Lesion Detection in Retinal Images using Back Propagation Neural Network

Exudates are one of the chief signs of diabetic retinopathy, which is a main root of blindness and can be prevented with an early screening process. In this article, authors have attempted to detect exudates using back propagation neural network. The publicly available diabetic retinopathy dataset DIARETDB1 has been used in the evaluation process. To prevent the optic disk from interfering with exudates detection, the optic disk is eliminated. Significant features are notorious from the similes after preprocessing by using two methods: Decision tree and GA-CFS method are used as input to the BPN model to detect the exudates and non-exudates at pixel level. The results attest that, BPN recital with features identified by Decision tree and GA_CFS loom has out performed the performance of BPN with all inputs. The BPN classifier best recital was found with Sensitivity of 96.97 %, Specificity of 100% and classification accuracy of 98.45%.