Retinal thickness estimation from SD-OCT macular scans

Abstract

Glaucoma, a leading cause of blindness worldwide, can be detected using retinal thicknesses from spectral-domain optical coherence tomography (SD-OCT) scans of the macula. We calculate the desired thickness maps as the distance between the inner-limiting membrane (ILM) and retinal pigmented epithelium (RPE) of the retina. To delineate these two layers, we use a set of two deformable open surfaces that are driven by intensity contrast, while preserving their shape and topology properties, i.e. local surface smoothness and inter-surface distance smoothness. To evaluate our method, qualified graders manually segmented 30 random sections from 20 OCT image stacks, in triplicate; we make comparisons with obtained ground-truth and the clinically tested Heidelberg Spectralis segmentation. We show the superiority of our method with respect to accuracy and average execution time (~7 secs), validating it as a clinical tool.