An Ecg Signal Synthesis Approach Based on the Synergy of Wgan and Attention Mechanisms

作     者：Zhou, Hongfang Li, Yijie Guo, Yifan Xue, Haolin 

作者机构：School of Computer Science and Engineering Xi’an University of Technology Xi’an710048 China Shaanxi Key Laboratory of Network Computing and Security Technology Xi’an710048 China 

出 版 物：《SSRN》 

年 卷 期：2024年

核心收录：

主　　题：Generative adversarial networks 

摘      要：Electrocardiogram (ECG) synthesis plays a key role in the diagnosis of heart diseases. However, ECG data involves patient privacy and has obvious imbalance issue. This leads to a disconnect between theoretical research methods and practical applications. Based on this, the paper will explore the application of generative adversarial networks (GANs) in ECG signal synthesis. First, the paper proposes a WGAN-SD-GP model that uses the conditional WGAN networks incorporating an attention mechanism to enhance the ability to capture contextual information and effectively focus contextual information by scaling dot products. Secondly, the paper explores whether the design concepts of circular convolution, feature extraction and layer-hopping connection in R2U-Net can be applied to the synthesis of ECGs, and proposes the WGAN-SKD-GP model. The model combines the hopping layer connection with the attention mechanism, which enables it to process the input data more flexibly and enhances the feature expression ability. Finally, the experiments were designed to automatically classify arrhythmia using 1DResNet50 on a mixed dataset of synthetic and real signals, which was evaluated using several metrics, including Precision, Recall, F1 Score, Macro Avg, Weighted Avg, and Overall Accuracy. The experimental results not only verify the effectiveness of WGAN-SKD-GP in ECG signal synthesis, but also provide a new idea for solving the problem of insufficient data of ECG signals, which has significant potential application value. © 2024, The Authors. All rights reserved.