Semi-supervised GAN for enhancing electrocardiogram time series diagnostics

Dawei Li

doi:10.1016/j.bspc.2025.108058

Semi-supervised GAN for enhancing electrocardiogram time series diagnostics

2025-06-06

0

PRE

AI

DOI：10.1016/j.bspc.2025.108058

Original

Original request for help

Save

Abstract

En 中文

Despite the remarkable progress made by Deep Neural Networks (DNN) in biomedical areas, challenges such as data scarcity and unbalanced samples persist. In this paper, we propose a series generation network that combines a self-coding model with a Generative Adversarial Network (GAN). This innovative approach addresses the limitation of GAN networks in overlooking time dynamic features. Our proposed network aims to achieve two main objectives: (a) generate high-quality synthetic data for state-of-the-art neural networks using a small dataset, and (b) enhance the performance of the ECG diagnosis model when trained with synthetic data. The dimension reduction analyses are used to visualize the comparison between the synthesized and real data, and the residual neural network (ResNet) is used to verify whether the synthesized data can improve the classification accuracy of neural network. Experimental results show that, compared with the real data model, the difference between sensitivity and specificity of the model using synthetic data are decreased by 4% on average. The model accuracy of synthesized data has improved to 97.2%, which has increased by 2% compared with similar works. This work is beneficial to tackling the data scarcity and unbalance and can even improve the diagnostic accuracy of ECG dataset classification, making it a potential solution for practical applications.

Semi-supervised GAN for enhancing electrocardiogram time series diagnostics

Abstract

Journal

Organization