BRAC University2025 – Dec 2025
Brain Tumor Segmentation & Classification
Course project: U-Net variants for brain MRI segmentation and tumor classification.
My own course project for CSE428 Image Processing at BRAC University (Fall 2025). I trained and compared U-Net, Attention U-Net, Attention U-Net + BiFPN and a joint segmentation–classification Attention U-Net in PyTorch on BRISC2025 (6,000 T1-weighted brain MRI slices, 4 classes). Separately trained models beat the joint model on both tasks.
Highlights
- Compared a standard preprocessing pipeline with CLAHE, bilateral filtering and z-score normalization, with and without augmentation.
- Separately trained models beat the joint segmentation–classification model on both tasks.
- Ownership
- Solo course project
- Stack
- PyTorchtorchvision