Abstract 16139: Artificial Intelligence to Identify Acute Coronary Syndroms: Detection of Vulnerable Coronary Plaques in OCT Images With Deep Learning.

Introduction: Optical coherence tomography (OCT) is an intravascular imaging modality which is widely used clinically for coronary plaque characterization. It has the ability to measure the fibrous cap thickness and to detect lipid content. Therefore, it can be used to identify in vivo thin-cap fibroatheroma(TCFA) and to detect plaque vulnerability. However, it takes much time for a doctor to learn to understand OCT images. We proposed a novel method of detecting vulnerable coronary plaques in OCT images of coronary artery disease (CAD) patients with deep learning. Methods and Results: The procedure includes: 1) preprocessing of the original images including coordinates transformation, image denoising and data augmentation; 2) deep learning network training on a dataset of 2000 images including classification, object detection and segmentation; 3) postprocessing of the detected images including image filtering and expert knowledge based adjustment. The model was tested on a separate series of 300 coronary OCT images. The results indicate that applying the image detecting method provided by this paper, the accuracy rate was 85.0% while the sensitivity (recall rate) was 93%. Conclusions: Our artificial intelligence method can help improve clinical decision-making and reduce human medical errors of Cardiologists, thus help provide a more individualized treatment for each CAD patients. [ABSTRACT FROM AUTHOR]