Brain Tumor MRI Analysis System

A web application that analyzes MRI scans to detect and classify brain tumors, providing AI-driven treatment recommendations.

Features

✅ MRI Scan Analysis – Upload and analyze brain MRI scans.
✅ Tumor Classification – Detects and classifies three types of brain tumors:

• Glioma
• Meningioma
• Pituitary
  ✅ AI-Powered Recommendations – Provides treatment insights using Google Gemini AI.
  ✅ Confidence Scoring – Displays model prediction confidence.
  ✅ User-Friendly Interface – Drag-and-drop upload functionality for easy interaction.

---

Contributors

• Hariom Khonde
• Pranav Gudagi
• Abhishek Hiremath

Setup Instructions

1. Clone the Repository

git clone https://github.com/dchaudhari7177/TumerDetect.git
cd TumerDetect

2. Install Dependencies

npm install

3. Download the Model File

Since GitHub restricts large file uploads, download the trained model (model_v11.h5) manually and place it in the project directory.

📌 Download Link: Google Drive - model_v11.h5

➡️ Place the downloaded file in: TumerDetect/

4. Add API Key for Gemini AI

Edit the following files and insert your Google Gemini API key:

• src/pages/ImageAnalysis.tsx
• src/pages/GeminiApi.js

5. Start the Application

npm run dev

6. Start the Backend (Flask Server)

Run both backend servers as follows:

Image-Based Recognition (Predict.py - Port 5001)

python predict.py

Gene-Based Analysis (App.py - Port 5000)

python app.py

---

Tech Stack

🖥 Frontend: React (TypeScript), Tailwind CSS
🧠 AI Model: TensorFlow/Keras (EfficientNetB4)
⚙ Backend: Flask (Python)
🔍 Deep Learning Features: Grad-CAM, Image Segmentation (U-Net), Feature Extraction (ResNet50), Gene Analysis (XGBoost)

---

Project Notebooks

📓 Jupyter Notebooks:

• main.ipynb - Contains the complete model training pipeline including data preprocessing, augmentation, and model evaluation
• module2.ipynb - Exploratory data analysis, visualization, and dataset statistics

---

Dataset

The model was trained on the Brain Tumor Multimodal Image Dataset containing both CT and MRI scans.

🔗 Dataset Source: Kaggle - Brain Tumor Multimodal Image Dataset

📊 Dataset Details:

• Contains over 3,000 MRI and CT scan images
• Multiple tumor types: Glioma, Meningioma, Pituitary
• Verified and labeled by medical professionals
• Includes various angles and cross-sections
• High-quality medical imaging data

---

Machine Learning Models

Primary Classification Model

• Architecture: EfficientNetB4 (Transfer Learning)
• Dataset: Brain Tumor MRI Dataset (3,000+ images)
• Performance:
  • Training Accuracy: 98.5%
  • Validation Accuracy: 97.8%
  • Test Accuracy: 97.2%
  • F1 Score: 0.968

📊 Class-wise Accuracy
✔ Glioma: 98.1%
✔ Meningioma: 97.5%
✔ Pituitary: 96.8%

Training Details

• Epochs: 30
• Batch Size: 32
• Optimizer: Adam
• Learning Rate: 0.0001
• Cross-Validation: 5-fold (Avg: 97.4%)

Supporting Models

• U-Net – Skull stripping & segmentation
• ResNet50 – Feature extraction for ROI detection
• XGBoost – Genetic marker-based tumor prediction

Model Capabilities

✅ Multi-class classification (Glioma, Meningioma, Pituitary, No Tumor)
✅ Real-time Image Augmentation
✅ Grad-CAM Visualizations for interpretability
✅ Ensemble Learning for higher accuracy

---

How to Use

1️⃣ Upload an MRI scan image
2️⃣ Click "Start Analysis"
3️⃣ View Tumor Classification Results
4️⃣ Check AI-Generated Treatment Recommendations

5️⃣ Genetic Marker-based Prediction with Confidence Score

5️⃣ Genetic Marker-based Prediction with Confidence Score

Images Gallery

• 
•