Clinical-Grade Precision Without the Complexity: A YOLO-Based Brain Tumor Detection Framework

Abstract

The precise and efficient diagnosis of brain tumors in magnetic resonance (MR) images is extremely important as it enable to treat these kinds of the disease earlier and more effectively. Although models from the YOLO (You Only Look Once) series of real-time object detection algorithms have attained favorable results in medical image analysis, performance comparison of the most recently released YOLO versions is less explored, particularly in multi-class brain tumor classification tasks. In this paper, we assess and compare the performance of three YOLO versions YOLOv10, YOLOv11, and YOLOv12 on an annotated MRI dataset with 308 brain tumor images including glioma, meningioma, and pituitary adenoma cases. Models are finetuned throughout on same settings and evaluated with standard metrics like precision, recall, mAP@0. 5, and mAP@0. 5-0.95. YOLOv11 has the best average recall (89.2%) and mAP @0. 5 (91.6%), although YOLOv12 achieves the best precision (89.8%), and performs best in diagnosing pituitary tumor. YOLOv10...is of good performance.., but it is obviously deficient in glioma detection. There are trade-offs between detection sensitivity and calss-specific accuracy across models, with YOLOv11 representing the best in trade- offs and the most suitable in the real-time clinical deployment. This work provides a practical guide to choose the optimal YOLO architecture for automated brain tumor localization and has potential to enhance diagnostic speed and reliability in neuro- oncology pipelines.