🧠 TumorNet — Brain Tumor Classifier

TumorNet is a deep learning project that uses a Convolutional Neural Network (CNN) to detect and classify brain MRI scans into four categories:
Glioma, Meningioma, No Tumor, and Pituitary.

The model is trained on the publicly available Brain Tumor MRI Dataset from Kaggle.

---

🚀 Features

• 🩺 Automatic tumor detection from MRI scans
• 🧠 Multi-class classification: Glioma, Meningioma, No Tumor, Pituitary
• 💡 User-friendly Streamlit app for quick predictions
• 🧩 Trained model (.h5 / .keras) for immediate inference
• ⚙️ End-to-end pipeline: Training → Evaluation → Deployment

---

🧱 Folder Structure

TumorNet/
│
├── venv/ # Virtual environment (excluded via .gitignore)
│
├── Training/ # Contains training images
├── Testing/ # Contains testing images
│
├── app.py # Streamlit web app for predictions
├── Brain_Tumor_Classifier.ipynb # Jupyter notebook for model training
│
├── Brain Tumors.h5 # Trained model (HDF5 format)
├── Brain Tumors.keras # Keras model format
│
├── requirements.txt # Project dependencies
├── .gitignore # Files/folders to ignore
└── README.md # Project documentation

---

🧩 Model Overview

• Architecture: Convolutional Neural Network (CNN)
• Framework: TensorFlow / Keras
• Optimizer: Adamax (lr=0.001)
• Loss Function: Categorical Crossentropy
• Accuracy: ~93% overall, with balanced class performance

---

🧠 How It Works

1. The user uploads an MRI image through the Streamlit web interface.
2. The app preprocesses the image (resizing to 224×224, normalizing).
3. The CNN model (Brain Tumors.h5) predicts the tumor type.
4. The app displays both the predicted class and class probabilities.

---

🖼️ Dataset Information

Dataset used:

Brain Tumor MRI Dataset — Kaggle Link

• Classes: Glioma, Meningioma, Pituitary, No Tumor
• Images: 5,500+ labeled MRI scans
• Format: JPEG
• Split: 80% training, 20% testing

---

🛠️ Installation & Setup

1️⃣ Clone the repository

git clone https://github.com/Sahil0015/TumorNet.git
cd TumorNet

2️⃣ Create and activate virtual environment

python -m venv venv
venv\Scripts\activate      # On Windows
# source venv/bin/activate # On Mac/Linux

3️⃣ Install dependencies

pip install -r requirements.txt

4️⃣ Run the Streamlit app

streamlit run app.py

5️⃣ Open in browser

Once the server starts, open:

http://localhost:

Upload any MRI image and get instant predictions.

---

🧠 Training the Model

If you don't have the pre-trained model file (Brain Tumors.h5 or Brain Tumors.keras), you can easily train it from scratch by following these steps:

1. Download the dataset
   Get the dataset from Kaggle:
   👉 Brain Tumor MRI Dataset

2. Prepare the dataset folders
   After downloading, extract the dataset and place the following folders inside your project directory:

   Training/
   Testing/
   

3. Train the model using the notebook
   Open and run all cells in the Jupyter notebook file:

   Brain_Tumor_Classifier.ipynb
   

   This will automatically train the CNN model and save it as:

   Brain Tumors.h5
   

   or

   Brain Tumors.keras
   

4. Run the Streamlit app
   Once training is complete, you can launch the app and test the model using:

   streamlit run app.py

💡 Tip: Training may take several minutes depending on your hardware. For faster results, you can use a GPU runtime on Google Colab.

---

🧰 Technologies Used

Category	Libraries / Tools
Language	Python
Deep Learning	TensorFlow, Keras
Visualization	Matplotlib, Seaborn
Data Handling	NumPy, Pandas
Frontend	Streamlit
Model Persistence	H5, Keras

---

📜 .gitignore Notes

venv/
__pycache__/
*.h5
*.keras
*.zip
.DS_Store
.ipynb_checkpoints/

The large model files (.h5, .keras) can be excluded from Git if you plan to share your repo publicly — you can always re-upload them via release or Google Drive.

---

🧩 Results Summary

Metric	Value
Accuracy	~93%
Precision	0.92
Recall	0.93
F1-Score	0.93
Dataset Size	~5500 images

✅ The model demonstrates balanced performance across all tumor categories.

---

⚙️ Future Improvements

• 🔧 Add Grad-CAM for model explainability
• 🌐 Deploy app using Streamlit Cloud / Hugging Face Spaces
• 🧮 Experiment with EfficientNet / MobileNet architectures
• 📱 Create a lightweight mobile-compatible version

---

⚖️ Disclaimer

This project is for educational and research purposes only.
It is not a substitute for professional medical diagnosis or advice.
Always consult a qualified radiologist or medical expert for clinical interpretation.

---

👨‍💻 Author

Sahil Aggarwal
📂 GitHub
🔗 LinkedIn
📧 sahilaggarwal1532003@gmail.com

---

📝 License

This project is licensed under the MIT License — feel free to use, modify, and share with attribution.