N. Naga Raju, Kankanala Srinivas, Chilukamari Rajesh, Balaram Murthy Chintakindi
Accurate identification of brain tumor regions in MRI scans will help in early diagnosis and further treatment line. But difficulties like small tumor size, low contrast, and inconsistent morphology often complicate detection accuracy within real-time clinical practice. Here, we introduce Y O L O β 11 , a state-of-the-art deep learning-based model developed over the YOLOv11 framework to increase detection performance at low computational costs. The new model presents a new TwinFormer module that captures global and local attention mechanisms for enhancing multi-scale feature representation. It also incorporates a Segmented Cross-Flow Stage (SCFStage) to maintain semantic-spatial features for better tumor localization. In addition, a combination loss function is introduced based on CIOU loss, focal loss and BCE loss to improve training strength and stability.The model was tested on the BR35H brain tumor MRI dataset and Real Brain Tumor dataset. The Y O L O β 11 model demonstrated enhanced and state-of-the-art performance by recording a 0.936 Precision, 0.927 Recall, mAP@50 of 0.950, and mAP@50:95 of 0.744 on BR35H. These results indicate the robustness of Y O L O β 11 in dealing with complex tumor patterns, thus proving its superiority over existing methods for precise real-time brain tumor detection in MRI scans.