Probabilistic Intra-Retinal Layer Segmentation in 3-D OCT Images Using Global Shape Regularization

With the introduction of spectral-domain optical coherence tomography (OCT),\nresulting in a significant increase in acquisition speed, the fast and accurate\nsegmentation of 3-D OCT scans has become evermore important. This paper\npresents a novel probabilistic approach, that models the appearance of retinal\nlayers as well as the global shape variations of layer boundaries. Given an OCT\nscan, the full posterior distribution over segmentations is approximately\ninferred using a variational method enabling efficient probabilistic inference\nin terms of computationally tractable model components: Segmenting a full 3-D\nvolume takes around a minute. Accurate segmentations demonstrate the benefit of\nusing global shape regularization: We segmented 35 fovea-centered 3-D volumes\nwith an average unsigned error of 2.46 $\\pm$ 0.22 {\\mu}m as well as 80 normal\nand 66 glaucomatous 2-D circular scans with errors of 2.92 $\\pm$ 0.53 {\\mu}m\nand 4.09 $\\pm$ 0.98 {\\mu}m respectively. Furthermore, we utilized the inferred\nposterior distribution to rate the quality of the segmentation, point out\npotentially erroneous regions and discriminate normal from pathological scans.\nNo pre- or postprocessing was required and we used the same set of parameters\nfor all data sets, underlining the robustness and out-of-the-box nature of our\napproach.\n