A novel anchored GAN approach separating temporal and amplitude variations improves synthetic ECG generation and downstream classification.
Electrocardiogram (ECG) synthesis is a challenging task due to the complex dynamic nature of ECG signals. In this paper, we present a novel approach for ECG synthesis based on Generative Adversarial Networks. We study how to incorporate prior knowledge on morphological ECG patterns into synthetic data generation. We demonstrate how to improve the modeling of the complex dynamics of ECG signals by separating between modeling temporal and amplitude variations. To evaluate our proposal, we consider ECG signals taken from the MIT-BIH arrhythmia database for training our models. The obtained experimental results demonstrate the efficacy of our approach for generating synthetic ECG signals and improving ECG classification.
Neifar et al. (Mon,) studied this question.