Cardiovascular diseases are among the leading causes of death worldwide, demanding accurate, timely, and automated diagnostic solutions. Despite advancements in medical imaging, early detection of heart conditions remains challenging due to the complex spatiotemporal patterns present in cardiac data. This research aims to develop a real-time deep learning framework for continuous and precise heart disease detection, addressing limitations of conventional models in handling temporal dependencies and feature complexity. A total of 15,000 cardiac images were collected, including 8,000 MRI and 7,000 CT scans, from publicly available datasets such as MIMIC-CXR and Cardiac MRI Dataset, complemented by hospital acquired images to ensure diversity in age, gender, and pathological conditions. Collected images were preprocessed using normalization, noise reduction, contrast enhancement, and region-of-interest segmentation, ensuring optimized feature extraction and consistent input quality for model training. The proposed model, HRAE-LSTM (Hybrid Residual Attention-Enhanced Long Short-Term Memory), incorporates residual connections and attention mechanisms to capture both temporal and spatial cardiac features, enabling early detection of myocardial infarction, arrhythmia, and cardiomyopathy. Comparative analysis with conventional LSTM, CNN, and RNN models demonstrated that the proposed framework achieves superior accuracy, sensitivity, and specificity, while providing real-time inference and continuous monitoring. The study establishes a clinically viable, automated approach for heart disease detection, with significant potential for integration into hospital imaging workflows and telemedicine platforms.