P. Murali, S. Meenatchi
ABSTRACT Heart disease remains the leading cause of morbidity and mortality worldwide, highlighting the urgent need for accurate and timely diagnostic methods. Recent advances in machine learning (ML), deep learning (DL), and federated learning (FL) have enabled automated heart disease diagnosis by leveraging diverse data modalities, including ECG signals, cardiac imaging, and electronic health records (EHR). However, the high dimensionality, multipartite nature, and variability of these datasets present challenges in model generalization, often introducing biases that reduce robustness. This study presents a systematic literature review (SLR) of heart disease prediction research published between 2015 and 2025, initially screening 550 references and narrowing to 72 studies based on strict inclusion and exclusion criteria. The review categorizes studies by evidence type and analyzes trends in predictive methods, contrasting single‐modality approaches (ECG or imaging) with multimodal techniques that integrate multiple data sources. Key research gaps identified include the need for improved data fusion strategies, augmentation methods, privacy‐preserving models, and explainable AI. Building on these insights, the study proposes a diagnostic framework that integrates ECG and imaging data using CNN and BiLSTM for feature extraction, enhanced with Grad‐CAM and SHAP for model interpretability. The framework also incorporates diffusion models and transformer architectures to reconstruct missing data, enabling the combination of ECG, imaging, and EHR for improved predictive accuracy and robustness. Furthermore, federated learning combined with homomorphic encryption is explored to support secure, multi‐institutional deployment. Overall, this review synthesizes state‐of‐the‐art techniques, highlights critical gaps, and provides actionable strategies for developing explainable, secure, and multimodal AI‐based heart disease prediction systems, advancing both diagnostic performance and clinical trust.