Automated Classification of Calcification and Stent on Computed Tomography Coronary Angiography Using Deep Learning

Abstract

In computed tomography coronary angiography (CTCA), calcification and stent make it difficult to evaluate intravascular lumen. This is a cause of low positive-predictive value of coronary stenosis. Therefore, it is expected to develop a computer-aided diagnosis (CAD) system that can automatically detect stenosis in coronary arteries. The purpose of this study is to automatically recognize calcifications or stents in coronary arteries and classify them from the normal coronary artery in CTCA. We used 4960 coronary-cross-sectional images, which consisted of 1113 images with calcification, 1353 images with a stent, and 2494 normal artery images. These images were automatically classified using the deep convolutional neural network (LeNet, AlexNet, and GoogLeNet). The classification accuracy of LeNet, AlexNet, and GoogLeNet were 58.4%, 75.9%, and 81.3%, respectively. The proposed method would be a fundamental technique of CAD in CTCA.