Automatic Diagnosis and Grading of Diabetic Retinopathy using Bat Optimization Algorithm-Refined Deep Residual Network

Diabetic Retinopathy (DR) is one of the most severe sight-threatening disorders resulting from diabetes, and can eventually lead to blindness and visual impairment. Early detection and medical therapy can assist in controlling and preventing DR progression. However, manual grading is extremely difficult and time-consuming because of the retina’s complicated structure. In this research, the Bat Optimization Algorithm- Refined Deep Residual Network (BOA-RDRN) is proposed for automatic diagnosis and grading of DR. Initially, the DIARETDB0 dataset is employed for the proposed method, and then pre-processing is performed for image denoising which eliminates noise. The Attention-based Fusion Network (AFU-Net) is used to segment the lesion region. The Grey-Level Co-occurrence Matrix (GLCM) and Local Ternary Pattern (LTP) are used to extract the features and BOA is employed to find an optimal subset of features in retinal images. Finally, the RDRN is utilized to classify the anomalies as normal or abnormal. The BOA-RDRN achieves an accuracy of 99.42% compared to the existing methods such as Tunicate Swarm Spider Monkey optimization-based Refined Deep Residual Network (TSSMO-RDRN), Weighted Kernel Fuzzy C-Means Clustering and Dilation-Based Function (WKFM-DBF), Improved Harris Hawk Optimization-based Convolution Neural Network (IHHO-CNN), and Enhanced Fuzzy C-Means Clustering-Resnet 152 respectively.