Coronary artery stenosis detection algorithm based on improved yolov5

Cardiovascular disease caused by coronary artery stenosis has high morbidity and mortality, and has become an important factor threatening human life and health. Therefore, early diagnosis of coronary artery stenosis is particularly critical. In actual medical diagnosis, manual detection of vascular stenosis is inefficient. Screening vascular stenosis by deep learning algorithm is one of the effective methods. However, in the field of medical image processing, it is difficult to obtain publicly available coronary artery CTA image datasets due to the protection of patient privacy. To this end, this paper establishes a available coronary artery CTA dataset and annotates it under the guidance of doctors. Coronary artery edge features affect the judgment of stenosis. In view of the problem that YOLOv5 is easy to ignore the details of vessel edge contours in CTA images, the MFFN module is used to replace the SPP module in the YOLOv5 feature extraction layer to achieve deep and shallow feature fusion; the Neck layer adds PAM-ECA-N attention force mechanism to improve the target detection effect. Experiments show that the improved YOLOv5-MFFN-PAM-ECA-N model has increased by 2.92% on F1-score and 1.64% on mAP@.5, which proves that this improvement can improve model performance.