기본 정보
연구 분야
프로젝트
논문
구성원
article|
인용수 0
·2025
Generated polarization-sensitive optical coherence tomography image using contrastive unpaired translation
Thanh Dat Le, Yong-Jae Lee, Tae Joong Eom, Chang-Ho Lee
초록

Polarization-sensitive (PS-OCT) optical coherence tomography provides valuable insight information of tissue birefringence by analyzing the sensitive light scattering. Current PS-OCT systems require intricate optical alignment and an additional spectrometer, making them complex and resource-intensive. To address these challenges, we propose a novel approach that leverages contrastive unpaired translation (CUT) to synthesize PS-OCT images directly from standard OCT intensity data. In our study, we applied the patch-based CUT model to generate cross-sectional PS-OCT images. A dataset was created to capture the tendon healing process in mice over six weeks, resulting in 10,000 paired B-scan samples of PS-OCT and OCT images. These were divided into training and testing datasets with a 4:1 ratio for all GAN models. Our findings demonstrate that CUT is an effective deep learning method for generating high-quality synthetic PS-OCT images. Validation through classification models showed that up to 90% of the synthetic samples meet the same structure as PSOCT B-scans provided from SD-PS-OCT system. This work highlights CUT as a reliable and efficient solution for creating synthetic PS-OCT images, supported by consistent classification and repeatability results.

키워드
Optical coherence tomographyPolarization (electrochemistry)Translation (biology)Optical tomographyComputer scienceOpticsArtificial intelligenceComputer visionMaterials sciencePhysics
타입
article
IF / 인용수
- / 0
게재 연도
2025