Rapid lipid-laden plaque identification in intravascular optical coherence tomography imaging based on time-series deep learning

Significance: Coronary heart disease has the highest rate of death and morbidity in the Western world. Atherosclerosis is an asymptomatic condition that is considered the primary cause of cardiovascular diseases. The accumulation of low-density lipoprotein triggers an inflammatory process in focal areas of arteries, which leads to the formation of plaques. Lipid-laden plaques containing a necrotic core may eventually rupture, causing heart attack and stroke. Lately, intravascular optical coherence tomography (IV-OCT) imaging has been used for plaque assessment. The interpretation of the IV-OCT images is performed visually, which is burdensome and requires highly trained physicians for accurate plaque identification. Aim: Our study aims to provide high throughput lipid-laden plaque identification that can assist in vivo imaging by offering faster screening and guided decision making during percutaneous coronary interventions. Approach: An A-line-wise classification methodology based on time-series deep learning is presented to fulfill this aim. The classifier was trained and validated with a database consisting of IV-OCT images of 98 artery sections. A trained physician with expertise in the analysis of IV-OCT imaging provided the visual evaluation of the database that was used as ground truth for training and validation. Results: This method showed an accuracy, sensitivity, and specificity of 89.6%, 83.6%, and 91.1%, respectively. This deep learning methodology has the potential to increase the speed of lipid-laden plaques identification to provide a high throughput of more than 100 B-scans/s. Conclusions: These encouraging results suggest that this method will allow for high throughput video-rate atherosclerotic plaque assessment through automated tissue characterization for in vivo imaging by providing faster screening to assist in guided decision making during percutaneous coronary interventions. (C) The Authors. Published by SPIE under a Creative Commons Attribution 4.0 International License.