Electrocardiosignal anomaly classification method and apparatus

Abstract

The application discloses an electrocardiosignal abnormality classification method and device, which comprises the following steps: receiving a multi-lead electrocardiosignal data set acquired by a multi-lead acquisition device; constructing an intra-lead graph structure for representing the internal rhythm change of an electrocardiosignal waveform for each lead electrocardiosignal sequence; constructing an inter-lead external graph structure based on the waveform synchronism and correlation between different leads; further determining the external distribution mode of the electrocardiosignal under different physiological states or individual differences based on the intra-lead graph structure and the inter-lead external graph structure; then fusing the graph structure and the external distribution mode to generate a global graph representation of the electrocardiosignal; and inputting the graph representation into a classifier to obtain a heart rhythm abnormality classification result. The application can model the local waveform morphological features, cross-lead correlation features and distribution changes caused by cross-population differences of the electrocardiosignal at the same time, thereby effectively improving the accuracy, robustness and cross-subject generalization ability of the electrocardiosignal abnormality classification task.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical signal analysis technology, and more specifically, to a graph structure representation method and apparatus for classifying arrhythmias in electrocardiogram (ECG) signals. Background Technology

[0002] Electrocardiogram (ECG) is an important physiological signal reflecting the electrical activity of the heart, and multi-lead ECG is widely used in clinical screening and diagnosis of arrhythmias. Common arrhythmias such as premature atrial contractions (PACs), premature ventricular contractions (PVCs), and atrial fibrillation show significant differences in waveform morphology, periodic rhythm, and translead potential distribution. Therefore, accurately identifying arrhythmias from multi-lead ECG signals is an important research direction for the auxiliary diagnosis of cardiovascular diseases.

[0003] Existing ECG abnormality classification methods are mainly based on deep models such as convolutional neural networks, recurrent neural networks, or Transformers, directly representing ECG signals end-to-end. However, these methods typically focus only on waveform features within a single lead, neglecting the physiological correlations between different leads, such as the coupling relationship of ECG potential spatial distribution on the body surface. Furthermore, individual differences among subjects, changes in physiological state, and variations in acquisition equipment can lead to inconsistencies in ECG waveform distribution, i.e., significant out-of-distribution (OOD) across individuals. Traditional methods struggle to maintain stable classification performance across different population groups and scenarios, exhibiting insufficient generalization ability.

[0004] Therefore, existing technologies lack a method for classifying ECG abnormalities that can simultaneously model the internal periodic morphological structure of leads, the cross-spatial correlation structure between leads, and explicitly model cross-individual distribution differences, in order to improve the robust classification ability of multi-lead ECG signals. Summary of the Invention

[0005] The present invention aims to overcome the above-mentioned shortcomings of the prior art and provide a method and apparatus for classifying abnormal electrocardiogram signals.

[0006] A first aspect of the present invention provides a method for classifying abnormal electrocardiogram signals, comprising:

[0007] The system receives a multi-lead ECG signal dataset acquired by a multi-lead ECG acquisition device; for each lead signal in the multi-lead ECG signal dataset, it determines the internal graph structure of each lead; wherein each internal graph structure represents the waveform and rhythm information within a single lead ECG signal; for the multi-lead ECG signal dataset, it determines the external graph structure between leads; wherein the external graph structure between leads represents the synchronization and morphological correlation of ECG waveforms between multiple leads; based on the internal graph structure and the external graph structure between leads, it determines the external distribution pattern of the ECG signal dataset, which is used to distinguish the differences in ECG waveform distribution caused by different subjects, different physiological states, different levels of autonomic nerve excitation, or different myocardial conduction conditions; based on the internal graph structure, the external graph structure between leads, and the external distribution pattern, it generates a graph representation of the multi-lead ECG signal dataset; it inputs the graph representation into an ECG abnormality classifier to obtain the ECG abnormality classification result output by the classifier, such as atrial premature beats, ventricular premature beats, atrial fibrillation, or other abnormal rhythm types.

[0008] The construction of the lead internal diagram structure in step S2 includes:

[0009] S21. Divide the ECG signal of each lead into multiple waveform segments according to a fixed time window or heartbeat cycle, and each segment is a node in the figure;

[0010] S22. Calculate the waveform cosine similarity between the target segment m and other segments n;

[0011] S23. Based on the cosine similarity, select a specified number of segments with the highest similarity as neighbor nodes;

[0012] S24. Construct a fragment-level adjacency matrix based on the target fragment node and its neighboring nodes;

[0013] S25. Use a graph neural network to aggregate the features between nodes in the multiple segment-level adjacency matrices to obtain the internal graph structure of the lead.

[0014] Step S3, which involves constructing the inter-lead external graph structure, includes:

[0015] S31. Identify a target lead among multiple leads;

[0016] S32. Calculate the waveform morphology cosine similarity between the target lead and other leads;

[0017] S33. Generate a similarity matrix between leads by taking all leads as target leads in sequence;

[0018] S34. Based on the similarity matrix, select the specified number of leads with the highest similarity as neighbor nodes;

[0019] S35. Add a location embedding feature to each lead node in the global graph;

[0020] S36. Based on graph Laplacian calculation, the feature vectors are fused into the global graph, and the graph neural network is used to aggregate information between nodes to obtain the external graph node features between the leads.

[0021] Among them, step S4 extracts the external distribution pattern of the electrocardiogram signal and obtains a global graph representation, including;

[0022] S41. Construct a neighborhood subgraph for each node, and treat each neighborhood subgraph as an independent distribution;

[0023] S42. Based on a graph neural network, information is aggregated from each of the neighborhood subgraphs to obtain global graph features;

[0024] S43. Perform max pooling on global graph features to enhance feature representation.

[0025] Step S5 involves classifying the distribution of the electrocardiogram signal, including:

[0026] S51. Use a domain classifier to perform domain identification on the global graph representation to obtain the domain identification result;

[0027] S52. The domain identification results are used as a reference for parameter optimization to update part of the modeling parameters for the lead internal graph structure and the inter-lead external graph structure.

[0028] In step S6, the electrocardiogram signal is classified based on the global graph representation:

[0029] S61. Use a classification model to perform the final classification of ECG abnormality types on the global graph representation;

[0030] S62. The anomaly classification result is used as a reference for parameter optimization to update part of the modeling parameters for the lead internal graph structure and the inter-lead external graph structure;

[0031] S63. Combining the anomaly classification results and distribution classification results, the modeling parameters for updating the internal graph structure of leads and the external graph structure between leads are updated.

[0032] In a second aspect, the present invention provides a cardiac signal abnormality classification device, comprising:

[0033] The receiving unit is used to receive multi-lead electrocardiogram signal datasets acquired by the multi-lead electrocardiogram acquisition device;

[0034] The lead internal structure construction unit is used to construct a lead internal diagram structure to characterize the internal rhythm change features of the electrocardiogram for each lead signal sequence in the multi-lead electrocardiogram signal dataset.

[0035] The inter-lead structure building unit is used to construct an inter-lead external graph structure to characterize the relationship between leads based on the waveform synchronicity and morphological correlation between the ECG signals of each lead.

[0036] The external distribution pattern extraction unit is used to extract external distribution patterns based on the internal lead graph structure and the external lead graph structure to distinguish differences in electrocardiogram waveforms caused by different subjects, different physiological states, different myocardial conduction conditions, or different levels of autonomic nerve excitation. The external distribution pattern extraction unit performs this extraction in the following manner:

[0037] For each node, construct a neighborhood subgraph and treat it as an independent distribution domain;

[0038] Information is aggregated from each neighborhood subgraph based on graph neural network to obtain node domain features;

[0039] Max pooling is performed on the aggregated node domain features to enhance feature representation, resulting in the external distribution pattern.

[0040] A global graph representation generation unit is used to generate a global graph representation of the electrocardiogram signal based on the internal graph structure of the leads, the external graph structure between the leads, and the external distribution pattern.

[0041] The distribution classification unit is used to identify the distribution differences within the electrocardiogram signal based on the domain classification results corresponding to the external distribution pattern, and to generate a distribution classification output for updating model parameters.

[0042] The classification unit is used to input the global graph representation into the ECG abnormality classifier to obtain the ECG abnormality classification result.

[0043] The working principle of this invention is:

[0044] This invention achieves unified modeling of the spatiotemporal characteristics of ECG signals by converting multi-lead ECG signals into a multi-layer graph structure that simultaneously reflects local rhythm structure, interlead spatial relationships, and external distribution differences. Specifically, firstly, the internal graph structure of each lead is used to characterize the correlation between the heartbeat rhythm pattern and waveform morphology within a single lead; further, based on the external graph structure between leads, the synchronicity, temporal consistency, and morphological similarity of different leads in the conduction process of cardiac electrical activity are modeled to form a global correlation structure across leads. Subsequently, by constructing subgraphs in the neighborhood of nodes and performing domain-level feature aggregation, external distribution patterns are introduced to characterize the differences in overall waveform distribution caused by different subjects, different physiological states, or different myocardial conduction conditions.

[0045] Building upon this foundation, this invention identifies the sub-graph domain features corresponding to each node through a distribution classification unit, thereby establishing cross-sample distribution constraints. These constraints, along with the global graph representation output by the global graph representation generation unit, serve as input to the classifier, enabling the detection of cardiac rhythm abnormalities. Ultimately, this invention achieves this by jointly optimizing the domain classification loss and the task classification loss, allowing the model to simultaneously learn internal rhythms, inter-lead correlations, and cross-individual distribution differences, thus ensuring the stability and generalization ability of ECG abnormality detection.

[0046] The innovation of this invention is:

[0047] First, this invention models multi-lead electrocardiogram signals into a unified multi-layer graph structure composed of internal graph structure of leads, external graph structure of leads, and cross-population distribution pattern, so that the local rhythm characteristics, spatial conduction characteristics and cross-sample distribution characteristics of electrocardiogram signals can be synergistically expressed in the same framework.

[0048] Secondly, this invention proposes an external distribution pattern extraction method based on node neighborhood subgraphs. By constructing local subgraphs, graph neural network aggregation, and max pooling enhancement, it achieves explicit modeling of distribution differences among different subjects and different electrocardiographic states, effectively improving the robustness of the model in cross-individual and cross-condition scenarios.

[0049] Third, this invention introduces a distribution classification unit, which generates distribution labels for each node domain subgraph and designs a domain classification loss, so that the model simultaneously constrains the temporal internal structure, translead structure and cross-sample distribution structure during the training process, which significantly improves the generalization performance of ECG abnormality classification.

[0050] Fourth, the global graph representation constructed in this invention realizes the structured fusion of multidimensional features of ECG signals, enabling the classification model to simultaneously perceive local rhythm abnormalities, translead conduction abnormalities, and cross-individual distribution abnormalities in a unified space, thereby achieving more accurate and robust ECG abnormality detection.

[0051] The present invention has the following beneficial effects: The ECG signal abnormality classification method proposed in this application generates a global graph representation of a multi-lead ECG signal dataset by extracting and fusing the internal graph structure of leads, the external graph structure between leads, and the external distribution pattern. This method not only captures the ECG waveform structure information within a single lead and the physiological conduction correlation between leads, but also considers the external distribution variations caused by differences in ECG waveforms among different subjects. Therefore, it can more comprehensively represent ECG signals and improve the accuracy and cross-population generalization ability of arrhythmia classification tasks. In particular, the external distribution pattern modeling fully considers the physiological variation characteristics of ECG signals, improving its adaptability to actual clinical screening and long-term monitoring scenarios. Attached Figure Description

[0052] Figure 1A flowchart of a method for classifying abnormal electrocardiogram signals according to an embodiment of this application is shown;

[0053] Figure 2 A block diagram of an electrocardiogram signal abnormality classification device according to an embodiment of this application is shown;

[0054] Figure 3 This is a block diagram of an electronic device for classifying abnormal electrocardiogram signals, according to an exemplary embodiment. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0056] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the application. The singular forms “a,” “said,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0057] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0058] It should be understood that although the terms first, second, third, etc., may be used in the embodiments of this application, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, first may also be referred to as second without departing from the scope of the embodiments of this application, and similarly, second may also be referred to as first.

[0059] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0060] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0061] It should be noted that any symbols and / or numbers present in the specification that are not marked in the accompanying drawings are not reference numerals.

[0062] The classification of abnormal electrocardiogram (ECG) signals is of great significance in practical medical monitoring and health management, especially when using multi-lead ECG acquisition equipment for real-time detection. Multi-lead signals reflect the distribution of cardiac electrical activity at different locations on the body surface. Their waveform structure not only changes with the heart rate but is also affected by differences in myocardial conduction pathways, body size, electrode position, and physiological state.

[0063] Current ECG classification methods typically employ convolutional neural networks (CNNs), recurrent neural networks (RNNs), or Transformer models to directly extract features from ECG waveforms. However, these methods often focus only on the waveform structure within a single lead, neglecting the synchronization relationships and complementary information between different leads during cardiac electrical conduction. In multi-lead ECG signals, there are clear physiological conduction relationships between leads. If this information is ignored, the classification results are easily affected by factors such as waveform variations, noise, and electrode contact conditions.

[0064] Furthermore, there are significant differences in electrocardiogram waveforms among different populations. Factors such as myocardial thickness, conduction velocity, heart rate variability, fatigue, and emotional state can all lead to different manifestations of the same type of arrhythmia in different individuals, i.e., there is a significant cross-subject distribution difference (OOD). Traditional methods tend to suffer from decreased classification performance and insufficient generalization ability when dealing with these physiological differences.

[0065] Based on the above, this application models the structural information of the multi-lead electrocardiogram signal itself: (1) extracting the relationship of heartbeat morphology changes within each lead; (2) modeling the spatial conduction correlation of cardiac electrical activity between leads; and (3) considering the differences in electrocardiogram distribution brought about by different subjects.

[0066] By incorporating the internal rhythmic structure of ECG signals, physiological correlations between leads, and cross-population differences into graph structure modeling, a more stable, robust, and clinically meaningful signal representation can be obtained, thereby improving the accuracy of arrhythmia classification and cross-population generalization ability.

[0067] The optional embodiments of this application are described in detail below with reference to the accompanying drawings.

[0068] Example 1

[0069] Figure 1 A flowchart of a method for classifying abnormal electrocardiogram signals according to an embodiment of this application is shown.

[0070] The ECG signal abnormality classification method proposed in this application generates a global graphical representation of a multi-lead ECG signal dataset by extracting and fusing the internal graph structure of leads, the external graph structure between leads, and the external distribution pattern. Figure 1 As shown, this method not only captures detailed changes in waveforms within a single lead and the electrophysiological conduction correlations between leads, but also considers the differences in ECG distribution among different subjects, thus enabling a more comprehensive characterization of ECG signals and improving the accuracy of arrhythmia classification tasks and cross-population generalization performance. In particular, the modeling of external distribution patterns fully considers the physiological variability of ECG signals, improving the stability and robustness of the method in real-world monitoring, dynamic recording, and long-term continuous monitoring scenarios.

[0071] A method for classifying abnormal electrocardiogram signals includes the following steps:

[0072] Step S1: Receive the multi-lead ECG signal dataset acquired by the multi-lead ECG acquisition device.

[0073] This step is used to obtain the raw ECG signals input to the model. Specifically, multi-lead ECG acquisition devices (such as Holter monitors, ECG patches, clinical 12-lead ECG devices, etc.) are used to collect the subjects' ECG signals in resting, exercise, or daily monitoring scenarios, forming a multi-lead ECG signal dataset. Assume the acquired ECG leads constitute a multivariate time series, a dataset with... A multi-lead ECG signal can be represented as Each contains ECG signal sequence at each time step One of the datasets for classifying multi-lead ECG signal abnormalities, containing M samples, can be defined as follows: Each sample Each has a corresponding category label. Based on this, a graph can be defined as... ,in, and These are sets of nodes and edges, respectively. An electrocardiogram (ECG) signal, after conversion, can be represented as a graph.

[0074] Example, a node With another node The edge between them is defined as Adjacency matrix It can represent the connection relationship between nodes, where Represents the matrix dimension, if but Otherwise, it is 0.

[0075] Step S2: For each lead signal sequence, construct the lead internal graph structure to characterize the internal rhythm characteristics of the ECG waveform.

[0076] This step aims to capture the waveform morphology relationships and rhythm dependence within a single lead ECG signal.

[0077] S21. The first lead signal The graph is divided into multiple segments based on a fixed window or a heartbeat cycle, with each segment being a node in the graph, as shown in formula (1):

[0078]

[0079] in This refers to the segment length (which can correspond to a heartbeat segment).

[0080] S22. In this application, cosine similarity is used to characterize whether the waveforms of two segments are similar, as shown in formula (2):

[0081]

[0082] in, For the first Similarity between leads It is the Frobenius norm.

[0083] S23. Based on the similarity matrix, retain only the most similar segments for each segment. The connections form a sparse adjacency matrix, as shown in formula (3):

[0084]

[0085] in, For the first The adjacency matrix of each lead. This represents the number of segments.

[0086] S24. Based on the method for constructing neighbor nodes, apply it to all leads to construct a segment-level adjacency matrix.

[0087] S25. Subsequently, a graph neural network is used to aggregate node features of the segment sequence. Graph neural networks are a class of neural network models used to process graph-structured data. Examples include GCN (Graph Convolutional Network) proposed in 2017, suitable for local feature propagation in structured graph signals; GAT (Graph Attention Network) proposed in 2018, which learns the importance of neighboring nodes through an attention mechanism; and GraphSAGE proposed in 2017, which supports inductive graph representation learning. This invention, without limiting itself to a specific graph neural network model, may choose any of the above-disclosed techniques.

[0088] The fragment nodes and their adjacency matrices are input into the graph neural network for information aggregation to obtain the internal representation of the leads, as shown in formula (4):

[0089]

[0090] in, For graph neural network models, Features of the lead internal diagram, The feature dimensions of the lead internal diagram.

[0091] Step S3: Based on the waveform synchronicity and morphological correlation between the signals of each lead, construct an inter-lead external graph structure to characterize the inter-lead correlation.

[0092] This step is used to capture spatial correlation, conduction synchronicity, and morphological coupling characteristics between different leads.

[0093] S31. Identify a target lead among multiple leads.

[0094] S32. The node representations of the internal diagram of each lead serve as the overall characteristics of that lead. For the first lead... The and the first The cosine similarity of each lead is calculated to determine the graph structure between variables, as shown in formula (5):

[0095]

[0096] in, For the first Similarity between leads It is the Frobenius norm.

[0097] S33. Based on the aforementioned similarity calculation method, extend it to all leads, and generate an inter-lead similarity matrix by sequentially using all leads as target leads.

[0098] S34. After obtaining the similarity between leads, retain the most relevant lead. One lead, Given the pre-set parameters, the adjacency matrix is ​​obtained as shown in formula (6):

[0099]

[0100] in, This is the inter-lead adjacency matrix. The number of leads.

[0101] S35. To preserve the body surface spatial location of the lead adjacency matrix, this application uses graph Laplacian embedding, assuming... For the inter-lead graph structure, its graph Laplace matrix is ​​calculated as shown in formula (7):

[0102]

[0103] in, For the graph Laplace operator, For the image eigenvector matrix, For the image eigenvalues.

[0104] S36. To enhance feature representation, lead location information is encoded into node features. This embodiment introduces location embedding in the following manner, as shown in formula (8):

[0105]

[0106] in, This indicates a feature concatenation operation. This is a dot product function for sampling eigenvectors based on a random sign matrix. It is a random symbol matrix, whose elements take values ​​of or This is used to maintain the consistency of the feature vector sign.

[0107] The above steps yield the inter-lead external diagram structure containing location information. And by using a graph neural network to aggregate information between nodes, the aggregated features of external nodes between leads are obtained, as shown in formula (9):

[0108]

[0109] in, For graph neural network models, For the internal graph node features of the j-th lead, This refers to the preset feature dimensions of the internal diagram nodes of the lead.

[0110] Constructed from this It also includes: local rhythm correlation information within leads (from step S2); inter-lead correlation structure between leads (from similarity matrix E); and the relative positional relationships of leads in the body surface space (from position embedding). Therefore, This approach provides a comprehensive representation of the global patterns of multi-lead ECG signals, effectively describing the synchronicity, morphological differences, and spatial topological structure among different leads, and providing a basic representation for subsequent steps such as external distribution pattern extraction and anomaly classification.

[0111] Step S4: Based on the internal graph structure of leads and the external graph structure between leads, extract the external distribution pattern of ECG signals and obtain a global graph representation.

[0112] The purpose of step S4 is to further extract the external distribution pattern that can distinguish the differences in ECG waveforms produced by different subjects and different physiological states, based on the internal lead graph structure and external lead graph structure obtained in steps S2 and S3. This step achieves explicit modeling of distribution differences by constructing node-level local sub-graph domains and aggregated domain features.

[0113] S41. For external graph structures For each node q in the graph, select the one with the highest correlation. Each neighboring node constructs a node-level distribution domain. Each sub-domain is treated as a distribution, thereby extracting the external distribution pattern of the electrocardiogram signal. This is the preset field size parameter.

[0114] S42. The sub-region Input graph neural network model Information aggregation is performed to obtain the subgraph domain embedding, as shown in formula (10):

[0115]

[0116] in, Represents a node Subgraph domain embedding, The aggregated feature dimensions.

[0117] S43. Max pooling is applied to the features of the q nodes in the updated global graph to enhance feature representation, as shown in formula (11):

[0118]

[0119] in, This is for max pooling.

[0120] After obtaining the updated global graph Then, a graph neural network is used to process the global graph to obtain the final node-level representation, as shown in formula (12):

[0121]

[0122] in For the final node-level representation, The number of features is a pre-defined hyperparameter.

[0123] Step S5: Based on the external distribution pattern and global graph representation of the ECG signal, classify the distribution existing inside the ECG signal;

[0124] This step classifies the distribution based on the local sub-graph domain and aggregate domain features constructed in step S4, thereby extracting the differences in the distribution.

[0125] S51. Move the node Features Input domain classifier This is used to identify the external distribution domain to which it belongs, as shown in formula (13):

[0126]

[0127] S52. And construct the domain classification loss using cross-entropy loss, as shown in formula (14):

[0128]

[0129] in, Let cross-entropy be the loss function. It also serves as a distribution label for the node itself, used to construct monitoring signals.

[0130] Step S6: Input the global graph representation into the classifier to obtain the ECG signal abnormality classification result.

[0131] This step obtains the final ECG signal abnormality classification result through the global graph representation.

[0132] S61. Represent the global graph obtained in step S5. Input classification model As shown in formula (15):

[0133]

[0134] in, It can be a multilayer perceptron, convolutional neural network, or other publicly available classification structure. Output the heart rate state category for each sample.

[0135] S62. This implementation example uses cross-entropy loss for training, as shown in formula (16):

[0136]

[0137] in, For real labels, This is the cross-entropy loss function.

[0138] Combining the distribution classification loss and the task classification loss, the total loss function of this application is obtained as shown in formula (17):

[0139]

[0140] Among them is the loss weight coefficient, used to balance the contributions of distribution classification loss and task classification loss.

[0141] Example 2

[0142] This embodiment provides a device for detecting abnormal electrocardiogram (ECG) signals, which can be implemented through a combination of hardware and software. The device includes multiple functional modules, which interact with each other to complete the abnormal detection process. This device can be deployed in servers, wearable devices, medical monitoring terminals, or other electronic devices with computing capabilities.

[0143] This application also provides apparatus embodiments corresponding to the above embodiments, used to implement the above method steps. The functions of each unit or module in the apparatus are the same as those in the corresponding method embodiments, and the same technical effects can be achieved, which will not be repeated here.

[0144] like Figure 2 As shown, this application provides an electrocardiogram signal abnormality classification device 200, comprising:

[0145] The receiving unit 201 is used to receive a multi-lead ECG signal dataset acquired by a multi-lead ECG acquisition device. The determining unit 202 is used to determine the internal graph structure of each lead in the multi-lead ECG signal dataset; wherein each internal graph structure is used to characterize the morphological and rhythmic changes of the heartbeat waveform within a single lead. For the multi-lead ECG signal dataset, the inter-lead external graph structure is determined; wherein the inter-lead external graph structure is used to characterize the waveform synchronicity and physiological correlation between multiple leads during cardiac electrical activity conduction. Based on the internal graph structure and the inter-lead external graph structure, the unit determines the external distribution pattern of the multi-lead ECG signal dataset to reflect the differences in ECG waveform distribution caused by different subjects, different physiological states, or myocardial conduction conditions. The generating unit 203 is used to generate a global graph representation of the multi-lead ECG signal dataset based on the internal graph structure, the inter-lead external graph structure, and the external distribution pattern. The global graph representation is used to input the ECG abnormality classifier to obtain the cardiac rhythm abnormality classification result output by the ECG abnormality classifier.

[0146] In one embodiment, the lead-internal graph structure and the inter-lead external graph structure each contain multiple different nodes. The determining unit 202 determines the external distribution pattern of the multi-lead ECG signal dataset based on the lead-internal graph structure and the inter-lead external graph structure as follows: for each node, a neighborhood subgraph is constructed, and each neighborhood subgraph is treated as an independent distribution domain; information is aggregated from each neighborhood subgraph using a graph neural network to obtain aggregated graph information; the aggregated graph information is subjected to max pooling to obtain enhanced graph information; for the enhanced graph information, domain information is extracted from each neighborhood subgraph, and the enhanced graph information is further differentiated into distribution domains based on a domain classifier, thereby obtaining the external distribution pattern.

[0147] In one embodiment, the determining unit 202 determines the inter-lead external graph structure as follows: a target lead is identified among multiple leads; the waveform cosine similarity between the target lead and other leads is calculated, and a specified number of leads with the highest similarity are designated as neighboring nodes; a lead-level adjacency matrix is ​​constructed based on the corresponding node of the target lead and its neighboring nodes; multiple leads are used as the target lead to obtain multiple adjacency matrices; a global lead graph structure of the multi-lead ECG signal dataset is generated based on the multiple leads and multiple adjacency matrices; wherein each lead is considered a node in the graph, and the connection relationship between nodes is determined by the adjacency matrix; location embedding features are added to each node in the global graph to reflect the electrophysiological location distribution of the lead in the body surface space; the feature vector is incorporated into each node in the global graph based on the Laplacian graph calculation, and the inter-node information in the global graph is aggregated based on a graph neural network to obtain the inter-lead external graph structure.

[0148] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the corresponding method embodiments, and will not be repeated here.

[0149] Figure 3 This is a block diagram of an electronic device 300 for classifying abnormalities in multi-lead electrocardiogram signals, according to an exemplary embodiment.

[0150] like Figure 3 As shown, one embodiment of this application provides an electronic device 300. The electronic device 300 includes a memory 301, a processor 302, and an input / output (I / O) interface 303. The memory 301 is used to store instructions. The processor 302 is used to execute the electrocardiogram (ECG) signal abnormality classification method of this application embodiment by calling the instructions stored in the memory 301. The processor 302 is connected to both the memory 301 and the I / O interface 303, for example, via a bus system and / or other forms of connection mechanisms (not shown). The memory 301 can be used to store programs and data, including the program for the ECG signal abnormality classification method involved in the embodiments of this application. The processor 302 executes various functional applications and data processing of the electronic device 300 by running the program stored in the memory 301.

[0151] In this embodiment, the processor 302 can be implemented in at least one of the following hardware forms: Digital Signal Processor (DSP), Field Programmable Gate Array (FPGA), Programmable Logic Array (PLA). The processor 302 can be one or a combination of several of the following: Central Processing Unit (CPU) or other processing units with data processing capability and / or instruction execution capability.

[0152] The memory 301 in this embodiment may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0153] In this embodiment, the I / O interface 303 can be used to receive input instructions (such as numeric or character information, and to generate key signal inputs related to user settings and function control of the electronic device 300), and can also output various information (such as images or sounds) to the outside. In this embodiment, the I / O interface 303 may include one or more of the following: a physical keyboard, function keys (such as volume control keys, power buttons, etc.), a mouse, a joystick, a trackball, a microphone, a speaker, and a touch panel.

[0154] In some embodiments, this application provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, perform any of the methods described above.

[0155] In some embodiments, this application provides a computer program product comprising a computer program that, when executed by a processor, performs any of the methods described above.

[0156] Although the operations are described in a specific order in the accompanying drawings, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all of the operations shown to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0157] The methods and apparatus of this application can be implemented using standard programming techniques, utilizing rule-based logic or other logic to implement various method steps. It should also be noted that the terms "apparatus" and "module" as used herein and in the claims are intended to include implementations using one or more lines of software code and / or hardware implementations and / or devices for receiving input.

[0158] Any step, operation, or procedure described herein may be performed or implemented using one or more hardware or software modules, either alone or in combination with other devices. In one embodiment, the software module is implemented using a computer program product comprising a computer-readable medium containing computer program code, which is executable by a computer processor to perform any or all of the described steps, operations, or procedures.

[0159] The foregoing description of implementations of this application has been provided for illustrative and descriptive purposes. The foregoing description is not exhaustive and is not intended to limit this application to the exact forms disclosed. Various modifications and variations may exist in accordance with the foregoing teachings, or may arise from practice of this application. These embodiments were chosen and described to illustrate the principles of this application and its practical application, enabling those skilled in the art to utilize this application in various implementations and modifications to suit the specific purpose of the concept.

[0160] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0161] It can be further understood that, unless otherwise specified, "connection" includes both direct connections where no other components exist between the two parties and indirect connections where other components exist between them.

[0162] It is further understood that although the operations are described in a specific order in the accompanying drawings in the embodiments of this application, this should not be construed as requiring these operations to be performed in the specific order or serial order shown, or requiring all the operations shown to be performed to obtain the desired result. In certain environments, multitasking and parallel processing may be advantageous.

[0163] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the field of this application that are not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0164] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

[0165] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for classifying abnormal electrocardiogram signals, characterized in that, include: S1. Receive the multi-lead electrocardiogram (ECG) signal dataset acquired by the multi-lead ECG acquisition device; S2. For each lead signal sequence, construct a lead internal diagram structure to characterize the internal rhythm characteristics of the ECG waveform. The lead internal diagram structure is a representation of the waveform and rhythm information inside a single lead ECG signal. S3. Based on the waveform synchronization and morphological correlation between the signals of each lead, construct an inter-lead external graph structure to characterize the correlation between leads; S4. Based on the internal lead diagram structure and the external lead diagram structure, extract the external distribution pattern and global diagram representation of the electrocardiogram signal. The external distribution pattern is used to distinguish the differences in electrocardiogram waveform distribution caused by different subjects, different physiological states, different levels of autonomic nerve excitation, or different myocardial conduction conditions. The extraction of the external distribution pattern and global diagram representation of the electrocardiogram signal includes: S41. Construct a neighborhood subgraph for each node, and treat each neighborhood subgraph as an independent distribution; S42. Based on a graph neural network, information is aggregated from each of the neighborhood subgraphs to obtain global graph features; S43. Perform max pooling on global graph features to enhance feature representation; S5. Based on the external distribution pattern and global graph representation of the electrocardiogram (ECG) signal, classify the distributions existing within the ECG signal, including: S51. Use a domain classifier to perform domain identification on the global graph representation to obtain the domain identification result; S52. The domain identification results are used as a reference for parameter optimization to update part of the modeling parameters for the lead internal graph structure and the inter-lead external graph structure; S6. Input the global graph representation into the classifier to obtain the ECG abnormality classification results, including: S61. Use a classification model to perform the final classification of ECG abnormality types on the global graph representation; S62. The anomaly classification result is used as a reference for parameter optimization to update part of the modeling parameters for the lead internal graph structure and the inter-lead external graph structure; S63. Combining the anomaly classification results and distribution classification results, the modeling parameters for updating the internal graph structure of leads and the external graph structure between leads are updated.

2. The method according to claim 1, characterized in that, Step S2, which involves constructing the internal diagram structure of the lead, includes: S21. Divide the ECG signal of each lead into multiple waveform segments according to a fixed time window or heartbeat cycle, and each segment is a node in the figure; S22. Calculate the waveform cosine similarity between the target segment m and other segments n; S23. Based on the cosine similarity, select a specified number of segments with the highest similarity as neighbor nodes; S24. Construct a fragment-level adjacency matrix based on the target fragment node and its neighboring nodes; S25. Use a graph neural network to aggregate the features between nodes in the multiple segment-level adjacency matrices to obtain the internal graph structure of the lead.

3. The method according to claim 1, characterized in that, Step S3, which involves constructing the inter-lead external graph structure, includes: S31. Identify a target lead among multiple leads; S32. Calculate the waveform morphology cosine similarity between the target lead and other leads; S33. Generate a similarity matrix between leads by taking all leads as target leads in turn; S34. Based on the similarity matrix, select the specified number of leads with the highest similarity as neighbor nodes; S35. Add a location embedding feature to each lead node in the global graph; S36. Based on graph Laplacian calculation, the feature vectors are fused into the global graph, and the graph neural network is used to aggregate information between nodes to obtain the external graph node features between the leads.

4. A device for classifying abnormal electrocardiogram signals, characterized in that, include: The receiving unit is used to receive multi-lead electrocardiogram signal datasets acquired by the multi-lead electrocardiogram acquisition device; The lead internal structure construction unit is used to construct a lead internal diagram structure to characterize the internal rhythm change features of the electrocardiogram for each lead signal sequence in the multi-lead electrocardiogram signal dataset. The inter-lead structure building unit is used to construct an inter-lead external graph structure to characterize the relationship between leads based on the waveform synchronicity and morphological correlation between the ECG signals of each lead. The external distribution pattern extraction unit is used to extract external distribution patterns based on the internal map structure of the leads and the external map structure between the leads to distinguish the differences in electrocardiogram waveforms caused by different subjects, different physiological states, different myocardial conduction conditions, or different levels of autonomic nerve excitation. The external distribution pattern extraction unit performs this extraction in the following manner: For each node, construct a neighborhood subgraph and treat it as an independent distribution domain; Information is aggregated from each neighborhood subgraph based on graph neural network to obtain node domain features; Max pooling is performed on the aggregated node domain features to enhance feature representation, resulting in the external distribution pattern. A global graph representation generation unit is used to generate a global graph representation of the electrocardiogram signal based on the internal graph structure of the leads, the external graph structure between the leads, and the external distribution pattern. A distribution classification unit is used to identify distribution differences within the electrocardiogram signal based on the domain classification results corresponding to the external distribution pattern, and to generate a distribution classification output for model parameter updates, including: S51. Use a domain classifier to perform domain identification on the global graph representation to obtain the domain identification result; S52. The domain identification results are used as a reference for parameter optimization to update part of the modeling parameters for the lead internal graph structure and the inter-lead external graph structure; A classification unit, used to input the global graph representation into an ECG abnormality classifier to obtain ECG abnormality classification results, includes: S61. Use a classification model to perform the final classification of ECG abnormality types on the global graph representation; S62. The anomaly classification result is used as a reference for parameter optimization to update part of the modeling parameters for the lead internal graph structure and the inter-lead external graph structure; S63. Combining the anomaly classification results and distribution classification results, the modeling parameters for updating the internal graph structure of leads and the external graph structure between leads are updated.

5. A device for classifying abnormal electrocardiogram signals, 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 computer program, it implements the electrocardiogram signal abnormality classification method according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the electrocardiogram signal abnormality classification method according to any one of claims 1 to 3.

Images