Intelligent System for Screening Diabetic Retinopathy by Using Neutrosophic and Statistical Fundus Image Features.

Abstract

Diabetic retinopathy (DR) is considered as one of the global diseases of blindness, especially for aged people. The main reason behind this disease is the complication of diabetes in retinal blood vessels. Usually, the early warning signs are not observed. Screening is an important key for the diagnosis of early stages of diabetic retinopathy. In this work, a new technique for automatically screening three categories; Normal, Non-Proliferative Diabetic Retinopathy (Non-PDR), and Proliferative Diabetic Retinopathy (PDR) disease is presented that is may help doctors and physicians to make a preliminary decision. Neutrosophic set (NS) domain based on statistical features, Gray Level Cooccurrence Matrix (GLCM), Gray Level Run Length Matrix (GLRLM), and difference statistics were used for features extraction. More than thirty statistical textural features derived from the NS set domain and spatial domain have been tested using a features selection scheme named one-way analysis of variables (ANOVA1) with significance value (p<0.001). After feature selection, about sixteen features were passed the test and introduced to the classification stage which is made up of three techniques Multi-class support vector machine (MUSVM), Naïve Bayes (NB), and Decision Forest (DF) classifiers. Over 50 images from each category were downloaded from Digital Retinal Images for Vessel Extraction (DRIVE) database. The performance resulted for this proposed method shows the system robustness in identifying each stage of diabetic retinopathy within the accuracy, sensitivity, and specificity about 95.5%, 100%, and 93.3% respectively. The results of this method were compared with other considered systems. The fair comparison of results shows system superiority and can be used in clinical observation.