Automated Digital Biomarker Discovery Pipeline for Cardiovascular Diseases

Cardiovascular Diseases (CVDs) are the leading cause of mortality worldwide, necessitating early and accurate diagnosis to prevent severe outcomes such as Heart Failure (HF). Despite the widespread use of Electrocardiogram (ECG) for cardiac monitoring, traditional methods often miss subtle preclinical changes. In this paper, we present an automated digital biomarker discovery pipeline that leverages explainable artificial intelligence (XAI) to enhance the interpretability and clinical applicability of ECG-based biomarkers for CVDs. Using an inter-pretable feature extractor combined with unsupervised clustering and Particle Swarm Optimisation (PSO), our method identifies both known and novel ECG features associated with high CVD risk. These include established markers like RR Interval Sample Entropy and the discovery of novel biomarkers such as T-Wave Multiscale Entropy, which we found to be significantly associated with CVD risk. Our pipeline enhances early detection by bridging Artificial Intelligence (AI) methods with clinical relevance, providing interpretable insights that align with physiological principles. This transparency promotes clinician trust and supports the integration of AI into routine medical practice. Our results demonstrate that this approach can significantly improve the prediction and understanding of heart diseases, thus offering a powerful tool for reducing the global burden of CVDs.