AI-based intelligent hybrid framework (BO-DenseXGB) for multi- classification of brain tumor using MRI

Abstract

A brain tumor is one of the most deadly tumors in the world and can affect both adults and children. According to its shape, severity, or region affected, it comes in different types or grades. The precise treatment strategy necessitates the early detection and classification of the correct type and grade of the tumor. Magnetic Resonance imaging (MRI) is the most extensively used medical imaging technique for examining tumors. The manual examination in clinical practices is constrained by the huge amount of data generated by MRI, which makes tumor classification challenging and time-consuming. Hence, automated methods are the need of the hour for precise and timely diagnosis. This paper proposes Artificial Intelligence (AI) based automated framework to classify tumors into meningioma, glioma, and pituitary classes. The proposed framework exploits the hierarchical feature learning capabilities of the Convolutional Neural Network (CNN) in combination with an optimized boosting classifier. Hyper-parameters of the boosting classifier are tuned with Bayesian Optimization. An overall accuracy of 99.02% is obtained during the experimental evaluation of the proposed model using the benchmark Figshare dataset, which comprises 3064 MRI images. The experimental outcomes confirm that the proposed deep learning strategy outperforms the existing approaches in a convincing manner.