基于深度学习的心电信号生成技术研究  

Study in generating electrocardiogram signals by deep learning

作　　者：林铭俊 肖中举[2] 肖慧 张鑫 洪永 陈超敏[1] 郑东宏 Lin Mingjun;Xiao Zhongju;Xiao Hui;Zhang Xin;Hong Yong;Chen Chaomin;Zheng Donghong(School of Biomedical Engineering,Southern Medical University,Guangzhou 510515,China;School of Basic Medical Sciences,Southern Medical University,Guangzhou 510515,China;Department of Equipment,Heyuan People’s Hospital,Heyuan 517000,China)

机构地区：[1]南方医科大学,生物医学工程学院,广州510515 [2]南方医科大学,基础医学院,广州510515 [3]河源市人民医院设备科,河源517000

出　　处：《现代仪器与医疗》2023年第5期44-48,共5页Modern Instruments & Medical Treatment

基　　金：广州市科技项目(2023B03J1337)。

摘　　要：随着人工智能技术的快速发展,运用心电图自动诊断模型辅助临床诊断成为一大研究热点,但仍存在着心电医疗数据难以获取、样本不平衡、数据隐私限制与泄露等问题。而大型免费医疗数据集的出现以及深度生成模型的发展,使得生成高质量的心电数据成为解决上述问题的有效方法。本文介绍了三种深度生成模型,分别为变分自编码器、生成对抗网络和去噪概率扩散模型,对其原理与特点进行了归纳和总结,然后将深度生成模型在心电信号生成领域中的应用进行了系统的分析研究,最后讨论了人工智能技术在心电信号生成领域的发展前景与趋势。With the rapid development of Artificial intelligence,it is of great value to attach importance to the research and development of electrocardiogram(ECG)automatic diagnosis.But it is still limited by such as difficulty in collecting ECG medical data,imbalanced training data,data privacy restrictions and disclosure.At the same time,the well-developed deep generative model and large-scale publicly available open-access medical datasets makes generating high-quality synthetic ECG has become a solution to alleviate the challenges above.In this paper,we first introduce the mechanism of three typical algorithms:variational autoencoder(VAE),generative adversarial networks(GANs)and Denoising diffusion probabilistic model(DDPM).Then we review their application in ECG generation.Our view about future potential development of deep generative model in ECG generation is stated in the final part of this paper.

关 键 词：深度生成模型 心电信号生成 变分自编码器 对抗生成网络 去噪概率扩散模型 

分 类 号：TH776[机械工程—仪器科学与技术] R318[机械工程—精密仪器及机械]