An electrocardiogram sub-waveform identification method, system, device and storage medium

Abstract

The present application relates to the technical field of region identification of electrocardiogram subwaveform, and provides an electrocardiogram subwaveform identification method, system, device and storage medium. The method comprises the following steps: based on acquired electrocardiogram data, an identification model is used to obtain an identification result; the process of using the identification model is as follows: the identification model uses an improved U^2-Net model, which comprises an encoder, a classification module and a decoder; based on the electrocardiogram data, an improved encoder is used to obtain a first feature map; based on the first feature map, the classification module and the decoder are used respectively to obtain a second feature map and a third feature map; based on a fusion feature map of the second feature map and the third feature map, the identification result is obtained; wherein the improved encoder is as follows: the output end of each layer of the encoder except the last layer is connected with an SPD-Conv module, and the output of the SPD-Conv module of the previous stage is taken as the input of the current stage of the encoder layer.

Description

Technical Field

[0001] This invention relates to the field of region recognition technology for electrocardiogram (ECG) sub-waveforms, and more particularly to a method, system, device, and storage medium for recognizing ECG sub-waveforms. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In automated electrocardiogram (ECG) analysis, the main types and morphologies of sub-waveforms that need to be detected include: P wave, QRS complex, T wave, and U wave. Accurate detection of these sub-waveforms provides crucial parameters for disease diagnosis, such as the frequency and interval of the QRS complex.

[0004] The morphology of sub-waveforms varies greatly depending on the patient's condition and is highly susceptible to external errors. The ratio of points between sub-waveform regions and background points, as well as the ratio of points between sub-waveform categories, are both relatively large, indicating a severe data imbalance. For example, the U-wave occurs very infrequently in the data, far less frequently than the other three sub-waveforms. These issues significantly increase the difficulty of identifying sub-waveform categories.

[0005] There is some research on detecting various sub-waveforms in electrocardiogram (ECG) waveforms, which can be categorized as waveform instance segmentation. In instance segmentation, there are generally two approaches. One is to first infer multiple candidate boxes and categories of instances using a detection model, and then use post-processing techniques such as non-maximum suppression to select the best candidate box, thus segmenting the specific region of the sub-waveform. The other is to use a "one-stage" approach, directly identifying the specific shape and category of the instance through an instance segmentation model.

[0006] Regardless of the method used, in the current field of recognition, good segmentation results largely depend on pre-trained models derived from massive amounts of data; the sub-model recognition of electrocardiograms is a relatively new topic, and how to develop a highly accurate solution without a pre-trained model is a significant challenge.

[0007] Electrocardiogram (ECG) data is a time-series signal. How to extract its time-series information to better identify these sub-waveforms is a key area of ​​research. In the field of instance segmentation, segmentation models will inevitably infer some non-existent small objects, which manifests in this field as the identification of some non-existent small P waves. How to solve this problem is also a challenge.

[0008] In the identification of sub-waveforms, how to establish a good evaluation benchmark for the model's recognition performance is a crucial issue that directly impacts its clinical effectiveness. However, a theoretical solution for this aspect has been lacking.

[0009] In addition, the data volume is relatively large in the sub-waveform recognition of patients' electrocardiograms, especially in dynamic electrocardiograms, which are often on the order of 24 hours and 200Hz. How to quickly improve the inference speed of the model is also an urgent problem to be solved. Summary of the Invention

[0010] To address the technical problems mentioned above, this invention provides a method, system, device, and storage medium for recognizing electrocardiogram (ECG) sub-waveforms. Based on the U^2-Net framework, this invention improves the model by incorporating the characteristics of ECG data, thereby enhancing both the accuracy and speed of waveform recognition.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] The first aspect of the present invention provides a method for identifying electrocardiogram (ECG) sub-waveforms.

[0013] A method for identifying electrocardiogram (ECG) sub-waveforms, comprising:

[0014] Based on the acquired electrocardiogram data, a recognition model is used to obtain the recognition results;

[0015] The process of adopting the recognition model is as follows: the recognition model adopts an improved U^2-Net model, including an encoder, a classification module, and a decoder; based on electrocardiogram data, the improved encoder is used to obtain a first feature map; based on the first feature map, the classification module and the decoder are used respectively to obtain a second feature map and a third feature map; based on the fusion feature map of the second feature map and the third feature map, the recognition result is obtained.

[0016] The improved encoder is as follows: except for the last layer, the output of each encoder layer is connected to an SPD-Conv module, and the output of the previous SPD-Conv module is used as the input of the current encoder layer.

[0017] Furthermore, the improved encoder is as follows: the encoder in the first part adopts the RSU module, and the encoder in the second part adopts the RSU_3F module, where F indicates that the dilated convolution operation of RSU is used instead of the downsampling operation inside the RSU module; and before adopting the RSU module, the SCSE module is used to extract the sequence features of the electrocardiogram data, and the extracted feature sequence is input into the RSU module for encoding.

[0018] Furthermore, the structure of the decoder in each layer corresponds to the structure of the encoder in the same level.

[0019] Furthermore, the input to each encoder layer is: the upsampled feature map output by the previous level decoder and the feature map output by the encoder of the same level corresponding to the current level decoder.

[0020] Furthermore, the classification module includes a pooling layer, a BatchNorm1d module, a multilayer perceptron, a sigmoid module, and a threshold module connected in sequence.

[0021] Furthermore, the second feature map obtained by the classification module is the feature map output by the sigmoid module;

[0022] Furthermore, the second feature map obtained by the classification module is the feature map output by the threshold module.

[0023] Furthermore, the training process of the recognition model includes: using the FocalLoss loss function based on the second feature map output by the classification module; using the FocalLoss loss function and the DiceLoss loss function based on the third feature map output by the decoder; and optimizing the hyperparameters of the recognition model based on the loss results.

[0024] A second aspect of the present invention provides a system for recognizing electrocardiogram sub-waveforms.

[0025] A system for recognizing electrocardiogram (ECG) sub-waveforms, comprising:

[0026] The recognition module is configured to: obtain recognition results based on the acquired electrocardiogram data using a recognition model;

[0027] The recognition model module is configured as follows: the recognition model adopts an improved U^2-Net model, including an encoder, a classification module, and a decoder; based on electrocardiogram data, the improved encoder is used to obtain a first feature map; based on the first feature map, the classification module and the decoder are used respectively to obtain a second feature map and a third feature map; based on the fused feature map of the second feature map and the third feature map, the recognition result is obtained.

[0028] The improved encoder is as follows: except for the last layer, the output of each encoder layer is connected to an SPD-Conv module, and the output of the previous SPD-Conv module is used as the input of the current encoder layer.

[0029] A third aspect of the present invention provides a computer-readable storage medium.

[0030] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the electrocardiogram sub-waveform recognition method as described in the first aspect above.

[0031] A fourth aspect of the present invention provides a computer device.

[0032] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the steps of the electrocardiogram sub-waveform recognition method as described in the first aspect above.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] This invention addresses the waveform classification problem of electrocardiograms. Based on the U^2-Net framework, it adaptively improves the encoder structure of the recognition model by taking into account the characteristics of electrocardiogram data, and introduces a classification module, which not only improves the accuracy of the recognition results, but also ensures the inference speed of the model.

[0035] To address the issues of multiple categories and imbalanced category data in waveform region classification, this invention selects FocalLoss and DiceLoss as loss functions, which improves the accuracy of the recognition model and brings high efficiency and convenience to the training of the recognition model.

[0036] This invention proposes a reasonable evaluation mechanism for the electrocardiogram sub-waveform recognition task, so as to better meet the needs of clinical analysis. Attached Figure Description

[0037] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0038] Figure 1 This is a structural diagram of the improved recognition model shown in this invention;

[0039] Figure 2 This is a structural diagram of the SPD-Conv module shown in this invention;

[0040] Figure 3 This is a diagram illustrating the internal structure of the Classifier Head as shown in this invention;

[0041] Figure 4 This is a comparison chart of the evaluation criteria, in waveform units, between the old scheme and the scheme of this invention. Detailed Implementation

[0042] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0043] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0044] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0045] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and systems according to various embodiments of this disclosure. It should be noted that each block in a flowchart or block diagram may represent a module, segment, or portion of code, which may include one or more executable instructions for implementing the logical functions specified in the various embodiments. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than that shown in the drawings. For example, two consecutively represented blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, may be implemented using a dedicated hardware-based system that performs the specified functions or operations, or using a combination of dedicated hardware and computer instructions.

[0046] Example 1

[0047] This embodiment provides a method for identifying electrocardiogram (ECG) sub-waveforms. This embodiment uses the application of this method to a server as an example for illustration. It is understood that this method can also be applied to a terminal, or to a system including a terminal, server, and system, and is implemented through interaction between the terminal and server. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network servers, cloud communication, middleware services, domain name services, CDN security services, and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein. In this embodiment, the method includes the following steps:

[0048] Based on the acquired electrocardiogram data, a recognition model is used to obtain the recognition results;

[0049] The process of adopting the recognition model is as follows: the recognition model adopts an improved U^2-Net model, including an encoder, a classification module, and a decoder; based on electrocardiogram data, the improved encoder is used to obtain a first feature map; based on the first feature map, the classification module and the decoder are used respectively to obtain a second feature map and a third feature map; based on the fusion feature map of the second feature map and the third feature map, the recognition result is obtained.

[0050] The improved encoder is as follows: In order to alleviate the problem of misidentification of small waveforms and retain as much detail information as possible, the downsampling module in the encoder is replaced with an SPD-Conv model. That is, except for the last layer, the output of each encoder layer is connected to the SPD-Conv module, and the output of the previous SPD-Conv module is used as the input of the current encoder layer.

[0051] To address the imbalanced data sample problem among the four identified waveforms, this invention employs a multi-label classification strategy, using both the binary Focal Loss and Dice Loss loss functions for model training.

[0052] This invention proposes a one-stage instance segmentation scheme for model selection. This scheme draws on the U^2-Net framework, leveraging the advantages of the embedded (Residual U-block, RSU) module structure to retrain a high-accuracy sub-waveform segmentation model without a pre-trained model. To address the imbalance in the number of points between sub-waveform categories, a classification head strategy is introduced, and the information extracted by the classification head module is fused into the segmentation head module for further feature filtering and enhancement. To mitigate the high-resolution feature loss caused by the original downsampling operation, the SPD-Conv module is used. For sequence information extraction, the SCSE attention mechanism module is employed to enhance meaningful features and suppress useless features.

[0053] To evaluate the model's performance, this invention uses sub-waveforms as the identification unit and establishes a soft-F1 evaluation benchmark.

[0054] Ultimately, the model not only achieved a higher accuracy rate than traditional mainstream models, but also significantly improved inference speed.

[0055] The technical solution of this embodiment will be described in detail below with reference to the accompanying drawings:

[0056] like Figure 1 As shown, the ECG sub-waveform recognition method proposed in this embodiment includes the following steps:

[0057] Step (1): Collect electrocardiogram (ECG) data. Based on the actual situation, the data used here is 10-20 seconds long, 200Hz, 12-lead ECG waveform data.

[0058] Step (2): Perform baseline removal and other data preprocessing on the data to obtain two-dimensional waveform data. The dimension of each data point is represented as [C, L]. Where C represents the number of ECG leads collected; L represents the data truncation length.

[0059] Step (3): Input the data into the recognition model in batches for training. The model architecture process is as follows: Figure 1 As shown, the dimensions of batch data are represented as [B, C, L], and the output data dimension of the model is the same as the input data dimension. Here, B represents the batch size.

[0060] Step (3-1): The recognition model includes an encoder, a classifier, and a decoder. The recognition model is based on U^2-Net and optimized to take into account the characteristics of ECG data, such as... Figure 1 As shown.

[0061] In the encoder, Residual U-shaped modules, i.e., RSU modules, are used in En_1, En_2, and En_3, namely RSU_5, RSU_4, and RSU_3, respectively. Here, "5", "4", and "3" refer to the height L of the RSU module. In the encoder, this invention uses a downsampling factor of 4 within and between the RSU modules to effectively accommodate the extremely long number of ECG data points. Since the data resolution is relatively low in En_4 and En_5 modules, further downsampling within them would lead to the loss of useful semantic information. Therefore, the RSU_3F module is used here, where F indicates that a dilated convolutional version of RSU is used instead of the traditional downsampling operation within the RSU.

[0062] In the decoder, the corresponding stage and the corresponding stage in the encoder have similar RSU module settings. For example, De_4 uses an RSU_3F module similar to that in En_4 and En_5.

[0063] Step (3-2): In order to enhance the feature extraction of the data, the SPD-Conv module is integrated on the basis of U^2-Net.

[0064] As is well known, high-resolution features play a crucial role in segmentation tasks. However, directly performing a 4x downsampling operation between RSU modules in the encoder can potentially lead to a loss of high-resolution information. Therefore, the SPD-Conv module is used here to replace the traditional 4x downsampling operation. For example... Figure 2 As shown, firstly, the input data dimension of this module is set to [B, C', L'], where C' represents the number of channels and L' represents the length of the data. Sampling is performed every 3 positions along the length direction. By continuously changing the first position of the sampling, four data samples are obtained, with a dimension of [B, C', L' / 4]. These four data samples are then concatenated along the channel dimension C to obtain one data sample with a dimension of [B, 4C', L' / 4]. Then, a CNN module with a single convolutional kernel maps the number of channels. Finally, the output data dimension of this module is [B, C', L'], achieving a 4x downsampling effect.

[0065] Step (3-3): In order to incorporate sequence features into the model, this invention adds the SCSE module.

[0066] like Figure 1 As shown, each decoder level takes as input the concatenation of the upsampled feature map from its previous level and the feature map from its symmetric encoder level. To enhance the extraction of sequence features from the data, this input undergoes SCSE feature enhancement before entering the RSU module.

[0067] Step (3-4): In order to better solve the problem of identifying false small P-waves, the segmentation head module is optimized here.

[0068] like Figure 2 As shown, the Classifier Head is the classification head module, and its internal structure is as follows: Figure 3 As shown, the output feature information of the En_5 module is used as the input of the classification head. This feature first undergoes an adaptive global max pooling process, then passes through the BatchNorm1d module, Multilayer Perceptron (MLP), sigmoid module, and thresholding module to obtain the output feature. Next, the output feature of the classification head and the output feature of the segmentation head are multiplied. Regarding the internal structure of the classification head module, there are two schemes:

[0069] Option 1: Multiply the features processed by the thresholding module after the sigmoid module by the output features of the segmentation head;

[0070] Option 2: Multiply the features after sigmoid by the output features of the segmentation head;

[0071] The advantage of Option 1 is that the model can be trained directly, and some false small objects identified by the segmentation head can be filtered out by the segmentation head features after thresholding. However, since the segmentation head features output in this way will bring some difficulties to the convergence of the model, the best inference results are often not obtained.

[0072] The advantage of Option 2 is that the model has better convergence and can achieve a higher inference score, but it cannot solve the problem of the model inferring false small P-waves.

[0073] Through experimentation, a combination of Scheme 1 and Scheme 2 was adopted. First, Scheme 2 was used for model training, resulting in good model recognition results. Then, the encoder, bottleneck module, and classification head module were frozen, and Scheme 1 was used to train the decoder features, ultimately yielding relatively good recognition results.

[0074] Steps (3-5): The model has two output vectors. The segmentation head's output vector has dimensions [B, C, L], where B is the batch size; C is the number of categories, set to 4 here; and L is the original length of the data, responsible for the main task of sub-waveform segmentation and recognition. The classification head's output vector has dimensions [B, C], where B is the batch size; and C is the number of categories, set to 4 here, responsible for the auxiliary task of determining whether there is a sub-waveform of that category in the data.

[0075] Step (4): Feed the model's output vector into the loss function. For the output vector of the classification head, the loss function used is FocalLoss; for the output vector of the segmentation head, the loss functions used are FocalLoss and DiceLoss. The default weight ratio is 1:1:1.

[0076] Step (5): Evaluation mechanism.

[0077] The first is to use data points as the evaluation unit to judge the accuracy of the model's prediction of point categories, i.e., the F1 score of point classification.

[0078] The second approach is to use sub-waveforms as the evaluation unit, develop a soft-F1 scheme, and determine the accuracy of the model's prediction of sub-waveforms, i.e., the F1 / IOU value of the sub-waveforms. For example... Figure 4 As shown in the figure, the numbers represent the number of points in the length dimension of the ECG data, and the blue area represents the positive sample points identified for a certain type of sub-waveform. There are old and new schemes; since the new scheme considers the number of predicted sub-waveforms, it is used here.

[0079] Old plan:

[0080]

[0081] New proposal:

[0082]

[0083]

[0084]

[0085] Example 2

[0086] This embodiment provides a system for recognizing electrocardiogram (ECG) sub-waveforms.

[0087] A system for recognizing electrocardiogram (ECG) sub-waveforms, comprising:

[0088] The recognition module is configured to: obtain recognition results based on the acquired electrocardiogram data using a recognition model;

[0089] The recognition model module is configured as follows: the recognition model adopts an improved U^2-Net model, including an encoder, a classification module, and a decoder; based on electrocardiogram data, the improved encoder is used to obtain a first feature map; based on the first feature map, the classification module and the decoder are used respectively to obtain a second feature map and a third feature map; based on the fused feature map of the second feature map and the third feature map, the recognition result is obtained.

[0090] The improved encoder is as follows: except for the last layer, the output of each encoder layer is connected to an SPD-Conv module, and the output of the previous SPD-Conv module is used as the input of the current encoder layer.

[0091] It should be noted that the identification module and identification model module described above implement the same examples and application scenarios as those in Embodiment 1, but are not limited to the content disclosed in Embodiment 1. It should also be noted that the above modules, as part of a system, can be executed in a computer system such as a set of computer-executable instructions.

[0092] Example 3

[0093] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the electrocardiogram sub-waveform recognition method as described in Embodiment 1 above.

[0094] Example 4

[0095] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the electrocardiogram sub-waveform recognition method as described in Embodiment 1 above.

[0096] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0097] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0098] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0099] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0100] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0101] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for identifying sub-waveforms in an electrocardiogram (ECG), characterized in that, include: Based on the acquired electrocardiogram data, a recognition model is used to obtain the recognition results; The process of adopting the recognition model is as follows: the recognition model adopts an improved U^2-Net model, which includes an encoder, a classification module and a decoder; Based on electrocardiogram data, an improved encoder is used to obtain a first feature map; based on the first feature map, a classification module and a decoder are used respectively to obtain a second feature map and a third feature map; based on the fusion feature map of the second and third feature maps, the recognition result is obtained. The improved encoder is as follows: except for the last layer, the output of each encoder layer is connected to an SPD-Conv module, and the output of the previous SPD-Conv module is used as the input of the current encoder layer.

2. The method for identifying electrocardiogram sub-waveforms according to claim 1, characterized in that, The improved encoder is as follows: the encoder in the first part adopts the RSU module, and the encoder in the second part adopts the RSU_3F module. F indicates that the dilated convolution operation of RSU is used instead of the downsampling operation inside the RSU module. Before using the RSU module, the SCSE module is used to extract the sequence features of the electrocardiogram data, and the extracted feature sequence is input into the RSU module for encoding.

3. The method for identifying electrocardiogram sub-waveforms according to claim 1, characterized in that, The structure of the decoder in each layer corresponds to the structure of the encoder in the same level.

4. The method for identifying electrocardiogram sub-waveforms according to claim 1, characterized in that, The input to each encoder layer is: the upsampled feature map output by the previous level decoder and the feature map output by the encoder of the same level corresponding to the current level decoder.

5. The method for identifying electrocardiogram sub-waveforms according to claim 1, characterized in that, The classification module includes a pooling layer, a BatchNorm1d module, a multilayer perceptron, a sigmoid module, and a threshold module connected in sequence.

6. The method for identifying electrocardiogram sub-waveforms according to claim 5, characterized in that, The second feature map obtained by the classification module is the feature map output by the sigmoid module; Alternatively, the second feature map obtained by the classification module may be the feature map output by the threshold module.

7. The method for identifying electrocardiogram sub-waveforms according to claim 1, characterized in that, The training process of the recognition model includes: using the FocalLoss loss function based on the second feature map output by the classification module; using the FocalLoss loss function and the DiceLoss loss function based on the third feature map output by the decoder; and optimizing the hyperparameters of the recognition model based on the loss results.

8. A system for recognizing electrocardiogram (ECG) sub-waveforms, characterized in that, include: The recognition module is configured to: obtain recognition results based on the acquired electrocardiogram data using a recognition model; The recognition model module is configured such that the recognition model adopts an improved U^2-Net model, including an encoder, a classification module, and a decoder; Based on electrocardiogram data, an improved encoder is used to obtain a first feature map; based on the first feature map, a classification module and a decoder are used respectively to obtain a second feature map and a third feature map; based on the fusion feature map of the second and third feature maps, the recognition result is obtained. The improved encoder is as follows: except for the last layer, the output of each encoder layer is connected to an SPD-Conv module, and the output of the previous SPD-Conv module is used as the input of the current encoder layer.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the ECG sub-waveform recognition method as described in any one of claims 1-7.

10. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the ECG sub-waveform recognition method as described in any one of claims 1-7.

Images