Deep Learning based Coronary Stent Detection and Residual Stenosis Assessment in Optical Coherence Tomography

Intravascular optical coherence tomography (OCT) is an optical-based imaging modality with high axial resolution that enables in vivo intra-vascular assessment of the coronary arteries and of the stent implantation results. Herein, we introduce and evaluate a deep neural network based methodology for stent detection. A binary classification is performed for each slice (stent / no stent). Additionally, we perform coronary lumen segmentation to pre- and post-process the images. Stents were detected with a precision of 96.4% and an F1-score of 94%. At slice level the detection had an accuracy of 94%. No bias was observed in the detection of the start and end locations of the stent, while the mean absolute error was 1.05 and 1.66 mm, respectively. Furthermore, the proposed approach can accurately assess the residual stenosis, which is a clinically important measure, associated with an increased risk for future revascularization requirement. An important advantage of the proposed method is that strut detection is no longer required: annotations were performed by marking only the start and the end OCT slice of the stent.