Integrated Transfer Learning and Nature-Inspired Optimization for Enhanced Feature Extraction in Diabetic Retinopathy Image Analysis

Abstract

This research aims to detect diabetic retinopathy using optimized features extracted from deep learning model. Initially, several deep learning architectures are trained using retinal image dataset and the best model is determined. Regarding transfer learning approaches for diabetic retinopathy patients, SqueezeNet seems to be the best model. The proposed model in this research relies on a two-stage optimization process to enhance the features extracted by SqueezeNet. Deep features obtained by SqueezeNet are optimized using Particle Swarm Optimization (PSO) and the Crow Search Algorithm (CSA). Merging the results of the two optimization methods with a value-maximizing solution is essential for producing an accurate and resilient feature vector. The proposed hybrid model employs a variety of machine-learning algorithms to classify diabetic retinopathy and non-diabetic retinopathy cases. The experimental findings indicate that the suggested method is effective with correct classification accuracy of 96.8%.