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

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.