AI-based object detection for brain tumour using YOLO model

Brain tumour detection in MRI is still hard because tumours look very different, good labels are limited, and clinics need fast results. The purpose of this study is to evaluate whether a lightweight Deep learning-based YOLOv8 object detection pipeline can deliver reliable brain tumour classification with coarse localisation when only image-level labels are available. Using the public Kaggle Brain Tumour MRI Dataset (glioma, meningioma, pituitary tumour, and no tumour), we convert classification labels into weak bounding boxes and fine-tune a YOLOv8n detector. We compare a baseline YOLOv8n model against an enhanced version that adds CBAM attention and an Inner GIoU-based IoU loss. Performance is reported using precision, recall, F1 score, and mAP at IoU 0.5 and 0.5 to 0.95, alongside GPU latency. The baseline achieves of 0.991 and F1 of 0.953. The CBAM plus IoU loss variant improves throughput (10.88 ms per image vs. 12.90 ms), but reduces accuracy metrics (of 0.974). Overall, the results show that Brain Tumour Object Detection with YOLOv8 under weak supervision can work as a fast slice-level classifier, while localisation-focused add-ons such as CBAM and IoU loss need tighter spatial labels to consistently improve accuracy.