Recent Trends in ECG-Data Based Solutions for CVD Forecast Using Explainable Artificial Intelligence: A Systematic Review

Abstract

The field of cardiovascular disease (CVD) prediction utilizing electrocardiogram (ECG) data experienced a substantial change, simply since Explainable Artificial Intelligence (XAI) has emerged. The crucial transition from conventional "black-box" deep learning models to transparent, clinician-centric frameworks is examined in this paper. Factor-ECG, which breaks down complex signals into interpretable latent factors, and Generative Counterfactual XAI (GCX), which offers "what-if" morphological representations, are two significant technological developments. Beyond qualitative heatmaps, objective auditing of model trustworthiness turned out to be enabled by adding a mix of quantifiable XAI metrics (qxAI).

With intermediate (feature-level) fusion tactics outperforming late fusion methods, multimodal integration—specifically, the fusion of ECG with medical imaging and wearable data—has shown improved prediction accuracy. With a growing physician acceptance rate of 66% as of 2024, the global AI in cardiology market is expected to reach USD 1.66 billion by 2031, according to market data. Despite these developments, issues with data privacy, regulatory compliance, and external validation continue to impede clinical application. Future paths are increasingly focused on Federated Learning to provide safe, multi-institutional model training and Digital Twins for customized diagnoses.