Coronary Artery Disease (CAD) is one of the leading causes of global mortality, and its diagnosis primarily relies on the visual interpretation of angiographic images by specialists—a process that is both time-consuming and prone to human error. This study aims to develop an automated and efficient diagnostic framework based on deep learning for precise localization of vascular stenosis. In this study, the performance of two advanced object detection architectures, namely RetinaNet and EfficientDet-D3, was evaluated. The models were trained and validated using a dataset comprising 8,325 angiographic images obtained from 100 patients diagnosed with single-vessel CAD. To enhance generalizability, image preprocessing and data augmentation techniques were applied. Experimental results demonstrated that the EfficientDet-D3 model achieved a superior performance, with a mean Average Precision (mAP) of 96.6%. The RetinaNet model also exhibited strong robustness, achieving an mAP of 93.2% while maintaining high processing speed. The proposed framework, by achieving a balance between accuracy and computational efficiency, demonstrates strong potential for deployment in clinical settings as a decision support system. It may facilitate timely and accurate diagnosis by reducing reliance on manual interpretation.