GO-DBN: Gannet Optimized Deep Belief Network Based wavelet kernel ELM for Detection of Diabetic Retinopathy

Document Type


Publication Date



Diabetic retinopathy (DR) is an irreversible disease that damages blood vessels and results in permanent visual impairment. Timely detection and proper diagnosis can mitigate severe eye impacts and slow DR progression. Color fundus images are commonly used by eye specialists to diagnose DR. However, manual operations are tedious, laborious, time-consuming, and error-prone. As a result, automated techniques are gaining increased attention in the medical domain due to their ability to precisely analyze unstructured clinical notes. In this paper, a novel Gannet-optimized deep belief network-based wavelet kernel extreme learning machine (GO-DBN-WKELM) technique is proposed for detecting DR occurrences and assessing its progressive stages. The GO-DBN-WKELM approach detects and classifies DR severities into six distinct classes: no DR (i.e., normal), early DR, mild non-proliferative DR, moderate non-proliferative DR, severe non-proliferative DR, and proliferative DR. Initially, detailed information from the original datasets is extracted by reducing feature dimensions using the Deep Belief Network (DBN) model. Subsequently, the extracted images are analyzed by the proposed GO-DBN-WKELM classification model, which effectively detects and accurately classifies fundus images based on their severities into distinct classes. The enhanced detection performance is primarily due to the application of the GO algorithm with a wavelet kernel extreme learning machine (WKELM). The GO algorithm not only optimizes the kernel parameters of the WKELM but also increases the convergence speed of the classifier. The proposed classifier is analyzed using three different types of datasets, namely MESSIDOR, DIARETDB1, and IDRiD datasets. The effectiveness of the proposed classification model is determined by evaluating its ability to detect DR in terms of diverse performance metrics, such as accuracy, precision, recall, and F-measure. Simulation results reveal that the proposed GO-DBN-WKELM classifier achieved a high accuracy rate of approximately 98% for the MESSIDOR dataset and 97.8% for the DIARETDB1 dataset. These outcomes demonstrate the potential of the GO-DBN-WKELM approach in detecting and classifying diabetic retinopathy severities, offering an efficient, automated alternative to manual methods that can aid eye specialists in providing timely diagnoses and interventions for patients with this condition.

Publication Title

Expert Systems with Applications



This document is currently not available here.