Detection of CAD using XCA images(CADICA)

Coronary artery disease (CAD) is a leading cause of fatality worldwide. Early detection of CAD is crucial for timely diagnosis and prevention of severe health risks associated with it. This work focuses on detecting coronary artery disease (CAD) through X-ray coronary angiography (XCA) images. A framework for automated lesion detection using the YOLOv5 deep learning model was implemented. The methodology began with acquiring a dataset, followed by its preparation and organization for effective training and validation. To optimize the performance, transfer learning was employed with pre-trained YOLOv5s weights. YOLOv5 is an object detection model known for its exceptional accuracy and speed. It can identify and localize multiple objects in images and videos, making it suitable for various real-time applications. The model was trained for 100 epochs, during which the mean average precision (mAP) was monitored to track progress. Once training was completed, the model was evaluated on new or unseen test data, enabling it to predict lesions of CAD with bounding boxes drawn around the detected areas with class labels and confidence scores. The model achieved a overall accuracy of 79.