MICRO STATISTICAL DESCRIPTORS FOR GLAUCOMA DIAGNOSIS USING NEURAL NETWORKS

Abstract

A fully automatic Computer Aided diagnosis (CAD) of glaucoma is developed that aims to reduce the false positive detection rate and increasing the sensitivity of classification. It consists of three main steps: Region of Interest (ROI) extraction (Optic Disc (OD) region), feature extraction (micro textures) and classification using Linear Vector Quantizer-Artificial Neural Network (LVQ-ANN). The search area for glaucoma is the OD region wherein the cupping occurs, so in the first step ROI is extracted from the whole image. Feature extraction and classification are the most challenging tasks as the performance of the system depend both of them. Laws defined five spatial filters to extract micro-statistical estimators such as Level, Edge, Spot, Wave, and Ripple. Fundus images in three databases; DRISHTI-GS1, ORIGA, and RIM-ONE are classified using LVQ-ANN classifier. Results indicate the strength of the LVQ-ANN classifier for glaucoma diagnosis with sensitivity of 95.71% (DRISHTI-GS1), 83.33% (ORIGA) and 94.87% (RIM-ONE).