🧠 Glioblastoma Detection Using Deep Learning

A deep learning model for detecting glioblastoma brain tumors from MRI scans using TensorFlow and Keras. This project implements a Convolutional Neural Network (CNN) to analyze brain MRI images and identify the presence of glioblastoma tumors with high accuracy.

📋 Table of Contents

• Overview
• Dataset
• Model Architecture
• Performance
• Requirements
• Installation
• Usage
• Project Structure
• Model Details
• Results
• License

🔍 Overview

This project implements a deep learning solution for detecting glioblastoma in brain MRI scans. The model is designed to assist medical professionals in early tumor detection and classification.

Key Features

• Advanced CNN architecture optimized for medical imaging
• Data augmentation to improve model robustness
• Confidence scoring for predictions
• High accuracy and AUC metrics
• Support for grayscale MRI images (256x256x1)

📊 Dataset

The project uses the MRI Glioma Dataset for Tumor Grade Classification by Kolar Luni, available on Kaggle. The dataset includes:

• 9,832 MRI brain tumor images
• Image dimensions: 256x256x1 (grayscale)
• All images are pre-processed and standardized

🏗️ Model Architecture

The CNN architecture consists of:

- Input Layer (256x256x1)
- 3 Convolutional Blocks:
  └─ Each block contains:
     - 2x Conv2D layers with ReLU
     - BatchNormalization
     - MaxPooling
     - Dropout (0.25)
- Dense Layers:
  - Flatten
  - Dense (512) + BatchNorm + Dropout
  - Dense (256) + BatchNorm + Dropout
  - Output Dense (1) with Sigmoid

📈 Performance

• Accuracy: 95.73%
• AUC: 97.87%
• Loss: 0.1116

💻 Requirements

tensorflow>=2.0.0
pillow
numpy
pandas
scikit-learn
tqdm
matplotlib
seaborn

🚀 Installation

1. Clone the repository

git clone https://github.com/internetmachinebroke/CancerPred.git
cd CancerPred

2. Install dependencies

pip install -r requirements.txt

3. Prepare your data

# Place your MRI images in the GT_path directory
mkdir GT_path
# Copy your images to GT_path/

🔧 Usage

To train the model:

python main.py

To make predictions on new images:

from predictor import predict_single_image
from tensorflow.keras.models import load_model

model = load_model('best_model.keras')
result = predict_single_image(model, 'path_to_image.jpg')
print(f"Tumor detected: {'Yes' if result['has_tumor'] else 'No'}")
print(f"Confidence: {result['confidence']:.2f}%")

📁 Project Structure

CancerPred/
├── GT_path/              # Directory for MRI images
├── main.py              # Main execution script
├── data_loader.py       # Data loading and augmentation
├── data_processor.py    # Data preprocessing
├── model.py            # CNN model architecture
├── trainer.py          # Model training logic
├── visualizer.py       # Training visualization
├── predictor.py        # Image prediction
└── requirements.txt    # Project dependencies

🔬 Model Details

The model employs several key techniques:

• Data augmentation for improved generalization
• Batch normalization for training stability
• Dropout layers to prevent overfitting
• Early stopping and learning rate reduction
• Confidence scoring for predictions

Training Parameters

• Batch Size: 32
• Initial Learning Rate: 0.001
• Optimizer: Adam
• Loss Function: Binary Cross-entropy
• Metrics: Accuracy, AUC

📊 Results

The model achieves:

• High accuracy in tumor detection (95.73%)
• Excellent discrimination ability (AUC: 97.87%)
• Low loss value (0.1116)
• Robust performance on validation data

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• Dataset provided by Kolar Luni on Kaggle

---

Note: This model is intended for research purposes only and should not be used as the sole means of medical diagnosis.