these expert models to form the final prediction or decision.

[0021] Furthermore, the expert model is evaluated for performance using the following metrics: accuracy and F1 score;

[0022] The formula for calculating accuracy is:

[0023]

[0024] The formula for calculating the F1 score is:

[0025]

[0026] FP, TN, TP, and FN represent the number of false positives, true negatives, true positives, and false negatives, respectively.

[0027] Furthermore, the expert model is implemented using ResNet18, DenseNet, VGG16, EfficientNet, ShufuleNetV2, Xresnet101, BNN_net, or Light networks.

[0028] The second technical solution adopted in this invention is:

[0029] A performance optimization system for an electrocardiogram (ECG) signal classification model includes:

[0030] The task division module is used to divide the ECG signal classification task into multiple sub-tasks;

[0031] The expert training module is used to build and train expert models for different subtasks;

[0032] The decision fusion module is used to fuse the decisions output by all expert models using a gating model to obtain an overall solution.

[0033] The third technical solution adopted in this invention is:

[0034] A performance optimization device for an electrocardiogram signal classification model, comprising:

[0035] At least one processor;

[0036] At least one memory for storing at least one program;

[0037] When the at least one program is executed by the at least one processor, the at least one processor implements the method described above.

[0038] The fourth technical solution adopted in this invention is:

[0039] A computer-readable storage medium storing a processor-executable program, which, when executed by a processor, performs the method described above.

[0040] The beneficial effects of this invention are: by utilizing the expertise of different expert models and the dynamic weight allocation of gating mechanisms, this invention can effectively cope with large-scale and variable electrocardiogram data. This flexibility and robustness make it a powerful tool for monitoring and diagnosing arrhythmias in clinical and telemedicine environments. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following description is provided with accompanying drawings of the relevant technical solutions in the embodiments of the present invention or the prior art. It should be understood that the accompanying drawings described below are only for the purpose of clearly illustrating some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a design diagram of the MOE integration framework in an embodiment of the present invention;

[0043] Figure 2 This is an expert GradCam visualization diagram in an embodiment of the present invention;

[0044] Figure 3 This is a flowchart illustrating the steps of a method for optimizing the performance of an electrocardiogram signal classification model in an embodiment of the present invention. Detailed Implementation

[0045] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0046] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., are based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0047] In the description of this invention, "several" means one or more, "multiple" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. If "first" or "second" is used, it is only for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features. Furthermore, "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0048] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0049] Mixture of Experts (MoE) systems are an ensemble learning technique developed in the field of neural networks. Unlike traditional deep learning models, MoE allows for the dynamic activation of parts of the neural network, thereby significantly increasing the number of model parameters without increasing computational cost. In electrocardiogram (ECG) arrhythmia detection, MoE decomposes the prediction task into multiple sub-tasks, each handled by an expert model. A gating model then determines which expert's prediction to trust, thereby optimizing classification performance.

[0050] Based on this, this invention proposes a novel deep convolutional neural network ensemble framework that utilizes MoE technology to further improve the performance of ECG signal classification. The MoE framework of this invention decomposes the complex task of arrhythmia classification into several simpler subtasks, trains an independent learner (expert) for each subtask, and then refers to the suggestions of multiple experts to form the final diagnostic decision. This enhances the model's sensitivity to subtle feature variability in ECG signals, thereby improving the model's classification accuracy.

[0051] like Figure 3 As shown, this embodiment provides a method for optimizing the performance of an electrocardiogram (ECG) signal classification model, including the following steps:

[0052] S1. Divide the ECG signal classification task into multiple sub-tasks;

[0053] S2. Build and train expert models for different sub-tasks;

[0054] S3. Use a gating model to fuse the decisions output by all expert models to obtain an overall solution.

[0055] The method in this embodiment mainly includes the following steps: dividing the ECG signal classification task into subtasks, developing subtask disease experts, and using a gating model to fuse all expert decisions. The method is described in detail below, step by step: the MOE ensemble framework, the definition of subtasks, the expert model, the gating function, model training and testing evaluation, and interpretability analysis.

[0056] (1) MOE Integration Framework

[0057] In actual diagnostic practice, clinicians often seek further opinions from patients with confusing ECG signals before making a final diagnosis. Different cardiologists have different insights into patients with different rhythmic heart diseases, and organizing expert consultations is an important way to diagnose complex cases. Therefore, weighting the suggestions of several experts can produce a reliable final decision. The MOE ensemble framework of this invention decomposes the complex task of arrhythmia classification into several simpler and smaller sub-tasks, then trains separate learners (i.e., expert models, or simply experts) for different sub-tasks, and then refers to the suggestions of several experts to obtain a diagnostic result. In this task, a key issue is how to find the natural segmentation of the task and derive the overall solution from the sub-solutions. This involves determining how to decompose the task into appropriate sub-tasks so that each sub-task can be solved by an expert model, and how to merge them to obtain an overall solution. Figure 1 As shown in the overall design scheme, we will achieve the design goals based on three elements:

[0058] 1) Dividing the task into subtasks; in ensemble-based systems, it is crucial to produce different errors on any given sample. After all, if all expert members provide the same output, there is no benefit to their combination. To ensure the diversity of expert decisions in the system, we adopted multiple strategies to construct different experts. By training on different subsets of training data, each expert gained stronger recognition capabilities on certain disease types, consistent with the physician learning process.

[0059] 2) Developing sub-task experts; the MoE concept is to have multiple experts with complementary strengths to ensure comprehensive coverage of the problem space. To better illustrate the performance improvement of the MOE ensemble framework on single neural networks, each expert model can be a neural network with the same architecture, trained on a subset of data, demonstrating superior performance in specific disease identification.

[0060] 3) Gated models are used to fuse all expert decisions; some expert models may be better at identifying specific types of arrhythmias, while others may be more sensitive to other types. Gated networks improve the overall system performance and accuracy by learning how to best route inputs to the appropriate experts.

[0061] Therefore, the MOE ensemble framework proposed in this invention first divides the ECG records containing arrhythmias and normal ECGs in the ECG database into different subsets according to labels. These subsets are then input into expert models to obtain experts with strong classification abilities for certain disease categories. For example, for atrial fibrillation arrhythmia signals, Expert 1 focuses on learning features between atrial fibrillation and normal rhythms, such as the disappearance of P waves. Expert 2 focuses on learning features between atrial fibrillation and other abnormal rhythms, such as QRS complexes. Expert 3 focuses on learning features between normal rhythms and other abnormal rhythms and therefore does not play a significant role. Finally, through a gating function, we select the features learned by Expert 1 and Expert 2, disregarding the opinion of Expert 3, and comprehensively consider the characteristic morphology of P waves and QRS complexes to obtain the final output.

[0062] (2) Definition of subtask

[0063] For the four types of arrhythmias on the CINC2017 database, the following four strategies were used to construct subsets: 1. Divide the dataset into any two disease types (C42) and generate 6 experts with strong classification ability for the two types of diseases; 2. Divide the dataset into any three disease types (C43) and generate 4 experts with strong classification ability for the two types of diseases; 3. Divide the dataset into all disease types and train experts using the randomness of neural network weight initialization (ALLin); 4. Add emergency experts (EC) trained from all types of data to the three types of disease experts to obtain specialized experts proficient in several disease types and all-round emergency experts with slightly less ability.

[0064] (3) Expert Model

[0065] This invention uses classification models that have recently performed well on ECG as expert models for comprehensive discussion to ensure that the neural network ensemble framework proposed in this invention has the ability to further improve classification performance. Specifically, it includes common convolutional neural network classification models such as ResNet18, DenseNet, VGG16, EfficientNet, ShufuleNetV2, and Xresnet101, or a universally applicable ultra-lightweight binary neural network (BNN_net) or a lightweight densely connected convolutional neural network (Light).

[0066] (4) Gating function (i.e., gating model)

[0067] By training a gating model alongside an expert model, the gating model learns when to trust specific experts, specifically through dynamic weight allocation. The "top-k" strategy is a common approach for selecting the most active expert model for a specific task. The basic idea is to select the k best-performing experts from among all available experts and combine their outputs to form the final prediction or decision. Furthermore, the "top-k" strategy allows the model to activate only a few experts when processing each input, creating an efficient "sparse" execution structure and improving system robustness because the system integrates the opinions of multiple experts rather than relying on the judgment of a single expert. Additionally, this approach increases system flexibility, allowing the system to select different expert combinations based on different tasks or data conditions.

[0068] (5) Model training and testing evaluation

[0069] The PhysioNet / CinC_2017Dataset contains 8,528 single-lead ECG recordings, ranging in duration from 9 seconds to slightly over 60 seconds, sampled at 300Hz. All signals were manually labeled by ECG experts as normal rhythm (H), atrial fibrillation rhythm (AF), other rhythms (O), and noisy recordings (N). We truncated all ECG signals into 30-second segments, padding any segments shorter than 30 seconds with zeros, and downsampled them to 200Hz to extract 6000 data points. Finally, the dataset was split in a 5:1 ratio, resulting in 6823 training samples and 1705 test samples.

[0070] We first tested the performance of individual models (Resnet18, Densenet, VGG16, EfficientNet, shufuleNetV2, Xresnet101, BNN_net, and Light) using the training data as benchmarks for comparison, as shown in Table 1 (Resnet18*, Densenet*, VGG16*, EfficientNet*, shufuleNetV2*, Xresnet101*, BNN_net*, and Light*). Next, we partitioned the training data according to different strategies (C42, C43, ALLin, EC) and fed each partition into the same expert model for learning, resulting in classification models capable of specifically handling different disease types. We then used a top-k strategy to dynamically select all experts. Finally, we validated the performance on the test set and compared it with the single-expert model. The performance and performance changes are shown in Table 1.

[0071] In this patent, we use the following metrics to evaluate model performance: accuracy (Acc) and F1-score (F1), which are defined as follows:

[0072]

[0073]

[0074] FP, TN, TP, and FN represent the number of false positives, true negatives, true positives, and false negatives, respectively.

[0075] As shown in Table 1, the MOE framework proposed in this invention improves the classification performance of baseline models by an average of 2% overall. The MoE framework can further enhance classification performance by focusing on expert models with different features. On ShuffleNetV2, a model with relatively weak classification performance, the MoE framework achieved an accuracy improvement of approximately 13% and an F1 score improvement of 18%, demonstrating that the MoE framework can fully unleash the potential of neural network models with weak classification capabilities. The performance differences of different baseline models on different data distributions mean that not all strategies can guarantee optimal results. Therefore, when using the MoE framework to improve performance, it is necessary to compare different strategies and select the best improvement. In Table 1, the strategies that achieve the best improvement results using the MOE strategy on a single expert model are highlighted in bold. It can be seen that on the cinc2017 database, the C43 strategy, the ALLin strategy, and the EC strategy have the best average phenotypic performance. These three strategies utilize model diversity to achieve the best classification results.

[0076] Observations revealed a significant drop in classification performance with the C42 strategy in the Light model. This is because the C42 strategy is typically considered to maximize model diversity, leading the model to sometimes prioritize creating diverse experts over classification accuracy. Therefore, this implies that an overly diverse (beyond optimal) set of experts may fail to provide reliable diagnostic decisions.

[0077] Table 1 Summary of Experimental Results

[0078]

[0079]

[0080] Note: * indicates that the model is used as a single expert for classification.

[0081] (6) Explainability

[0082] The electrocardiogram (ECG) findings of atrial fibrillation are as follows:

[0083] 1) In atrial fibrillation, the P wave disappears, replaced by F waves with inconsistent frequency, amplitude, and morphology. The isoelectric line disappears. The F wave frequency is between 450 and 600 bpm, and the F wave amplitude varies. Generally, in lead V1, an F wave amplitude greater than 1 mV is called coarse fibrillation, and less than 1 mV is called fine fibrillation. Generally, coarse fibrillation has a lower frequency, while fine fibrillation has a higher frequency. In atrial fibrillation, the amplitude of the F wave varies in different leads on the electrocardiogram, and it may be indistinguishable in some leads.

[0084] 2) During atrial fibrillation, rapid atrial excitations cannot be fully conducted to the ventricles. Most of them are conducted covertly in the atrioventricular node, resulting in a ventricular rate that is significantly slower than the atrial rate and an absolutely irregular ventricular rhythm.

[0085] 3) The morphology and amplitude of the QRS wave may change to some extent during atrial fibrillation. This is mainly due to two reasons: the superposition of the f wave and the QRS wave leads to changes in the amplitude of the QRS wave; and the different degrees of ventricular filling lead to changes in the amplitude of the QRS wave.

[0086] For other rhythms (O), the general category activation mapping pattern shows attention spikes associated with repetitive patterns in the ECG waveform. Other rhythm categories contain an unknown number of other arrhythmias, which may include ventricular tachycardia, ventricular bigeminy, ventricular trigeminy, junctional rhythms, etc. Our observations of the dataset suggest that other rhythm waveforms often contain premature beats, which sometimes have template morphologies different from those of other waveforms.

[0087] Gradcam was used to examine the learning characteristics of each expert, based on the best-performing ResNet18 network and C43 policy expert model.

[0088] See Figure 2 The training data contains 3983, 610, 2002, and 258 samples for H, A, O, and N, respectively. Figure 2 In the middle (a), there is expert one (HAO), an expert system that learns from three types of sample data from HAO. Figure 2 In the middle (b), there is expert 2 (HAN), which learns the expert system of three types of sample data from HAN. For expert 1 (HAO) and expert 2 (HAN), expert 1 (HON) focuses on observing the entire cycle of the P wave, QRS complex, and T wave in normal rhythms. Others include attention spikes related to premature beats and R wave peaks after the absence of P waves. Figure 2 In the middle (c), there is expert three (HON), an expert system that learns from HON's three types of sample data. For Figure 2 For Chinese (d) experts (AON), the main focus is on the areas with large QRS complex amplitude variations, which show the main distinguishing features between abnormal rhythms and atrial fibrillation rhythms. Figure 2In the middle (e), the emergency specialist (HAON) studies all types of samples, including all possible variations, with a more dispersed focus.

[0089] In summary, this invention developed and validated an ECG signal classification model based on a hybrid expert system (MoE). This model decomposes the complex task of arrhythmia classification into several simpler sub-tasks, resulting in multiple training expert strategies. It then uses a gating model to synthesize these expert strategies, significantly improving the performance of ECG signal classification compared to other ECG classification models. By leveraging the expertise of different expert models and the dynamic weight allocation of the gating mechanism, the MoE framework can effectively handle large-scale and variable ECG data. This flexibility and robustness make it a powerful tool for arrhythmia monitoring and diagnosis in clinical and telemedicine environments.

[0090] This embodiment also provides a performance optimization system for an electrocardiogram signal classification model, including:

[0091] The task division module is used to divide the ECG signal classification task into multiple sub-tasks;

[0092] The expert training module is used to build and train expert models for different subtasks;

[0093] The decision fusion module is used to fuse the decisions output by all expert models using a gating model to obtain an overall solution.

[0094] This embodiment of the ECG signal classification model performance optimization system can execute the ECG signal classification model performance optimization method provided in the method embodiment of the present invention, and can execute any combination of implementation steps of the method embodiment, and has the corresponding functions and beneficial effects of the method.

[0095] This embodiment also provides a device for optimizing the performance of an electrocardiogram signal classification model, including:

[0096] At least one processor;

[0097] At least one memory for storing at least one program;

[0098] When the at least one program is executed by the at least one processor, the at least one processor implements Figure 3 The method shown.

[0099] This embodiment of the electrocardiogram (ECG) signal classification model performance optimization device can execute the ECG signal classification model performance optimization method provided in the method embodiment of the present invention, and can execute any combination of implementation steps of the method embodiment, and has the corresponding functions and beneficial effects of the method.

[0100] This application also discloses a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium and execute the computer instructions, causing the computer device to perform... Figure 3 The method shown.

[0101] This embodiment also provides a storage medium storing instructions or programs that can execute the ECG signal classification model performance optimization method provided in the method embodiment of the present invention. When the instructions or programs are run, any combination of implementation steps of the method embodiment can be executed, and the method has the corresponding functions and beneficial effects.

[0102] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this invention are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is altered and sub-operations described as part of a larger operation are executed independently.

[0103] Furthermore, although the invention has been described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the invention. Rather, given the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed herein, the actual implementation of the module will be understood within the scope of conventional skill of an engineer. Therefore, those skilled in the art can implement the invention as set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of the invention, which is determined by the full scope of the appended claims and their equivalents.

[0104] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion 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.) to execute all or part of the steps of the methods described in the various embodiments of this invention. 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.

[0105] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0106] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0107] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0108] In the foregoing description of this specification, references to terms such as "one embodiment," "another embodiment," or "some embodiments" indicate that a specific feature, structure, material, or characteristic described in connection with an embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0109] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0110] The above is a detailed description of the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present invention. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for performance optimization of an electrocardiogram (ECG) signal classification model, characterized in that, Includes the following steps: The ECG signal classification task was divided into multiple sub-tasks; Train expert models for different subtasks; A gating model is used to fuse the decisions output by all expert models to obtain an overall solution; The above describes training expert models for different sub-tasks, including: Obtain an electrocardiogram (ECG) database, which includes data on arrhythmias and normal ECG data, and records them with tags; Based on the subtask partitioning results, the ECG database is divided into multiple different data subsets according to the labels, and different expert models are trained using different data subsets. The subtask-based partitioning results divide the ECG database into multiple different data subsets according to labels, and train different expert models using different data subsets, including: For the four types of arrhythmias in the ECG database, the following four strategies are used to construct data subsets: 1) Divide the dataset into two categories for any two disease types and generate six expert models that have strong classification capabilities for the two disease categories. 2) Divide the dataset into any three disease types and generate four expert models with strong classification capabilities for two types of diseases. 3) Divide the dataset into categories for all disease types and train expert models using the randomness of neural network weight initialization; 4) Based on the expert models trained on datasets divided into any three disease types, add emergency expert models trained on all types of data to obtain a specialized expert model proficient in several disease types and an all-round emergency expert model with slightly less ability.

2. The method for performance optimization of an electrocardiogram signal classification model according to claim 1, characterized in that, The gating model is trained in the following way: The gating model is trained together with the expert model so that the gating model can learn under what circumstances to trust a particular expert model.

3. The method for performance optimization of an electrocardiogram signal classification model according to claim 1, characterized in that, The gated model processes all expert models using a top-k strategy, selects the k best-performing expert models from among them, and combines the outputs of these expert models to form the final prediction or decision.

4. The method for performance optimization of an electrocardiogram signal classification model according to claim 1, characterized in that, The expert model is evaluated for performance using the following metrics: accuracy and F1 score; The formula for calculating accuracy is: The formula for calculating the F1 score is: FP, TN, TP, and FN represent the number of false positives, true negatives, true positives, and false negatives, respectively.

5. A method for optimizing the performance of an electrocardiogram signal classification model according to any one of claims 1-4, characterized in that, The expert model is implemented using ResNet18, DenseNet, VGG16, EfficientNet, ShufuleNetV2, Xresnet101, BNN_net, or Light networks.

6. A performance optimization system for an electrocardiogram signal classification model, used to implement the method described in any one of claims 1-5, characterized in that, include: The task division module is used to divide the ECG signal classification task into multiple sub-tasks; The expert training module is used to build and train expert models for different subtasks; The decision fusion module is used to fuse the decisions output by all expert models using a gating model to obtain an overall solution.

7. A device for optimizing the performance of an electrocardiogram (ECG) signal classification model, characterized in that, include: At least one processor; At least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method of any one of claims 1-5.

8. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to perform the method as described in any one of claims 1-5.

Images