The MGH-YOLO model, built from the YOLOv8 design, significantly improves brain tumour detection in MRI images. By integrating complex modules, including Ghost Convolution (GhostConv), C3Ghost and the Spatial Pyramid Pooling Efficient Layer Aggregation Network (SPPELAN), the suggested model performs better while maintaining processing efficiency. Integrating an x-small object recognition layer and a multi-scale feature fusion framework considerably refined its capacity to identify minute tumour areas. Through an assessment of the Br35H dataset, MGH-YOLO revealed a 1.2% increase in accuracy and a substantial 4.7% and 0.7% improvement in mAP(0.50) and mAP(0.50–0.95), respectively, over the baseline YOLOv8x. Further, across all criteria, it beat other cutting-edge models, such as BGF-YOLO and DAMO-YOLO-L*. With its revolutionary architecture, containing a new fourth detection head and attention-enhanced pooling, MGH-YOLO delivers a robust, resource-efficient key to brain tumour identification, demonstrating its promise for real-time applications in medical imaging.