Classification-Segmentation-Project

This repository contains our CSE428 course project on brain tumor detection, combining:

🌐 Live Demo (Streamlit)

We deployed the multitask U-Net model online using Streamlit + Hugging Face model hosting:

🔗 BriscApp: https://classification-segmentation-project-soicmh9wwa5dqkbwswufrq.streamlit.app/

---

• Tumor type classification (4-class)
• Tumor segmentation (pixel-level mask prediction)

We built a multitask deep learning pipeline using U-Net and evaluated it against Attention U-Net for segmentation.

---

Dataset

We used the BRISC 2025 Brain Tumor Dataset from Kaggle:

🔗 https://www.kaggle.com/datasets/briscdataset/brisc2025

Classes:

• No Tumor
• Glioma
• Meningioma
• Pituitary

---

Models Used

1) Multitask U-Net (Segmentation + Classification)

A standard U-Net backbone with:

• Segmentation head → predicts tumor mask
• Classification head → predicts tumor type (4-class)

2) Attention U-Net (Segmentation only)

Used as a segmentation baseline to compare performance.

---

🔗 Hugging Face Model: https://huggingface.co/MihirDas/brisc-unet
(contains unet_multitask_4cls_best.pth)

---

Team Members

• Mihir Das - https://github.com/MIhirDas10 (me)

• Digonta Das - https://github.com/DigontaDas

• Hasnain Ashraf - https://github.com/Hasu121

---