Accurate ECG Data Generation with a Simple Generative Adversarial Network. (September 2020)

• Record Type: Journal Article
• Authors: Wang, Haiyan; Ge, Zhaoyang; Wang, Zongmin
• Is Part Of: Journal of physics. Volume 1631(2020)
• Publication Date: 2020-09
• Subjects: ECG data generation -- heartbeat -- left bundle branch block -- deep learning model -- GAN; Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1631/1/012073
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: Since ECG data is highly sensitive medical data, the acquisition of ECG data is highly restricted. However, with the increasing demand for ECG big data research, the improvement of computer computing capabilities, and the development of deep learning, the direction of ECG intelligent analysis is facing a serious lack of standard clinical data. In order to generate more precise ECG data, this paper proposes a GAN architecture for generating ECG heartbeat data. The network structure is simple and does not require any domain knowledge. In this paper, the MIT-BIH Arrhythmia database is selected, from which all left bundle branch block heartbeats are selected to form a training dataset. The training process shows that the proposed GAN structure is effective and accurate, and the generated results show that not only the generated simulated ECG heartbeat data is diverse, but also highly similar to the real data.