Automated Hole and Non-hole Screening in Retinal OCT Images Using Local Binary Patterns with Support Vector Machine

Abstract

Macular hole (MH) formation in between the retinal layers causes distorted vision and decrease in the person’s visual acuity. MH is also the chronic stage of retinal disorder, significantly treated only by surgeries. This work addresses to the automated screening of the hole and non-hole images in the cross-sectional depiction of retinal optical coherence tomography (OCT) images. Screening between hole and non-hole comprises various pathological traces observed in the retinal OCT scans, aiming to automatically differentiating the retinal hole present in the macula and other macular pathologies. Machine learning utilizes local binary pattern with reduced dimension, as local descriptors, in which the texture information from the retinal OCT images is encoded. For the identification of the MH, support vector machine classifier is used. Dataset was prepared for 51 unseen OCT scans from 14 patients having orientation variabilities and used for extensive experimentation. Method effectiveness was verified through results, and the statement supported with the evaluations was performed across the dataset. Systems’ overall performance was validated stochastically where the system sensitivity is 92.3% and the specificity is 92%.