Finding brain tumors early using MRI scans is now a very important way that artificial intelligence is being used in health cared MRI)., because brains are complex, and it is hard to do diagnoses by hand using Magnetic Resonance Imaging (MRI). Over the past decade, ML, DL, and combinations of the two have greatly made brain tumor analysis more accurate and automatic. Traditional ML methodologies, such as SVM, Random Forests, and KNN, are highly dependent on hand-crafted features, but they tend to fail in various contexts. Some new DL methods, including convolutional neural network, U-Net architecture, capsule network, Efficient Net and sophisticated YOLOv7 using attention, have achieved accuracy greater than 99% on reference sets of data (e.g. BRATS, Fig Share, Harvard MRI).These models use techniques such as preprocessing, segmentation, transfer learning, and data augmentation to increase robustness across different tumor types (glioma, meningioma, pituitary, and metastatic tumors). Although considerable progress has been made, problems remain with respect to computational complexity, interpretability, identification of small tumors, and generalization to different datasets. This survey paper collates with 2014–2025 state-of-the-art approaches, contrasting methodology, datasets, and performance results. It indicates the trend in research, defines open challenges, and suggests directions for future work, such as explainable AI, multimodal fusion, and lightweight models for clinical deployment.