Hybrid hunter-prey ladybug beetle optimization enabled deep learning for diabetic retinopathy classification
Vidya Sagvekar, Manjusha Joshi, Minu Ramakrishnan, Ajay I Dudani
Abstract
The most common source of blindness and visual damage worldwide is Diabetic Retinopathy (DR). The use of mobile phones for DR identification is becoming more and more popular as mobile phone technology advances since it may be less expensive, more widely available, and simpler to use. Numerous attachments, the smartphone camera, and artificial intelligence can all be used with smartphones to gather and grade retinal images for use in ophthalmoscopy. Additionally, fundus photography is a helpful tool for treating diabetic eye disease. Preventing vision loss requires early diagnosis and treatment of DR. To address theaforementioned problem, a novel optimized Hunter-Prey Ladybug Beetle Optimizer Deep Maxout Network (HPLBO_DMN) is proposed for DR classification. The input image is initially retrieved from the database and preprocessed. The Kalman filter and Region of Interest (RoI) extraction are used in thepreprocessing stage to reduce noise in theinput image. The preprocessed image is next subjected to a wavelet transform, where theMeyer wavelet is used to separate the image into subbands for additional processing.Thereafter, the lesion segmentation is done using K-Net, which is trained using HPLBO. The Automatic Artery/Vein (A/V) Classification Network (AVNet) classifies the arteries and veins from the wavelet-transformed image simultaneously. Additionally, the feature extraction phase obtains the output of the AVNet. Finally, DR classification is carried out with the help of a DMN. Here, DMN is trained using theproposed HPLBO, which is developed using the Hunter–Prey Optimizer (HPO) and the Ladybug Beetle Optimization (LBO). The introduced HPLBO_DMN attained higher performance of accuracy of 93.6%, sensitivity of 93.8%, specificity of 92.9% as well as segmentation accuracy of 92.9%correspondingly.