Explainable AI-based system for automated detection of cardiovascular disease detection from ECG images: a theoretical review

Deep learning applied to electrocardiogram (ECG) images is changing cardiac diagnostics. But understanding how these systems work remains a major theoretical and practical challenge. This review focuses on explainable AI (XAI) models for automatic heart disease detection from ECG images, with attention to theoretical insights. A systematic thematic review was conducted using Scopus and Web of Science. The focus was on theoretical and quantitative aspects of XAI, ECG, and deep learning. Studies were grouped by model type, interpretation method, and theoretical contribution. The review found three trends: more use of saliency-based XAI methods such as Grad-CAM and SHAP; increasing use of CNN and transformer models for ECG analysis; and early development of theoretical frameworks assessing AI reliability. Few studies explored explainability in depth. A key challenge is linking model interpretability to clinical meaning. This review highlights the need for human-centered design, model castability, and cross-disciplinary theory to improve XAI use in cardiology.