Assessment of YOLO Variants for Multi-Type Brain Tumor Detection

Advancements in computer vision and AI technologies have brought significant changes to healthcare technologies. Classification and localization of brain tumors using various versions of the YOLO (You Look Only Once) models provide a better solution for improving diagnostic accuracy in medical imaging along with an Interpretable Grad-CAM (Gradient-weighted Class Activation Mapping) Technique. YOLO, a real-time object detection framework, effectively detects and categorizes tumors in brain MRI scans while identifying their locations. With the help of different YOLO versions, the following are YOLOv5, YOLOv8, YOLOv9, YOLOv10, and YOLOv11. The YOLO framework can better identify multiple types of tumors with higher accuracy, speed, and reliability. Every YOLO version includes improvements such as better feature extraction, spatial perception, and optimized neural networks, allowing accurate tumor location. This approach makes the diagnosis process much easier and provides radiologists with real-time automated analysis to enhance the patient’s care. Comparing different versions of YOLO shows the compromise between detection accuracy and time, proving the high potential of deep learning for detecting brain tumors. The comparative analysis indicates that YOLOv11 achieves the highest performance, with a mAP@0.5 of 97%. This top-performing model was further analyzed using Grad-CAM based Eigen-CAM technique to generate activation maps for each class of brain tumor in the MRI images.