Abstract
Coronary artery disease (CAD) remains one of the most prevalent causes of death worldwide, demanding accurate and timely diagnosis to guide treatment. Coronary angiography is the clinical reference standard for visualizing arterial patency; however, its manual interpretation is labour-intensive and subject to considerable inter-observer variability. This study introduces a hybrid deep learning framework that fuses convolutional neural networks with long short -term memory units to exploit both spatial and temporal cues in cine angiography. After automated frame extraction, selection and normalization, sequences are processed end-to-end to detect the presence and type of clinically significant coronary pathology. Validation on an expert-labeled video dataset demonstrates robust performance (accuracy=0.88; precision=0.85; recall=0.92; F1-score=0.88; AUC≈0.91). Gradient-weighted class activation mapping highlights lesion-relevant vessel segments, supporting model interpretability. These findings confirm that the proposed CNN-LSTM architecture can reduce observer variability and accelerate decision-making by delivering consistent, automated assessment of angiographic studies.