Cardiovascular diseases (CVDs) remain the leading cause of global mortality, highlighting the urgent need for accurate and early predictive systems. This paper presents a systematic review and comparative analysis of recent deep learning and hybrid approaches applied to CVD prediction. Following PRISMA guidelines, 28 relevant studies published between 2020 and 2025 were selected across multiple databases. The findings show a growing trend toward integrating advanced architectures such as transformers, convolutional recurrent networks, and graph neural networks as well as hybrid frameworks combining deep learning with feature selection, fuzzy logic, or machine learning techniques. These methods offer improved diagnostic accuracy and enhanced interpretability, especially when applied to multimodal and heterogeneous data sources. However, challenges remain, including lack of external validation, limited dataset diversity, and the need for explainable and generalizable models. This paper concludes with key insights on current limitations and outlines future research directions aimed at enhancing the clinical relevance and scalability of AI-based CVD prediction systems.
Baccouch et al. (Thu,) studied this question.