Hybrid Optimization for Enhanced Retinal Image Classification Using Neural Networks

The present study proposes a hybrid optimization algorithm that involves the integration of Neural Networks (NN), Genetic Algorithms(GA), and Particle Swarm Optimization(PSO) to improve the accuracy and efficiency of retinal image analysis. First, a wide range of features are extracted from retinal images and normalized. Then a NN is constructed with its weights and biases initialized to process these features. The fitness of each feature subset is measured with the help of a network's output, quantified by a Mean Squared Error(MSE)-based fitness function. Receiving fitness values, the GA component picks out the most important subsets of features, which are refined with crossover and mutation operations to retain diversity and improve the search process. In a parallel process, PSO dynamically adjusts the position of particles representing subsets of features through modification of velocity and position vectors under the influences of cognition and society. With the proposed integrated approach, iteratively, an optimal feature subset is generated by minimizing error and maximizing classification performance. Finally, after convergence, this last optimized subset of features is used in the final tuning of the NN to achieve a far-reaching improvement in the accuracy of classification. This method takes advantage of the complementary strengths of NN, GA, and PSO; therefore, it is able to provide an effective solution for the optimization of features within medical image analysis.