Cardiac stroke is among the most common causes of death globally and accounts for more than 6.5 million deaths every year. Early prediction is important in minimizing mortality and enhancing patient prognosis. This paper introduces an explainable deep learning model that leverages multi-modal clinical data—electronic health records, laboratory results, and imaging—to provide accurate and interpretable cardiac stroke risk prediction. The model achieved 94.2% accuracy, 91.8% precision, and AUC-ROC of 0.96 on a real-world dataset of 20,000 patient records. Feature attribution methods like SHAP and Grad-CAM provided interpretable and actionable insights to clinicians, thereby building trust and interpretability. The suggested system is superior to current models by over 12%, demonstrating its ability in enabling early clinical intervention through data-driven, interpretable decisions.