EfficientNet with Optimal Wavelet Neural Network for DR Detection and Grading

Diabetic retinopathy (DR) is an illness caused by diabetes, affecting irreversible injury to the retina blood vessels. Since the traditional diagnosis process is expensive and time consuming, automated DR detection models are necessary to identify the disease at the early stage. The recently developed deep learning (DL) models make it an effective approach for accomplishing interesting solutions for medical image analysis problems. This study introduces an EfficientNet with Chicken Swarm Optimization based Wavelet Neural Network (EN-CSOWNN) model for DR Detection and Grading. The goal of the EN-CSOWNN technique is to classify the different stages of DR using retinal fundus images. In addition, the EN-CSOWNN technique involves the design of U-Net based segmentation approach to determine the infected regions in the fundus image. Besides, EfficientNet model is used for deriving a set of feature vectors and WNN model is employed to allot distinct class labels. Finally, the CSO algorithm is utilized to properly adjust the initial parameters (connection weights, scaling factor, and translation factor) of the WNN model and thereby improves the classification performance significantly. In order to demonstrate the better performance of the EN-CSOWNN technique, a comprehensive set of simulations take place on benchmark MESSIDOR dataset and the results depict the enhanced classification outcome of the EN-CSOWNN technique with the maximum accuracy of 98.60%.