An Automated Deep Learning Framework for Multiclass Brain Tumor Detection in MRI Images

Using magnetic resonance imaging (MRI) to detect and classify brain tumors continues to be a crucial challenge in medical diagnostics, requiring precise, effective, and solutions that are easily available. This thorough study examines 25 cutting-edge research publications published between 2015 and 2025 with an emphasis on deep learning techniques for automated brain tumor identification, classification, and segmentation. Traditional CNNs, sophisticated YOLO architectures, U-Net variations, Transformer-based models, and hybrid ensemble methods are among the methodologies that are methodically examined in this research. According to performance measures from several research, 3D segmentation methods yield Dice coefficients of 93–98%, while 2D CNN algorithms achieve accuracy between 82 and 98%. YOLOv7 and YOLOv8 real-time detection systems have mean Average Precision (mAP) values ranging from 0.91 to 0.95, providing notable benefits in computing efficiency. Surgical planning benefits from improved spatial knowledge through the combination of augmented reality (AR) and 3D visualization approaches. Limited multi-institutional dataset validation, computational limitations in resource-constrained environments, class imbalance issues, and a lack of real-world clinical deployment studies are some of the major research gaps that have been found. This review lays the groundwork for future research directions in easily accessible, precise, and clinically feasible brain tumor diagnostic systems by offering a systematic comparative comparison of methodology, datasets, and performance indicators.