Segmentation of Optic Disc in Fundus Images using Convolutional Neural Networks for Detection of Glaucoma

The condition of the vascular network of human eye is an important diagnostic factor in ophthalmology. Its segmentation in fundus imaging is a difficult task due to various anatomical structures like blood vessel, optic cup, optic disc, macula and fovea. Blood vessel segmentation can assist in the detection of pathological changes which are possible indicators for arteriosclerosis, retinopathy, microaneurysms and macular degeneration. The segmentation of optic disc and optic cup from retinal images is used to calculate an important indicator, cup-to disc ratio( CDR) accurately to help the professionals in the detection of Glaucoma in fundus images.In this proposed work, an automated segmentation of anatomical structures in fundus images such as blood vessel and optic disc is done using Convolutional Neural Networks (CNN) . A Convolutional Neural Network is a composite of multiple elementary processing units, each featuring several weighted inputs and one output, performing convolution of input signals with weights and transforming the outcome with some form of nonlinearity. The units are arranged in rectangular layers (grids), and their locations in a layer correspond to pixels in an input image. The spatial arrangement of units is the primary characteristics that makes CNNs suitable for processing visual information; the other features are local connectivity, parameter sharing and pooling of hidden units. The advantage of CNN is that it can be trained repeatedly so more features can be found. An average accuracy of 95.64% is determined in the classification of blood vessel or not. Optic cup is also segmented from the optic disc by Fuzzy C Means Clustering (FCM). This proposed algorithm is tested on a sample of hospital images and CDR value is determined. The obtained values of CDR is compared with the given values of the sample images and hence the performance of proposed system in which Convolutional Neural Networks for segmentation is employed, is excellent in automated detection of healthy and Glaucoma images.