6 - Segmentation of retinal layers from OCT scans

Abstract

The accurate segmentation of retinal layers of the eye in a 3D optical coherence tomography (OCT) data provides relevant clinical information. This chapter introduces a three-dimensional (3D) segmentation approach that uses an adaptive patient-specific retinal atlas as well as an appearance model for 3D OCT data. To reconstruct that atlas of 3D retinal scan, we first segment the central area of the macula at which we can clearly identify the fovea. Markov Gibbs random field (MGRF) including intensity, shape, and spatial information of 12 layers of retina was all used to segment the selected area of retinal fovea. A set of coregistered OCT scans that were gathered from 200 different individuals were used to build a 2D shape prior. This shape prior was adapted in a following step to the first-order appearance and second-order spatial interaction MGRF model. After segmenting the center of the macula “foveal area,” the labels and appearances of the layers that have been segmented were used to have the adjacent slices segmented as well. The last step was then repeated recursively until the 3D OCT scan of the patient is segmented. This approach was tested on 30 individuals while some of them were normal and others were pathological, and then compared to a manually segmented ground truth and finally these results were verified by medical retina experts. Metrics such as dice similarity coefficient, agreement coefficient, and average deviation metrics were used to measure the performance of the proposed approach. Accomplished accuracy by the proposed approach shows promising results with noticeable advantages over the state-of-the-art 3D OCT approach.