A Critical Review on Segmentation of Glioma Brain Tumor and Prediction of Overall Survival

In recent years, the surge in glioma brain tumor cases has positioned it as the 10th most prevalent tumor affecting individuals across diverse age groups. Gliomas, characterized by their invasive nature, unpredictable localization, and heterogeneous subregions, pose a substantial threat to public health. Accurate segmentation of glioma subregions within magnetic resonance imaging (MRI) images is pivotal for the efficient planning of treatment and the prognostication of overall patient survival. This review examines recent advancements in glioma brain tumor segmentation (BTS) and overall survival (OS) prediction while addressing inherent biases and proposing innovative solutions. We explore the evolution of convolutional neural network (CNN) architectures, from traditional 2D CNNs to advanced 2.5D and 3D CNNs, which have greatly enhanced segmentation accuracy and efficiency. Furthermore, cutting-edge techniques for BTS, including attention mechanisms, ensemble methods, transformer-based models, and generative adversarial networks (GANs), are discussed. Additionally, we examine machine learning (ML) models for OS prediction, including support vector machines (SVM) and random forest regressors (RFRs), as well as pioneering methods such as radiomics-based approaches, consensus-based classifiers, and explainable artificial intelligence (XAI). By comparing different preprocessing techniques, model architectures, data sources, and evaluation metrics, we identify the most effective methods and emphasize the importance of collaboration in developing reliable prognostic tools. By consolidating current research, this paper advances ongoing investigations and offers a visionary path for future studies, providing guidance to healthcare stakeholders for refining patient care strategies in glioma management.