The estimation of optic disc location via a novel algortihm for diabetic retinopathy detection

In this study, localization of the Optic Disc (OD) on fundus images which is the basic step of many studies detecting some common retinal diseases like Diabetic Retinpathy (DR) is performed using digital image processing. The most of these studies have been on OD detection since the fact that it is the prerequisite for the extraction of the other image features and components of retinal diseases. First of all, Contrast Limited Adaptive Histogram Equalization (CLAHE) has been applied on intensity color channel of fundus images. Afterwards, the Red Green Blue (RGB) image has been converted to grayscale and applied Morphological Closing Operation (MCO) respectively. The vessels in the image has been able to get rid of by applying MCO but not extracting vessels like the former studies. Thereafter, the Canny Edge Detection (CED) algorithm has been applied to the closed image. More and more, because of the fact that OD edges may have been detected as disconnected, these edges have been applied MCO with a disk structruing element of a diameter value within 3 and 10 iteratively. Afterwards, all circular patterns in a predefined diameter range as an OD candidate has been localised by applying the Circular Hough Transform (CHT) algorithm over the detected edges. The threshold representing the yellowish region in green channel histogram is iteratively calculated by a novel algorithm. The detected circles as OD candidate have been masked over the green color channel and two novel features have been extracted from these masked regions using the calculated threshold. Each detected circle has been classified by applyinng its extracted features to a Multi Layer Perceptron (MLP) using two different training mothods. The success ratio of this Artificial Neural Network (ANN) classifier is 87.50% and 95.00 % for two different evaluation criteria.