Multi-classification of brain tumor by using deep convolutional neural network model in magnetic resonance imaging images

Abstract

Brain tumors are still diagnosed and classified based on the results of histopathological examinations of biopsy samples. The existing method requires extra effort from the user, takes too long, and can lead to blunders. These limitations underline the need of employing a fully automated deep learning system for the multi-classification of brain tumors. In order to facilitate early detection, this study employs a convolutional neural network (CNN) to multi-classify brain tumors. In this research, we present three distinct CNN models for use in three separate categorization tasks. The first CNN model can correctly categorize brain tumors 99.74% of the time. The second CNN model is 96.27% accurate in differentiating between normal, glioma, meningioma, pituitary, and metastatic brain tumors. The third CNN model successfully distinguishes between Grades II, III, and IV brain tumors 99.18% of the time. The Hybrid Particle Swarm Grey Wolf Optimization (HPSGWO) technique is used to quickly and accurately determine optimal values for all of CNN models most important hyperparameters. An HPSGWO algorithm is used to fine-tune all the necessary hyperparameters for optimal classification performance. The results are compared with standard existing CNN models across a range of performance measures. The proposed models are trained using publicly available large clinical datasets. To verify their initial multi-classification of brain tumors, clinicians and radiologists might use the proposed CNN models.