Synthesis of Multimodal Cardiological Signals using a Conditional Wasserstein Generative Adversarial Network

Abstract

Cardiovascular diseases (CVDs) are the leading cause of mortality worldwide. Recent advancements in machine learning have significantly enhanced early detection and treatment strategies for CVDs. While electrocardiogram (ECG) signals are commonly used for detection, additional signals like arterial blood pressure (ABP) and central venous pressure (CVP) provide a comprehensive view of the cardiovascular system. However, acquiring such extensive datasets is challenging due to resource constraints, privacy issues, and ethical considerations. This paper introduces a novel Multichannel Conditional Wasserstein Generative Adversarial Network (MC-WGAN) capable of simultaneously generating synthetic ECG, ABP, and CVP signals. The MC-WGAN model addresses the data scarcity issue by providing high-fidelity synthetic data that mirrors real physiological signals, facilitating better simulation, diagnosis, and treatment planning. Evaluation against the MIT-BIH Arrhythmia Database demonstrated the model’s strong performance, with competitive metrics such as RMSE, PRD, and FD, particularly excelling in the generation of ECG and ABP signals. MC-WGAN surpasses other generative models by simultaneously replicating multiple physiological signals, offering a comprehensive view of cardiovascular health. This advancement enhances diagnostic accuracy and risk stratification, setting a new standard in synthetic biomedical signal generation, and paving the way for more personalized and effective clinical interventions.