Fused Texture Feature of Segmented Retinal Image Based Multiretinal Disease Classification

Abstract

The examination of retinal blood vessels is crucial for ophthalmologists to diagnose various eye abnormalities, including diabetic retinopathy, glaucoma, cardiovascular diseases, high blood pressure, arteriosclerosis, and age-related macular degeneration. The manual scrutiny of retinal vasculature poses a significant challenge for medical professionals due to the intricate structure of the eye, the minuscule size of blood vessels, and the variability in vessel width. In recent literature, numerous automated techniques for retinal vessel extraction have been proposed, offering valuable assistance to ophthalmologists in promptly identifying and diagnosing eye disorders. The study introduces a comprehensive model that assesses and evaluates the performance of 13 state-of-the-art machine learning networks. This model aims to contribute to deep feature extraction and image classification for fundus images. The proposed approach extracts the Gray-Level Co-occurrence Matrix, Histogram of Oriented Gradients, Wavelet, Tamura, Law's of Texture Energy, and Local Binary Pattern texture feature vector from the segmented retinal blood vessel structure. After extracting the feature, create the group of all possible combinations by using six feature vectors. After an exhaustive experimental analysis, we select a suitable group of feature vectors and apply a machine Learning Classifier to classify the four Retinal blood vessel-related diseases, namely Hypertensive Retinopathy, Pathological myopia, Moderated Diabetic retinopathy, and Healthy retina. Finally, obtain the accuracy of 97.7% with Cubic SVM.