Automated Computational Diagnosis of Peripheral Retinal Pathology in Optical Coherence Tomography (OCT) Scans using Graph Theory

Abstract

Analysis of retinal shape with optical coherence tomography (OCT) has been valuable in describing different ophthalmic conditions. An effective method for retinal contour delineation is graph theory. This study compares the ability of two different implementations of graph theory, the Livewire (LVW) intelligent scissors developed for ImageJ and a purpose-built graph searching function (GSF), to determine retinal shape for a retinal disease classifier. Both methods require user interaction. Retinal shape features derived from both methods were used to diagnose eyes with posterior vitreous detachment (PVD) or retinal detachment (RD) via quadratic discriminant analysis. Classification with each method was the same in 49 out of 51 eyes. Processing time was faster with the GSF than LVW. In mean (µ) ± standard deviation (SD), GSF took 524 ± 62 s and LVW took 814 ± 223 s (p = 5.52 x 10−14). Conclusively, GSF was easier to use and is preferred for further retinal shape analysis.