Abstract
Coronary artery disease (CAD) is a prevalent cardiovascular condition and a leading cause of mortality. An accurate and timely diagnosis of CAD is crucial for treatment. In this study, we aimed to develop a novel stenosis detection algorithm using the YOLOv9c model to automate the detection of CAD for real-time applications. The dataset consisted of angiographic imaging series obtained from 100 patients with confirmed one-vessel CAD, comprising a total of 8,325 grayscale images. The YOLOv9c model was trained, tested, and validated using the Python API for YOLO and the ultralytics library, with fine-tuning and augmentations applied to improve detection accuracy. By automating the detection of multivessel disease, the proposed algorithm has the potential to enhance the workflow of operators. The proposed YOLOv9c model demonstrated superior performance in processing speed and detection accuracy, achieving an F1-score of 0.98 and an mAP@0.5 of 0.98, outperforming established models. The model had a weight of 25.3M, significantly lower than others, leading to reduced training time (11 hrs), fine-tuning time (3.5 hrs) and inference time (18 ms). Additionally, compared with SSD MobileNet V1, F1-score and mAP@0.5 improved by 1.36x and 1.42x, respectively. The proposed stenosis detection algorithm represents a significant advancement in the field of cardiovascular imaging and diagnostic algorithms. The integration of advanced algorithms in cardiovascular imaging represents a critical aspect of optimizing diagnostic efficiency and accuracy, emphasizing the need for ongoing advancements in medical imaging research and technology.