Abstract:
Brain tumor detection from magnetic resonance imaging (MRI) scans is crucial for early diagnosis and effective treatment planning. However, manual interpretation is slow, subjective and error-prone. With the rapid advancement of deep learning and object detection methods in recent years, automated methods have shown significant potential to overcome these challenges. This study evaluated the YOLOv12 architecture for automated brain tumor detection and compared its performance with YOLOv11, SSD and RT-DETR using a publicly available dataset. Among these models YOLOv12 demonstrated the highest performance, achieving 89.7% accuracy and 92.0% F1-score. In particular, significant improvements were observed in the detection of glioma, which are clinically challenging due to their irregular shapes and diverse textures. Architectural innovations in YOLOv12 models such as attention mechanisms, dynamic anchor boxes and advanced augmentation techniques played important roles in enhancing its robustness across multiple tumor classes. Also, YOLOv12 maintains strong computational efficiency, making it suitable for deployment in low-resource medical environments where real-time analysis is required. The comparative analysis highlights that SSD and Rt-DETR provided competitive results and YOLOv11 provided a strong baseline. YOLOv12 outperformed all other models in terms of accuracy, reliability and efficiency. These results provide the effectiveness of YOLOv12 in medical imaging strongly highlighting its research significance and practical value as a diagnostic support tool. Incorporating YOLOv12 into clinical workflows could help radiologists diagnose brain tumors faster more consistently and reliably.
