Significant challenges present in both diagnosis and therapy of brain tumors on account of their heterogeneity, infiltrative growth, and high morbidity and mortality. Medical imaging, particularly MRI, remains the mainstay for noninvasive detection and characterization of the tumors as well as for planning treatments. In the last decade, machine learning and deep learning have brought a sea change in automated brain tumor analysis that builds strong foundations for tumor detection, segmentation, grading, and prognostication. This review synthesizes the advances achieved by classical ML and modern DL pipelines; surveys public datasets and benchmarks; compares architectures and learning paradigms; analyzes evaluation metrics and clinical validation; and discusses limitations in terms of data quality, generalization, explainability, and fairness to contribution to the regulatory translation. We thus finish with open issues and promising avenues, including foundation models, multimodal and self-supervised learning, uncertainty aware systems, and human-AI collaboration for clinically trusted deployment.