Stenosis Detection in X-ray Coronary Angiography with Deep Neural Networks Leveraged by Attention Mechanisms

Abstract

Coronary artery disease (CAD) is one of the most prevalent causes of death worldwide. The automatic detection of coronary artery stenosis on X-ray images is important in coronary heart disease diagnosis. Coronary artery disease is caused by atherosclerotic plaques with subsequent stenosis (e.g. narrowing) of the coronary arteries. This makes the heart work harder, risking failure. Automated identification of stenosis may be used for triage or as a second reader in clinical practice, providing a valuable tool for cardiologists. In this paper, we evaluate the detection of stenosis in X-ray coronary angiography images with novel object detection methods based on deep neural networks. We trained and tested three promising object detectors based on different neural network architectures leveraging attention mechanisms (EfficientDet, RetinaNet ResNet-50- FPN, and Faster R-CNN ResNet-101) using clinical angiography data of 438 patients. The metrics obtained on this dataset, have shown an advantage of EfficientDet over alternative approaches, achieving a mean average precision of 0.67 in the task of detecting stenosis in X-Ray angiographies. This result provides evidence that attention mechanisms improve the performance of convolutional neural networks in a medical imaging context.