Brain Tumor Classification (4-Way)

This repository contains a comparative study for 4-class brain tumor classification using MRI images from the BRISC2025 dataset.

Project Goal

Build and compare multiple approaches for the same classification task:

• No Tumor
• Glioma
• Meningioma
• Pituitary Tumor

Current methods included:

• Random Forest (traditional ML with HOG + PCA)
• SVM (traditional ML with HOG + PCA)
• CNN (deep learning, PyTorch)
• Bayes (placeholder notebook currently empty)

Dataset

Reference dataset used in this project:

• BRISC2025 (Brain Tumor Segmentation and Classification)

Expected split used by notebooks:

• Train: 5,000 images
• Test: 1,000 images

Note:

• Dataset files are not committed to this repository.
• Update dataset paths inside notebooks before running.

Clean Project Structure

Brain-Tumor-Classification-4ways/
|-- README.md
|-- requirements.txt
|-- environment.yaml
|-- .gitignore
|-- data/
|   `-- labels/
|       |-- train_labels.csv
|       `-- test_labels.csv
|-- notebooks/
|   |-- bayes/
|   |   `-- bayes.ipynb
|   |-- cnn/
|   |   |-- cnn.ipynb
|   |   `-- dath_cnn.ipynb
|   |-- rf/
|   |   `-- rf.ipynb
|   `-- svm/
|       `-- multi_class_svm.ipynb
`-- docs/
    |-- BRISC-SwinHAFNet.pdf
    |-- final_report_nhom_1.pdf
    `-- final_slide_nhom_1.pdf

Environment Setup

Choose one of the two options below.

Option 1: Conda

conda env create -f environment.yaml
conda activate brain-tumor-4ways

Option 2: pip

python -m venv .venv
# Windows
.venv\Scripts\activate
# macOS/Linux
# source .venv/bin/activate

pip install -r requirements.txt

How To Run

1. Launch Jupyter:

jupyter lab

2. Open and run notebooks by method:

• Random Forest: notebooks/rf/rf.ipynb
• SVM: notebooks/svm/multi_class_svm.ipynb
• CNN: notebooks/cnn/dath_cnn.ipynb
• Bayes: notebooks/bayes/bayes.ipynb (currently empty)

3. Ensure dataset paths are updated in each notebook cell before training/evaluation.

Notes

• Some notebooks were originally authored in Google Colab and may include Colab-specific imports.
• If you run locally, remove or skip Colab-only cells (google.colab, Drive mount, etc.).
• Results can vary depending on random seed, hardware, and preprocessing setup.

Team Deliverables

• Technical paper: docs/BRISC-SwinHAFNet.pdf
• Final report: docs/final_report_nhom_1.pdf
• Final slides: docs/final_slide_nhom_1.pdf

License

No explicit license is currently defined in this repository.