MRI-Based Brain Tumor Detection System Using Pre-Trained YOLOv7 with Transfer Learning, Attention Mechanisms, and Multi-Scale Feature Fusion

MRI-Based Brain Tumor Detection Using YOLOv7 with Transfer Learning and Attention Mechanisms This work presents a deep learning-based system for automated brain tumor detection using MRI scans. The model is built upon the YOLOv7 object detection framework and fine-tuned using transfer learning on annotated brain MRI datasets. The system performs simultaneous tumor classification and localization across glioma, meningioma, and pituitary tumor categories. To enhance performance, the model integrates advanced architectural components including CBAM (attention mechanism), SPPF+ for multi-scale feature aggregation, and BiFPN for efficient feature fusion. These improvements enable better detection of tumors with varying sizes and characteristics. The dataset consists of 51,448 MRI images (including augmented samples), sourced from publicly available datasets. The model achieves strong performance with an accuracy of 97.2%, precision of 97.0%, recall of 96.8%, and F1-score of 96.9%. The project is implemented using PyTorch and includes preprocessing techniques such as noise reduction, contrast enhancement, and morphological refinement. A user-friendly inference interface is provided using Gradio. This system demonstrates the potential of deep learning in computer-assisted diagnosis, supporting radiological workflows through accurate tumor detection and localization. Keywords Brain Tumor Detection, YOLOv7, Transfer Learning, Attention Mechanisms, CBAM, BiFPN, MRI, Deep Learning, Medical Imaging, Object Detection