Performance Evaluation of ML Algorithms for Brain Tumor Classification on Brats Dataset

Brain neoplasms represent some of the most severe and life-threatening manifestations of cancer, with gliomas constituting the most prevalent subtype, which is further classified into Low-Grade Glioma (LGG) and High-Grade Glioma (HGG). The prompt identification, precise categorization, and accurate delineation of these tumors are imperative for effective diagnostic assessment, prognostic evaluation, and therapeutic strategy formulation. Nevertheless, the manual evaluation of magnetic resonance imaging (MRI) scans is labor-intensive and prone to inter-observer variability. In recent years, advancements in artificial intelligence (AI), particularly within the realms of machine learning (ML) and deep learning (DL) methodologies, have markedly enhanced the automation and precision of brain tumor identification and evaluation.This comprehensive review paper investigates contemporary developments in the automated classification and segmentation of brain tumors utilizing the BraTS dataset, which is widely recognized as a standard reference within the neuroimaging research community. We analyze a diverse array of methodologies, encompassing traditional machine learning algorithms, Convolutional Neural Networks (CNNs), UNet architectures, hybrid models, and ensemble-based strategies. The article underscores techniques that synergistically integrate handcrafted radiomic features with deep learning features to enhance both model robustness and interpretability. Documented accuracy rates in the studies reviewed vary from 66% to 99%, with Dice similarity coefficients employed to assess the segmentation efficacy of tumor sub-regions including tumor core (TC), whole tumor (WT), and enhancing tumor (ET). This scholarly endeavor aspires to furnish researchers and practitioners with valuable insights into cutting-edge methodologies for brain tumor analysis and to offer direction for forthcoming advancements in this domain.