Brain Tumor MRI Analysis System
I built an AI-powered medical image analysis system for my thesis that detects brain tumors in MRI scans using YOLOv8 deep learning. The project covers the full pipeline — from dataset analysis and model training to a live Flask web application.
Two datasets were used: a multi-class dataset (glioma, meningioma, pituitary) with ~3,000 annotated images and a multimodal tumor dataset with ~4,200 images. Training was done locally on a dedicated machine using PyTorch with GPU acceleration, supporting YOLOv8, YOLOv9, and YOLOv11 architectures with configurable epochs, batch size, early stopping, and checkpoint saving.
The resulting best. pt model was integrated into a Flask web app where users can drag-and-drop or upload MRI images (JPG, PNG, BMP, TIFF) and receive real-time inference results with detected tumor regions, bounding boxes, confidence scores, and class labels.
The training pipeline includes dataset statistics reporting, confusion matrices, and result curve visualizations.
Technologies
Project Features
- YOLOv8 object detection model trained on brain MRI tumor datasets (glioma, meningioma, pituitary)
- Support for YOLOv8, YOLOv9, and YOLOv11 architectures with GPU/CPU auto-detection
- Flask web app with drag-and-drop MRI image upload and real-time inference
- Tumor region bounding boxes with class labels and per-detection confidence scores
- Complete training pipeline: dataset analysis, multi-epoch training, early stopping, and checkpoint saving
- Training visualizations including confusion matrices, normalized confusion matrices, and result curves
- Trained on ~7,000+ annotated MRI images across two datasets with YOLO annotation format
