Prediction and Segmented Analysis on Glaucoma Disease Using Deep Learning Algorithms

Abstract

Diagnostic devices also pose difficulties in identifying glaucoma eye diseases. The ophthalmologists assess glaucoma during the course of screening by changing the shape of the optic disc, losing nervous fibers and atrophy of the peripapiliar zone. This paper describes the Convolutionary neural network unattended architecture to derive the characteristics from raw pixel intensities through multi-layer architecture. The deep-belief network (DBN) model was subsequently used to select the most discriminatory deep features based on the trained data set annotated. Finally, the ultimate recommendation is to distinguish between glaucoma and non- glaucoma retinal fundus picture using a softmax linear classification. This method is classified as the Glaucoma-Deep device and is validated with 1200 retinal images from data sets accessible publicly and privately. The intensity, precision, consistency and PRC (statistical measurements) have been used to test the efficiency of the Glaucoma Deep network. On average, 84.5% of SE, 98.01% of SP, 99% of ACC, and 84% of PRC have been reached.