🧠 TumorScope – Brain Tumor Classification with Deep Learning & Gemini AI

---

📖 Overview

TumorScope is a web application that classifies brain MRI scans into four categories:

• Glioma
• Meningioma
• Pituitary Tumor
• No Tumor

The app uses:

• Xception (Transfer Learning)
• Custom CNN Model
• Saliency Maps for interpretability
• Gemini 1.5 Flash (Google AI) to generate concise medical explanations for predictions

The project is deployed on Streamlit Cloud and provides an interactive user experience for radiologists, researchers, and students.

---

✨ Key Features

• Upload Brain MRI Images for classification
• Dual-Model Selection: Transfer Learning (Xception) or Custom CNN
• Saliency Map Generation to visualize important image regions
• AI-Powered Explanations via Gemini 1.5 Flash
• Interactive Probability Charts with Plotly
• Secure Deployment on Streamlit Cloud

---

🛠 Tech Stack

Component	Technology
Frontend	Streamlit
Model Training	TensorFlow / Keras
Explainability	Gemini 1.5 Flash
Visualization	OpenCV, Plotly
Deployment	Streamlit Cloud

---

🚀 Getting Started

1️⃣ Clone the Repo

git clone https://github.com/rahatmoktadir03/tumor-scope.git
cd tumor-scope

2️⃣ Install Dependencies

pip install -r requirements.txt

3️⃣ Add Gemini API Key

• Create a .env file in the project root:

  GOOGLE_API_KEY=your-gemini-api-key
  

4️⃣ Run Locally

streamlit run streamlit_app.py

---

📊 Models and Accuracy

• Xception (Transfer Learning) – ~99% accuracy
• Custom CNN – ~98% accuracy
• Saliency Maps highlight critical regions for model interpretability

---

🔮 Future Improvements

• Chat with MRI scan (using multimodal LLMs)
• Compare multiple models side-by-side
• Generate detailed diagnostic reports for doctors