Research on ST-T change recognition algorithm based on lead attention network
10.19745/j.1003-8868.2025117
- VernacularTitle:基于导联注意力网络的ST-T改变识别算法研究
- Author:
Liang WEI
1
;
Yun-chi LI
;
Jun XIE
;
Tong XU
;
Feng ZUO
;
Yong-qin LI
;
Bi-hua CHEN
;
Mi HE
;
Yu-shun GONG
Author Information
1. 陆军军医大学生物医学工程与影像医学系医学仪器与计量学教研室,重庆 400038
- Publication Type:Journal Article
- Keywords:
lead attention network;
ECG;
ST-T change;
deep learning;
12-lead ECG signal;
ECG monitoring
- From:
Chinese Medical Equipment Journal
2025;46(7):1-11
- CountryChina
- Language:Chinese
-
Abstract:
Objective To propose a lead attention network-based ST-T change recognition algorithm to detect ECG ST-T changes accurately.Methods Firstly,heartbeat signals were extracted through R-wave localization,and a 12-lead heartbeat matrix was generated by correlation-based screening and merging to realize data augmentation.Secondly,a lead attention module was constructed by combining depthwise convolution(DWConv)with the channel attention squeeze-and-excitation block(SE-block)structure to perceive the differences in ST-T status among electrocardiogram leads.Thirdly,the mapping output by two independent attention modules was fused and splicing with the original signal residual was carried out,so that attention information extraction and original information transfer were enhanced effectively.Finally,SE-ResNet was used as the backbone network to extract signal features to complete the classification and identification of ST-T changes.To validate the recognition performance of the proposed algorithm for ST-T changes in ECG,the 12-lead ECG data of 97 472 patients containing different ECG rhythms were collected for ablation and comparison experiments at the First Affiliated Hospital of Army Medical University.Results The proposed algorithm achieved an AUC of 0.965 with a sensitivity of 90.51%,specificity of 90.23%,positive predictive value of 89.24%and overall accuracy of 90.36%on an independent test set.Comparative analysis demonstrated superior performance to four benchmark architectures,including VGG16,ResNet18,MobileNetV3-Small and ShuffleNet,in terms of both classification accuracy and computational efficiency.Conclusion The algorithm designed can accurately detect ST-T changes and can be used for wearable ECG automatic analysis to assist in the early warning of cardiovascular diseases in both acute and chronic patients and highland residents.[Chinese Medical Equipment Journal,2025,46(7):1-11]
