Vitreo-retinal interface segmentation from spectral-domain OCT using change detection and belief propagation

Vitreo-retinal interface (VRI) segmentation is a common prerequisite of automated quantitative analysis of retinal structures.This work presents a new, fast and robust 3D VRI segmentation from SD-OCT images using change detection and belief propagation. The proposed technique breaks the 3D VRI segmentation into two sub-problems, each of which can be solved with linear time complexity. A modified change detection algorithm is first applied along the A-scan direction to locate potential VRI surface points. Subsequently, the VRI surface is constructed from the potential surface points by systematically incorporating contextual constraints with a belief propagation algorithm. The efficacy of the proposed 3D VRI segmentation is validated with 25 SD-OCT data sets.