Semi-automated system for cup to disc measurement for diagnosing glaucoma using classification paradigm

Abstract

Recently, Glaucoma has become one of the major retinal diseases. In order to detect such retinal diseases, cup to disc ratio measurement is a vital index of Glaucoma, as the Glaucomatous neuropathy increases the cup to disc ratio when the excavation of the optic cup is increased. In this paper, a semi-automated system to detect both of optic cup and optic disc and to measure cup to disc ratio has been proposed. The proposed system firstly, uses an object detection function from red channel of the retinal images. Then further using threshold values, the optic cup and optic disc are detected. Although, for several images manual tuning is needed as the object detection function as well as the threshold value fail to detect the optic cup and optic disc correctly. The manually tuned images and the automatically detected images are further used to determine the error in the system which leads to the categorizing of the images. These images are later post-processed using Haralick texture features. Haralick texture features’ obtained values are trained using back propagation neural network to determine the system’s accuracy. The proposed system was evaluated using RIM-ONE database. By increasing the absolute error, system’s accuracy is evaluated. The proposed system’s accuracy is 86.43% at 0.5 error value.