CN-121465604-B - Electrocardiogram signal abnormality classification method and device

CN121465604BCN 121465604 BCN121465604 BCN 121465604BCN-121465604-B

Abstract

The electrocardiosignal anomaly classification method and device comprise the steps of receiving a multi-lead electrocardiosignal data set acquired by multi-lead acquisition equipment, constructing a lead internal graph structure for representing the internal rhythm change of an electrocardiosignal for each lead electrocardiosignal sequence, constructing an inter-lead external graph structure based on waveform synchronism and correlation among different leads, further determining an external distribution mode of the electrocardiosignal under different physiological states or individual differences based on the lead internal graph structure and the 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 an abnormal heart rhythm classification result. The invention can model the local waveform morphological characteristics, transconductance association characteristics and distribution change caused by crowd crossing difference of the electrocardiosignal at the same time, thereby effectively improving the accuracy and robustness of the electrocardiosignal abnormal classification task and the generalization capability of a crossing subject.

Inventors

WANG HAISHUAI
LIU XIAO
SHEN QIANQIAN
GAO YANG
BU JIAJUN

Assignees

浙江大学

Dates

Publication Date: 20260512
Application Date: 20260108

Claims (6)

1. An electrocardiographic signal anomaly classification method is characterized by comprising the following steps: s1, receiving a multi-lead electrocardiosignal data set acquired by multi-lead electrocardiosignal acquisition equipment; s2, constructing a lead internal graph structure for representing internal rhythm characteristics of the electrocardiographic waveform according to each lead signal sequence, wherein the lead internal graph structure represents waveform and rhythm information inside a single lead electrocardiographic signal; S3, constructing an inter-lead external graph structure for representing the correlation between leads based on waveform synchronism and morphological correlation between the lead signals; S4, extracting an external distribution mode and a global graph representation of an electrocardiosignal based on the lead internal graph structure and the lead external graph structure, wherein the external distribution mode is used for distinguishing electrocardiosignal distribution differences caused by different subjects, different physiological states, different autonomic nerve excitation levels or different concentric muscle conduction conditions, and the extraction of the external distribution mode and the global graph representation of the electrocardiosignal comprises the following steps of; s41, constructing a neighborhood subgraph aiming at each node, and regarding each neighborhood subgraph as independent distribution; S42, information aggregation is carried out on each neighborhood subgraph based on the graph neural network, and global graph characteristics are obtained; S43, performing maximum pooling processing on the global graph features to enhance feature expression; s5, classifying distribution existing in the electrocardiosignals based on the external distribution mode of the electrocardiosignals and the global graph representation, wherein the classification comprises the following steps: s51, performing domain identification on the global map representation by using a domain classifier to obtain a domain identification result; s52, using the domain identification result as a reference item of parameter optimization, and updating a part of modeling parameters of the lead internal graph structure and the lead external graph structure; S6, inputting the global graph representation into a classifier to obtain an electrocardiographic anomaly classification result, wherein the method comprises the following steps: S61, classifying the global map representation by using a classification model to finally classify the electrocardiographic abnormal types; S62, using the abnormal classification result and a reference item serving as parameter optimization to update a part of modeling parameters of the lead internal graph structure and the lead external graph structure; S63, combining the abnormal classification result and the distribution classification result, and updating modeling parameters of the lead internal graph structure and the lead external graph structure.
2. The method of claim 1, wherein constructing the lead internal map structure of step S2 comprises: s21, dividing the electrocardiosignal of each lead into a plurality of waveform segments according to a fixed time window or a heart beat period, wherein each segment is used as a node in the graph; s22, calculating the similarity of waveform cosine between the target segment m and other segments n; s23, selecting a specified number of fragments with highest similarity as neighbor nodes based on the cosine similarity; S24, constructing a segment-level adjacency matrix based on the target segment node and the neighbor node; s25, aggregating the characteristics among the nodes in the plurality of segment-level adjacency matrixes by using a graph neural network to obtain the lead internal graph structure.
3. The method according to claim 1, wherein the constructing the inter-lead external graph structure in step S3 includes: S31, determining a target lead in a plurality of leads; s32, calculating waveform form cosine similarity between the target lead and other leads; S33, sequentially taking all leads as target leads to generate an inter-lead similarity matrix; S34, based on the similarity matrix, using a specified number of leads with highest similarity as neighbor nodes; s35, adding a position embedding feature for each lead node in the global map; s36, fusing the feature vector into the global graph based on the graph Laplace calculation, and carrying out information aggregation among nodes by utilizing the graph neural network to obtain the node features of the external graph among the leads.
4. An electrocardiographic signal anomaly classification device, characterized by comprising: A receiving unit for receiving a multi-lead electrocardiosignal data set acquired by multi-lead electrocardiosignal acquisition equipment; the lead internal structure construction unit is used for constructing a lead internal graph structure for representing the internal rhythm change characteristics of the electrocardiographic waveform aiming at each lead signal sequence in the multi-lead electrocardiographic signal data set; The inter-lead structure construction unit is used for constructing an inter-lead external graph structure for representing the association relationship between leads based on waveform synchronism and morphological correlation between the electrocardiosignals of the leads; An external distribution pattern extraction unit for extracting an external distribution pattern for distinguishing electrocardiographic waveform differences caused by different subjects, different physiological states, different muscle conduction conditions, or different autonomic nerve excitation levels based on the lead internal map structure and the inter-lead external map structure, the external distribution pattern extraction unit performing the following steps: constructing a neighborhood subgraph for each node and treating the neighborhood subgraph as an independent distribution domain; Information aggregation is carried out on each neighborhood subgraph based on the graph neural network, and node domain characteristics are obtained; performing maximum pooling treatment on the aggregated node domain features to enhance feature expression and obtain the external distribution mode; the global map representation generating unit is used for generating global map representations of electrocardiosignals based on the lead internal map structure, the inter-lead external map structure and the external distribution mode; The distribution classification unit is used for identifying distribution differences inside the electrocardiosignals based on domain classification results corresponding to the external distribution modes and generating distribution classification output for updating model parameters, and comprises the following steps: s51, performing domain identification on the global map representation by using a domain classifier to obtain a domain identification result; s52, using the domain identification result as a reference item of parameter optimization, and updating a part of modeling parameters of the lead internal graph structure and the lead external graph structure; the classification unit is used for inputting the global graph representation into an electrocardiographic anomaly classifier to obtain electrocardiographic anomaly classification results, and comprises the following steps: S61, classifying the global map representation by using a classification model to finally classify the electrocardiographic abnormal types; S62, using the abnormal classification result and a reference item serving as parameter optimization to update a part of modeling parameters of the lead internal graph structure and the lead external graph structure; S63, combining the abnormal classification result and the distribution classification result, and updating modeling parameters of the lead internal graph structure and the lead external graph structure.
5. An electrocardiographic signal anomaly classification device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the electrocardiographic signal anomaly classification method of any one of claims 1-3 when the processor executes the computer program.
6. A computer readable storage medium having stored thereon a computer program for implementing the electrocardiographic signal anomaly classification method of any one of claims 1 to 3 when executed by a processor.

Description

Electrocardiogram signal abnormality classification method and device Technical Field The invention relates to the technical field of intelligent analysis of medical signals, in particular to a graph structure characterization method and device for classifying heart rhythm abnormalities of Electrocardiosignals (ECG). Background Electrocardiograph (ECG) is an important physiological signal reflecting the electrical activity of the heart, and multi-lead electrocardiographs have been widely used for clinical arrhythmia screening and diagnosis. Common arrhythmias such as atrial premature beat, ventricular premature beat, atrial fibrillation, etc. have significant differences in waveform morphology, periodic rhythms, and transconductor potential distribution. Therefore, how to accurately identify arrhythmia from multi-lead electrocardiosignals is an important research direction for auxiliary diagnosis of cardiovascular diseases. The existing electrocardiosignal anomaly classification method is mainly based on a convolutional neural network, a recurrent neural network or a transducer and other depth models, and is used for directly carrying out end-to-end characterization on electrocardiosignals. However, such methods typically only focus on waveform characteristics within a single lead, ignoring physiological associations between different leads, such as the spatial distribution of electrocardiographic potentials across the body surface. In addition, individual differences, physiological state changes, differences in acquisition devices, and the like of different subjects can lead to inconsistent electrocardiographic waveform Distribution, i.e., there is a significant off-of-Distribution (OOD) shift. The traditional method is difficult to maintain stable classification performance under the conditions of cross-crowd and cross-scene, and has insufficient generalization capability. Therefore, the prior art is not provided with an electrocardio anomaly classification method capable of simultaneously modeling an internal periodic morphology structure of leads and a cross-space correlation structure among leads and explicitly modeling a cross-individual distribution difference so as to improve the robust classification capability of multi-lead electrocardiosignals. Disclosure of Invention The invention aims to overcome the defects of the prior art and provides an electrocardiosignal anomaly classification method and device. The first aspect of the invention provides an electrocardiosignal abnormality classification method, which comprises the following steps: the method comprises the steps of acquiring a multi-lead electrocardiosignal data set by multi-lead electrocardiosignal acquisition equipment, determining a lead internal graph structure for each lead signal in the multi-lead electrocardiosignal data set, wherein each lead internal graph structure is characterized by waveform and rhythm information inside a single lead electrocardiosignal, determining an inter-lead external graph structure for the multi-lead electrocardiosignal data set, wherein the inter-lead external graph structure is characterized by the synchronicity and morphological relativity of the electrocardiosignal between a plurality of leads, determining an external distribution mode of the electrocardiosignal data set based on the lead internal graph structure and the inter-lead external graph structure, wherein the external distribution mode is used for distinguishing the differences of the electrocardiosignal shape distribution caused by different physiological states, different levels of autonomic nerve excitation or different myocardial conduction conditions, generating graph characterization of the multi-lead electrocardiosignal data set based on the lead internal graph structure, the inter-lead external graph structure and the external distribution mode, and inputting the graph characterization of the multi-lead electrocardiosignal data set into an atrial fibrillation classifier, and obtaining an abnormal atrial fibrillation result, such as an atrial fibrillation abnormal atrial fibrillation classifier, an atrial fibrillation device, or other abnormal atrial beat type, or an atrial beat abnormal atrial beat classification. Wherein, the constructing the lead internal graph structure in step S2 includes: s21, dividing the electrocardiosignal of each lead into a plurality of waveform segments according to a fixed time window or a heart beat period, wherein each segment is used as a node in the graph; s22, calculating the similarity of waveform cosine between the target segment m and other segments n; s23, selecting a specified number of fragments with highest similarity as neighbor nodes based on the cosine similarity; S24, constructing a segment-level adjacency matrix based on the target segment node and the neighbor node; s25, aggregating the characteristics among the nodes in the plurality of segment-level adjacency matrixes by using a graph