Coronary arteries stenosis detection by deep learning methods

Coronary heart disease is a dangerous heart disease caused by coronary arteries. In clinical practice, X–ray coronary angiography is the main method of visualisation for diagnostics of coronary diseases. High cost and complexity of its results analysis by a cardiac surgeon made it necessary to automate the process of image processing and diagnostics of coronary artery stenoses. In this work we considered the models of deep detection, localisation and stenosis characterisation using popular models such as SSD, R–FCN, Faster–RCNN, RetinaNet, EfficientDet. Computational experiments on stenosis detection from X–ray images were performed on the coronary angiography data. A comparative analysis of the models in terms of the main performance indicators: mAP accuracy, image processing time, number of model parameters. The obtained results allow us to state that Faster R–CNN (ResNet101) and EfficientDet D4 (ResNet101) models are the detectors of choice in the detection of coronary artery stenosis. They have high detection accuracy and image processing speed compared to other models, as well as relatively low parameter values. Even though the speed of X–ray arterial images processing by both models does not exceed real time, their reliability is high enough to minimise the risk of false detections of coronary artery stenosis. A comparative analysis of their characteristics with the results of other researchers showed superior or comparable results obtained in this work.