Clinical validation of an algorithm for rapid and accurate automated segmentation of intracoronary optical coherence tomography images

Objectives: The analysis of intracoronary optical coherence tomography (OCT) images is based on manual identification of the lumen contours and relevant structures. However, manual image segmentation is a cumbersome and time-consuming process, subject to significant intra- and inter-observer variability. This study aims to present and validate a fully-automated method for segmentation of intracoronary OCT images.

Methods: We studied 20 coronary arteries (mean length=39.7±10.0 mm) from 20 patients who underwent a clinically-indicated cardiac catheterization. The OCT images (n=1812) were segmented manually, as well as with a fully-automated approach. A semi-automated variation of the fully-automated algorithm was also applied. Using certain lumen size and lumen shape characteristics, the fully- and semi-automated segmentation algorithms were validated over manual segmentation, which was considered as the gold standard.

Results: Linear regression and Bland-Altman analysis demonstrated that both the fully-automated and semi-automated segmentation had a very high agreement with the manual segmentation, with the semi-automated approach being slightly more accurate than the fully-automated method. The fully-automated and semi-automated OCT segmentation reduced the analysis time by more than 97% and 86%, respectively, compared to manual segmentation.

Conclusions: In the current work we validated a fully-automated OCT segmentation algorithm, as well as a semi-automated variation of it in an extensive "real-life" dataset of OCT images. The study showed that our algorithm can perform rapid and reliable segmentation of OCT images.