Coronary artery stenosis detection

Customer: Kemerovo Cardiology Center (Kemerovo, Russia)

Summary: Invasive coronary angiography is the gold standard for diagnosing coronary artery disease, but it can be complicated by patient-specific anatomy and image quality. Deep learning techniques that detect coronary artery stenoses can facilitate diagnosis, but previous studies have not achieved superior accuracy and performance for real-time labeling. This study aims to confirm the feasibility of real-time coronary artery stenosis detection using deep learning methods. To achieve this goal, eight promising detectors were trained and tested using clinical angiography data from 100 patients. These detectors were based on different neural network architectures, such as MobileNet, ResNet-50, ResNet-101, Inception ResNet, and NASNet. Three neural networks showed superior results. The network based on Faster RCNN Inception ResNet V2 was the most accurate, achieving a mean average precision of 0.95, an F1 score of 0.96, and a prediction rate of 3 fps on the validation subset. The relatively lightweight SSD MobileNet V2 network was the fastest, with a low mAP of 0.83, an F1 score of 0.80, and a prediction rate of 38 fps. The RFCN ResNet-101 V2 based model had an optimal accuracy-to-speed ratio with an mAP of 0.94, an F1 score of 0.96, and a prediction speed of 10 fps. The balance of performance and accuracy demonstrated by these advanced neural networks confirms the feasibility of real-time coronary artery stenosis detection and supports the decision-making process of the cardiac team interpreting coronary angiography findings.

Collaborators: Alejandro Frangi (University of Leeds, Leeds, United Kingdom), Evgeny Ovcharenko (Research Institute for Complex Issues of Cardiovascular Diseases, Kemerovo, Russia), Kirill Klyshnikov (Research Institute for Complex Issues of Cardiovascular Diseases, Kemerovo, Russia)

Project type: Commercial / Research

Media: Journal paper, Model testing during surgery, News 1 (rus), News 2 (rus), News 3 (rus)