Modified Social Group Optimization Based Deep Learning Techniques for Automation of Brain Tumor Detection–A Health Care 4.0 Application

Abstract

Now-a-days, Segmentation is essential in diagnosing severe diseases wherever there is a scope for image processing. In this work, hybridization of most popular and metaheuristic algorithms with Conventional Neural Network (CNN) has been proposed. As a part of the study, jelly fish and Modified Social Group Optimization Algorithms (MSGOA) are used. The CNN weights and the corresponding hyper parameters are modified or designed with the help of the respective metaheuristic approach of the algorithm. This certainly improved the efficiency of the segmentation which is measured in several metrics of bio-medical image processing. The accuracy, loss, Intersection over Union (IoU) are some of those metrics which are employed in this study for better understanding of the algorithm’s effectiveness. Further the detection process is simulated consuming 100 iterations uniformly in either of the algorithms. The proposed methodology has efficiently segmented the tumor portion. The simulation has been carried out in MATLAB and the results are presented in terms of computed metrics, convergence plots and segmented images.