Automated segmentation of optical coherence tomography images

Abstract

We propose a fast and accurate automated algorithm to segment retinal pigment epithelium and internal limiting membrane layers from spectral domain optical coherence tomography (SDOCT) B-scan images. A hybrid algorithm, which combines intensity thresholding and graph-based algorithms, was used to process and analyze SDOCT radial scans (120 B scans) images obtained from twenty patients. The relative difference in position of the layers segmented by the proposed hybrid algorithm and by the clinical expert was 1.49% ± 0.01%. The processing time of the hybrid algorithm was 9.3 s for six B scans. Dice’s coefficient of the hybrid algorithm was 96.7% ± 1.6%. The proposed hybrid algorithm for the segmentation of SDOCT images had good agreement with manual segmentation and reduced processing time.