NeuroSight

NeuroSight is a Streamlit application for classifying brain MRI slices into four classes with an EfficientNet-B0 model and Grad-CAM visual explanations.

This project was built by Team ThinkNova as part of the Samsung Innovation Campus AI Track.

What the app does

• Classifies MRI slices as Glioma, Meningioma, Pituitary, or No Tumor
• Generates Grad-CAM heatmaps to highlight model focus areas
• Displays class probabilities and simple clinical notes
• Exports a plain-text report from the Streamlit UI

The training notebook reports 97.86% test accuracy for the EfficientNet-B0 model used by the app.

Repository contents

• app.py: Streamlit inference app
• style.css: app styling
• neurosight_logo.png: branding asset
• notebooks/: training and experimentation notebooks
• requirements.txt: runtime dependencies for the app
• requirements-notebooks.txt: extra packages for notebook work
• LICENSE: repository license

Model files

The repository does not include trained weights or dataset files.

Download the model files here:

https://drive.google.com/drive/folders/1EOH-s1Iv_wDkRwUxAzm6B9MPb-1eZdjc?usp=sharing

Required for the app:

• last_model.pth

Optional for notebook training resume:

• last_optimizer.pth

Place last_model.pth in the project root before running the app.

Quick start

git clone <your-repo-url>
cd NeuroSight
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
streamlit run app.py

If you want to run the notebooks as well:

pip install -r requirements-notebooks.txt

Project structure

NeuroSight/
|-- app.py
|-- style.css
|-- neurosight_logo.png
|-- requirements.txt
|-- requirements-notebooks.txt
|-- LICENSE
|-- notebooks/
|   |-- Capstone_Project_Brain_Tumor_Classification_Part1.ipynb
|   `-- Capstone_Project_Brain_Tumor_Classification_Part2.ipynb
`-- .gitignore

Notes

• This project is intended for educational and research use.
• It is not a medical device and must not be used for clinical diagnosis or patient-care decisions.
• Notebook dependencies are broader than app runtime dependencies because the notebooks include training, dataset handling, and evaluation workflows.

Team

• Nadia Hafhouf
• Mohamed Dhia Chaouachi
• Mohammed Aziz Mhenni
• Amira Ouechtati
• Mariem Jlassi
• Mohamed Ayhem Zamouri

License

See LICENSE.