Statistical Parameters for Glaucoma Detection from Color Fundus Images

Abstract

Glaucoma is an ocular disease, which is considered to be the second main cause of blindness worldwide. This paper presents a glaucoma detection system based on the statistical parameters computed from the segmented optic disc and optic cup from the color fundus images. An improved FCM algorithm based on the morphological reconstruction and membership filtering (FRFCM) is used for optic disc and optic cup segmentation. The cup-to-disc ratio is one of the main discriminatory parameters of glaucoma infection and is calculated from the segmented results. In addition, several statistical features such as cup entropy, rim entropy, and kurtosis are calculated. Based on these values the glaucoma images are seperated from the non glaucoma images. The algorithm is evaluated on publicly available RIM-ONE and DRIONS-DB databases. The presented method has achieved an accuracy of 92.85%, 98.57%, 98.57%, and 87.14% in terms of the parameters CDR, cup entropy, kurtosis, and rim entropy, respectively for RIM-ONE database. The obtained results are compared with the ground truth values and an average error of 0.0402, 0.0473, 0.0462, and 0.0570 is achieved in terms of parameters CDR, cup entropy, rim entropy, and kurtosis, respectively. Classification of glaucoma and non glaucoma images by statistical parameters (cup entropy and kurtosis) gives better results compared to CDR based classification.