Automated Classification of Diabetic Eye Diseases Using Optimized Deep Convolutional Model

Abstract

Diabetic retinopathy is a diabetic complication caused by damage to the blood vessels of the light-sensitive tissue at the back of the retina. This disease has no early symptoms and can lead to blindness. In recent decades, the Deep Learning (DL) approaches have provided a way to accurately detect and classify this eye disease, which helps medical professionals plan treatment procedures. In this study, a novel automatic diabetic retinopathy classification model was proposed by leveraging the efficiency of the Honey Badger Algorithm and Deep Convolutional Neural Network (HBA-DCNN). The retinal images are initially collected and then preprocessed to standardize the images effectively for subsequent processes. Consequently, the image segmentation used superpixel mean orientation to capture the Region of Interest (ROI) from the preprocessed images. Finally, the retinopathy classification was performed using the proposed HBA-DCNN model. The DCNN in the proposed framework learns the patterns differentiating normal and diabetic retinopathy by capturing the most informative and distinct features. The HBA algorithm refines the parameters of the DCNN to its optimal range, which helps reduce the computational time and improves the training performances. The presented methodology was trained and validated using the publicly available two databases, namely Digital Retinal Image for Vessel Extraction (DRIVE) and Structured Analysis of the Retina (STARE), and the experimental results are assessed in terms of accuracy, precision, recall, and F-measure. In addition, the implementation outcomes are compared and validated with state-of-the-art techniques such as Deep Neural Network (DNN) in addition to Support Vector Machine (SVM).