Model-Driven 3-D Regularisation for Robust Segmentation of the Refractive Corneal Surfaces in Spiral OCT Scans

Measuring the cornea's anterior and posterior refractive surface is essential for corneal topography, used for diagnostics and the planning of surgeries. Corneal topography by Optical Coherence Tomography (OCT) relies on proper segmentation. Common segmentation methods are limited to specific, B-scan-based scan patterns and fail when applied to data acquired by recently proposed spiral scan trajectories. We propose a novel method for the segmentation of the anterior and posterior refractive surface in scans acquired by 2-D scan trajectories including but not limited to spirals. Key feature is a model-driven, three-dimensional regularisation of the region of interest, slope and curvature. The regularisation is integrated into a graph-based segmentation with feature-directed smoothing and incremental segmentation. We parameterise the segmentation based on test surface measurements and evaluate its performance by means of 18 in vivo measurements acquired by spiral and radial scanning. The comparison with expert segmentations shows successful segmentation of the refractive corneal surfaces.