Advancing Automated Brain Tumor Detection: A YOLOv11-Based Deep Learning Approach for Real-Time MRI Analysis

Accurate and rapid detection of brain tumors in magnetic resonance imaging (MRI) scans is critical for timely diagnosis and effective treatment planning. Manual interpretation of MRI data is time-consuming, subject to inter-observer variability, and limited in scalability, which highlights the need for automated solutions. This study presents a robust deep learning framework based on the latest YOLOv11 object detection architecture for real-time localization of brain tumors. A four-phase pipeline is implemented, consisting of dataset preparation, baseline training, hyperparameter optimization, and model evaluation. The Roboflow Universe Brain Tumor Dataset, including annotated categories of glioma, meningioma, pituitary tumor, and healthy cases, is preprocessed and partitioned into training, validation, and test sets to ensure unbiased assessment. Two YOLOv11 variants are systematically trained and evaluated. The YOLOv11m, achieved an mAP@50 of 0.9063, precision of 0.8858, and recall of 0.8614, delivering highly competitive results compared to YOLOv11s (mAP@50 = 0.9076). While YOLOv11s showed a marginal 0.14% higher detection accuracy. A comprehensive data analysis is performed using precision–recall curves, confusion matrices, ROC curves, and class-wise performance metrics to identify strengths and limitations across tumor categories. Notably, performance varied by tumor type: No Tumor (AP = 0.973) and Meningioma (AP = 0.964) achieved near-perfect detection, while Glioma (AP = 0.741) remained more challenging due to irregular shapes and contrast variations. These results demonstrate that YOLOv11m can deliver competitive detection accuracy with significantly faster inference than YOLOv11s and traditional CNN-based approaches, thereby enhancing both speed and reliability in automated neuro-oncological diagnostics. Future research will focus on integrating cross-dataset generalization, improving boundary localization (mAP@50–95), and extending the framework to multimodal MRI scans to support broader clinical applications. These results demonstrate the potential of advanced real-time object detection architectures in enhancing the speed and reliability of automated neuro-oncological diagnostics, supporting clinical workflows with consistent and precise tumor identification.