A Real-Time Method for Marking the Extent of a Lipid Plaque Based on IV-OCT Imaging

Abstract

Atherosclerotic plaques, the leading cause of heart attack, can be characterized from intravascular optical coherence tomography (IV-OCT) images by doctors. Since lipid accumulation is an important indication of atherosclerotic plaque, we introduced a new convolutional neural network, called Single Shot Plaque Marking Network (SSPM), to develop an automated method that highlights the extent of lipid plaques from IV-OCT images at real-time, which then would help doctors easily find the vulnerable plaque. Compared with previous available methods, our method is capable of marking the suspicious lipid plaque areas in real-time with better time-efficiency and competitive accuracy during the diagnosis. SSPM is tested on IV-OCT human coronary artery imaging dataset, and the result shows that our method is able to mark suspicious lipid-plaque areas at 91 fps on GPU, or 16 fps on CPU, with an accuracy of 87%.