YOLO-Scan: A Real-Time Deep Learning Framework for Brain Tumor Detection in MRI

Brain tumor poses a serious threat to global health, affecting individuals across various age groups. Early and accurate detection is crucial for improving patient outcomes and survival rates. While conventional diagnostic methods are effective, they often struggle with issues related to precision and speed. The advancement of Artificial Intelligence, particularly through Deep Learning, has opened new avenues for enhanced analysis of medical imaging. This paper introduces the improved YOLOv8 and the novel YOLOv11 models for brain tumor detection in MRI scans distinguished by its dual-stage instance segmentation and detection architecture. These models employ advanced object detection techniques to enhance precision, particularly in identifying small or overlapping tumor. YOLOv11 builds upon YOLOv8 by incorporating advanced feature extraction strategies and optimization methods. YOLOv11 employs advanced Multi-Scale Dilate Attention (MSDA) and GSConv modules, substantially improving feature extraction for small and overlapping tumors. The framework achieves real-time inference at 62 frames per second (FPS) on an NVIDIA RTX 2060 GPU, supported by TensorRT optimization that reduces memory usage by 18%, ensuring deployment efficiency. YOLOv11 achieves 0.91 F1-score, 0.90 precision, and 0.91 recall across glioma, meningioma, and pituitary tumors. The models presented here surpass previous approaches in terms of detection speed and reliability. They show promising results in brain tumor detection and could be adapted for other medical imaging applications. In summary, this study provides a valuable contribution to clinical diagnostics by offering a faster and more accurate tool for brain tumor detection.