(A) weakly-supervised deep learning network for lipid detection in spectroscopic optical coherence tomography

Abstract

Optical coherence tomography (OCT) is an imaging technique that utilizes optical interference to generate high-resolution cross-sectional images. In the field of cardiology, it plays a crucial role in identifying vulnerable vessels and lipids, which are the core causes of acute coronary syndrome. While specific tissues can be distinguished through analyzing grayscale OCT images, there are limitations to relying solely on morphological structures. Research incorporating deep learning has been conducted to improve accuracy, but creating manual labeling for fully-supervised learning has proven challenging. On the other hand, spectroscopic optical coherence tomography (S-OCT) has been successful in detecting and classifying lipids by utilizing spectroscopic information. However, there have been limitations to enhancing performance. In this study, leveraging the understanding of S-OCT and deep learning networks, a method for lipid detection based on weakly-supervised learning is proposed, and its performance and potential are demonstrated. It is anticipated that this research will overcome the limitations of previous studies and contribute to advancements in various technologies.