Abstract. We present a method for the incorporation of regional image information in a 3-D graph-theoretic approach for optimal multiple surface segmentation. By transforming the multiple surface segmentation task into finding a minimum-cost closed set in a vertex-weighted graph, the optimal set of feasible surfaces with respect to an objective function can be found. In the past, this family of graph search applications only used objective functions which incorporated "on-surface" costs. Here, novel "in-region" costs are incorporated. Our new approach is applied to the segmentation of seven intraretinal layer surfaces of 24 3-D macular optical coherence tomography images from 12 subjects. Compared to an expert-defined independent standard, unsigned border positioning errors are comparable to the inter-observer variability (7.8 ± 5.0 μm and 8.1 ± 3.6 μm, respectively).

No abstract

Abstract. An optimal 3-D graph search approach designed for simultaneous multiple surface detection is extended to allow for varying smoothness and surface interaction constraints instead of the traditionally used constant constraints. We apply the method to the intraretinal layer segmentation of 24 3-D optical coherence tomography (OCT) images, learning the constraints from examples in a leave-one-subject-out fashion. Introducing the varying constraints decreased the mean unsigned border positioning errors (mean error of 7.3 ± 3.7 μm using varying constraints compared to 8.3 ± 4.9 μm using constant constraints and 8.2 ± 3.5 μm for the inter-observer variability).

Abstract. We have developed a method for the automated segmentation of the internal limiting membrane and the pigment epithelium in 3-D OCT retinal images. Each surface was found as a minimum s-t cut from a geometric graph constructed from edge/regional information and a priori-determined surface constraints. Our approach was tested on 18 3-D data sets (9 from patients with normal optic discs and 9 from patients with papilledema) obtained using a Stratus OCT-3 scanner. Qualitative analysis of surface detection correctness indicates that our method consistently found the correct surfaces and outperformed the proprietary algorithm used in the Stratus OCT-3 scanner. For example, for the internal limiting membrane, 4% of the 2-D scans had minor failures with no major failures using our approach, but 19% of the 2-D scans using the Stratus OCT-3 scanner had minor or complete failures.