Self evolving intelligent system for classifying numerous retinal diseases

Abstract

Our eyes and the parts of our brain that allow us to understand the visual information onereceives from our eyes, comprise of a unique and awe inspiring sense known as sight. Theretina is a complex tissue at the back of the eye that contains specialized photoreceptor cells,called rods and cones. They are connected to a network of nerve cells for the local processingof visual information.Retina is susceptible to a variety of diseases that can lead to visual loss or complete blindness.Many a people are vulnerable to such diseases, but in the presence of more deadly diseasesinflicting humanity, like AIDS, cancer etc., eye diseases receive little in research funding. Thiswork is an effort in trying to come up with some intelligent ways of detecting diseases that iscost-effective and efficient.The signatures for the retinal image patterns were defined and then were used to train thenetwork. 45 retinal images were presented to a multilayer perceptron neural network, whichengaged a learning process using Levenberg Manquardt algorithm. The patterns out of theimage signatures wereextracted by the neural network, on the basis of which it classified testimages pattern that is specified the category to which that test image belongs.The sensitivity and specificity of the recognition for the three image patterns that is DiabeticRetinopathy, AMD Macular Hole and Normal Eye sampled for classification were 100%,83.5% and 75% respectively. Thus the overall performance regarding the correct classificationattained by the system was 85.7%.