Automated segmentation of retinal layers from optical coherence tomography images using geodesic distance. (December 2017)

• Record Type: Journal Article
• Authors: Duan, Jinming; Tench, Christopher; Gottlob, Irene; Proudlock, Frank; Bai, Li
• Is Part Of: Pattern recognition. Volume 72(2017:Dec.)
• Page Start: 158
• Page End: 175
• Publication Date: 2017-12
• Subjects: Optical coherence tomography -- Segmentation -- Geodesic distance -- Eikonal equation -- Partial differential equation -- Ordinary differential equation -- Fast sweeping; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2017.07.004
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We proposed an automated method that is able to segment healthy and pathological retinal layers from 2D/3D optical coherence tomography images. The method uses a weighted geodesic distance efficiently derived from an Eikonal equation via fast sweeping. Segmentation proceeds by solving an ordinary differential equation. We introduce an OCT-specific weight function into the geodesic distance framework. This is the first work on segmentation of intra-retinal layer structures using geodesic distance. Abstract: Optical coherence tomography (OCT) is a noninvasive imaging technique that can produce images of the eye at the microscopic level. OCT image segmentation to detect retinal layer boundaries is a fundamental procedure for diagnosing and monitoring the progression of retinal and optical nerve diseases. In this paper, we introduce a novel and accurate segmentation method based on geodesic distance for both two and three dimensional OCT images. The geodesic distance is weighted by an exponential function, which takes into account both horizontal and vertical intensity variations in the image. The weighted geodesic distance is efficiently calculated from an Eikonal equation via the fast sweeping method. Segmentation then proceeds by solving an ordinary differential equation of the geodesic distance. The performance of the proposed method is compared with manual segmentation. Extensive experiments demonstrate that the proposed method is robust to complex retinal structures with large curvature variations and irregularities and it outperforms the parametric active contour algorithm as well as graph based approaches for segmenting retinal layers in both healthy and pathological images.