Application of algorithms based on improved YOLO in MRI image detection of brain tumors

Jinghui Chen¹Tao Yang¹Lianxin Xie¹Lanlan Yang¹Hongjia Zhao^2*

• ¹The First Clinical Medical College, The Affiliated People’s Hospital of Fujian University of Traditional Chinese Medicine, Fuzhou, Fujian, China, Fuzhou, China
• ²The Affiliated People’s Hospital of Fujian University of Traditional Chinese Medicine, Fuzhou, China, Fuzhou, China

Objective: To help clinical practice rapidly and accurately identify brain tumor types and achieve target detection, this study improves the detection accuracy and model lightweighting of brain tumor MRI image detection based on the YOLO11n algorithm model. Methods: Firstly, 4,000 brain MRI images (glioma, meningioma, pituitary tumor, and non-tumor) from the public Kaggle dataset were annotated and divided into training, validation, and test sets in an 8:1:1 ratio. Secondly, the original YOLO11n model was improved and upgraded through the following ways: (1) integrating lightweight GhostConv modules; (2) embedding Online Convolutional Reparameterization (OREPA) in the C3k2 module; (3) introducing Efficient Multi-scale Attention (EMA) mechanism. Finally, after 200 training epochs, the optimal model was evaluated on both internal and external test sets. Results: The improved model achieved a mean average precision (mAP@50) of 97.2%, which is 2.1 percentage points higher than the original model (95.1%). Simultaneously, the computational load was reduced by 25%, with GFLOPS decreasing from 6.4 to 4.8. Conclusion: The improved YOLO11n model effectively enhances the performance of brain tumor detection while substantially reducing computational complexity, offering efficient and lightweight technical support for real-time clinical diagnostic assistance.