Deer Hunting Optimization with 3D-Convolutional Neural Network for Diabetic Retinopathy Classification Model

Abstract

A retina disease caused by high glucose levels in the blood is called Diabetic Retinopathy (DR) and is the world’s leading cause of blindness. To avoid or delay vision degradation and loss, early diagnosis and treatment are required. As a result, the creation of an automated method for accurate DR identification is essential. For this, in this paper, a 3D-Convolution Neural Network (3D-CNN) with Deer Hunting Optimization (DHO) algorithm is proposed for detecting and classifying DR images. The proposed 3D-CNN-DHO approach includes four phases such as pre-processing, segmentation, feature extraction, and classification. The contrast of the DR image is first improved using a Contrast-Limited Adaptive Histogram Equalization (CLAHE) approach. Subsequently, the threshold-based effective segmentation is carried out. Then, the Resnet50 model is implemented to extract the features from the image. Finally, 3D-CNN-DHO-based classifier model is implemented to categorize the various DR stages. The experiments are carried out in detail and evaluated on the Messidor DR benchmark dataset. The acquired experimental result demonstrated the 3D-CNN-DHO model’s outstanding qualities by achieving optimal specificity, sensitivity, recall, precision, and accuracy.