TumorVision

Automated 3D brain tumor segmentation from multi-modal MRI using Swin-UNETR, with interactive visualization.

Overview

Trained on the BraTS 2021 dataset to segment three tumor sub-regions — Tumor Core (TC), Whole Tumor (WT), and Enhancing Tumor (ET) — from 4-channel MRI input (T1, T1ce, T2, FLAIR).

Features

• 🧠 Swin-UNETR transformer architecture via MONAI
• 🎯 Multi-class segmentation: TC, WT, ET
• 📊 Sliding window inference on full 240×240×155 volumes
• 🌐 Interactive 3D neon tumor mesh visualization
• 🖥️ Gradio web UI for one-click inference

Results

Tech Stack

Python · PyTorch · MONAI · PyTorch Lightning · Gradio · Plotly

Setup

git clone https://github.com/yourusername/tumor-vision
cd tumor-vision
pip install -r requirements.txt

Usage

# Train
python train_brats.py

# Run web UI
python app.py

Dataset

BraTS 2021

import kagglehub
path = kagglehub.dataset_download(
    "dschettler8845/brats-2021-task1"
)
print("Path to dataset files:", path)