🧠 AI-Powered Brain Tumor Classification

Leveraging Convolutional Neural Networks (CNNs) and Transfer Learning to classify brain MRI images into four categories: Glioma, Meningioma, Pituitary Tumor, and No Tumor. This project achieves up to 97% accuracy using ResNet50V2, supporting early diagnosis and clinical decision-making.

---

📋 Table of Contents

• Features
• Model Performance
• Dataset
• Project Structure
• Installation
• Usage
• Technical Details
• Results
• Environment Variables
• Resources
• Contributing
• License

---

✨ Features

• Multi-Model Architecture: Implements and compares 5 different deep learning models

  • Custom CNN
  • VGG16 (Transfer Learning)
  • VGG19 (Transfer Learning)
  • ResNet50V2 (Transfer Learning)
  • MobileNetV2 (Transfer Learning)

• High Accuracy: Achieves up to 97% accuracy on test data with ResNet50V2

• Comprehensive Evaluation:

  • Accuracy, Precision, Recall, F1-Score metrics
  • Confusion matrices for detailed performance analysis
  • Training/validation loss and accuracy curves

• Data Augmentation: Robust preprocessing pipeline for better generalization

• Class Balancing: Automatic downsampling to handle class imbalance

• Visualization Tools:

  • Sample image visualization from training/testing sets
  • Model comparison charts
  • Training history plots

---

🎯 Model Performance

Model	Test Accuracy	Precision	Recall	F1-Score	Loss
ResNet50V2	96.98%	97.03%	96.91%	96.95%	0.1647
MobileNetV2	96.47%	96.49%	96.47%	96.45%	0.2659
VGG19	94.76%	94.85%	94.76%	94.67%	0.3124
VGG16	94.49%	94.70%	94.27%	94.30%	0.3187
Custom CNN	88.61%	88.86%	88.53%	87.63%	0.7931

Best Model: ResNet50V2 with 96.98% test accuracy 🏆

---

📊 Dataset

Sources

The project uses two Kaggle datasets:

1. Brain Tumor Classification (MRI) by Sartaj Bhuvaji

   • Dataset Link

2. Brain Tumor MRI Dataset by Masoud Nickparvar

   • Dataset Link

Dataset Structure

Training/
├── glioma_tumor/
├── meningioma_tumor/
├── no_tumor/
└── pituitary_tumor/

Testing/
├── glioma_tumor/
├── meningioma_tumor/
├── no_tumor/
└── pituitary_tumor/

Classes

• Glioma Tumor: Malignant brain tumors arising from glial cells
• Meningioma Tumor: Tumors arising from meninges (brain coverings)
• Pituitary Tumor: Tumors in the pituitary gland
• No Tumor: Normal brain MRI scans

Preprocessing

• Images resized to 224x224 pixels
• Normalization: pixel values scaled to [0, 1]
• Class balancing through downsampling
• Batch size: 32

---

📁 Project Structure

BRAIN-TUMOR-CLASSIFICATION/
│
├── src/
│   └── Models/
│       ├── CNN.ipynb              # Custom CNN implementation
│       ├── MobileNet.ipynb        # MobileNetV2 transfer learning
│       ├── ResNet.ipynb           # ResNet50V2 transfer learning
│       ├── VGG_16.ipynb           # VGG16 transfer learning
│       └── VGG_19.ipynb           # VGG19 transfer learning
│
├── results/
│   ├── CNN_Result/
│   │   └── CNN.png                # CNN performance visualization
│   ├── MobileNet_Result/
│   │   └── MobileNet.png
│   ├── ResNet_Result/
│   │   └── ResNet.png
│   ├── VGG_16_Result/
│   │   └── VGG_16.png
│   ├── VGG_19_Result/
│   │   └── VGG_19.png
│   ├── Comparison_Models/
│   │   ├── Comparison_1.png      # Model accuracy comparison
│   │   ├── Comparison_2.png      # Validation loss comparison
│   │   ├── Comparison_3.png      # Training curves
│   │   ├── Comparison_4.png      # Bar charts
│   │   ├── Comparison_5.png      # Horizontal comparison
│   │   ├── Comparison_6.png      # Final rankings
│   │   └── Comparison_7.png      # Detailed metrics
│   └── Data_Visualization/
│       ├── Testing_data.png       # Sample test images
│       ├── Training_data.png      # Sample training images
│       └── validtion_data.png     # Merged dataset samples
│
├── brain-tumor-classification-cnn-97-acc.py   # Main training script
├── models.txt                     # Links to trained models (.h5 files)
├── dataset.txt                    # Dataset download links
├── kaggle_notebook.txt            # Link to Kaggle notebook
├── requirements.txt               # Python dependencies
├── .env.example                   # Environment variables template
├── .gitignore                     # Git ignore rules
├── LICENSE                        # Project license
└── README.md                      # This file

---

🚀 Installation

Prerequisites

• Python 3.10 or higher
• CUDA-compatible GPU (recommended for faster training)
• 8GB+ RAM
• 10GB+ free disk space

Installation Steps

1. Install Python using MiniConda

Download and install MiniConda from here

Create a new environment:

conda create -n Brain_Tumor_Classification python=3.10

Activate the environment:

conda activate Brain_Tumor_Classification

2. Clone the Repository

git clone https://github.com/Abdelhady-22/brain_tumor_classification.git
cd brain-tumor-classification

3. Install Dependencies

pip install -r requirements.txt

4. Download Dataset

Option A: Using Kaggle API

pip install kagglehub
python -c "import kagglehub; kagglehub.dataset_download('sartajbhuvaji/brain-tumor-classification-mri')"
python -c "import kagglehub; kagglehub.dataset_download('masoudnickparvar/brain-tumor-mri-dataset')"

Option B: Manual Download

• Download datasets from links in dataset.txt
• Extract to project directory

5. Setup Environment Variables

cp .env.example .env

Edit .env file:

PROJECT_VERSION=1.0
IMAGE_SIZE=224
BATCH_SIZE=32
EPOCHS=20
LEARNING_RATE=0.0001

---

💻 Usage

Training Models

Run the main training script:

python brain-tumor-classification-cnn-97-accuracy.ipynb

Or train individual models using Jupyter notebooks:

jupyter notebook src/Models/CNN.ipynb

Model Inference

from tensorflow.keras.models import load_model
import numpy as np
from PIL import Image

# Load trained model
model = load_model('ResNet_best_model.keras')

# Load and preprocess image
img = Image.open('path/to/mri_scan.jpg')
img = img.resize((224, 224))
img_array = np.array(img) / 255.0
img_array = np.expand_dims(img_array, axis=0)

# Make prediction
prediction = model.predict(img_array)
classes = ['glioma_tumor', 'meningioma_tumor', 'no_tumor', 'pituitary_tumor']
result = classes[np.argmax(prediction)]

print(f"Prediction: {result}")
print(f"Confidence: {np.max(prediction) * 100:.2f}%")

Evaluating Models

# Load test data
from tensorflow.keras.preprocessing.image import ImageDataGenerator

test_datagen = ImageDataGenerator(preprocessing_function=lambda x: x/255.0)
test_generator = test_datagen.flow_from_directory(
    'path/to/Testing',
    target_size=(224, 224),
    batch_size=32,
    class_mode='categorical'
)

# Evaluate model
results = model.evaluate(test_generator)
print(f"Test Accuracy: {results[1]*100:.2f}%")

---

🔧 Technical Details

Architecture Overview

Custom CNN

Input (224x224x3)
→ Conv2D(64) + MaxPool
→ Conv2D(64) + MaxPool
→ Conv2D(128) + MaxPool
→ Flatten + Dropout(0.4)
→ Dense(128) + Dense(64) + Dense(64)
→ Output (4 classes, softmax)

Transfer Learning Models

• Base Models: Pre-trained on ImageNet
• Frozen Layers: All base layers non-trainable
• Custom Head:
  • GlobalAveragePooling2D / Flatten
  • Dense(256, relu)
  • Output(4, softmax)

Training Configuration

Parameter	Value
Image Size	224 × 224
Batch Size	32
Epochs	20 (with early stopping)
Optimizer	Adam
Learning Rate	0.0001
Loss Function	Categorical Crossentropy

Callbacks

• EarlyStopping: Patience=5, monitors val_loss
• ModelCheckpoint: Saves best model based on val_loss
• ReduceLROnPlateau: Reduces LR by 0.5 when val_loss plateaus

Hardware Requirements

• GPU: NVIDIA GPU with CUDA support (recommended)
• RAM: 8GB minimum, 16GB recommended
• Storage: 10GB for datasets and models

---

📈 Results

Model Comparison

Training History

All models were trained for up to 20 epochs with early stopping:

• ResNet50V2: Best validation loss at epoch 12
• MobileNetV2: Best validation loss at epoch 15
• VGG19: Best validation loss at epoch 18
• VGG16: Best validation loss at epoch 16
• Custom CNN: Best validation loss at epoch 19

Key Findings

1. Transfer Learning Superiority: Pre-trained models significantly outperform custom CNN
2. ResNet50V2 Excellence: Achieved best overall performance with 97% accuracy
3. Convergence Speed: Transfer learning models converge faster (fewer epochs)
4. Generalization: All models show good generalization on merged dataset

---

🌐 Environment Variables

Create a .env file with the following variables:

# Project Configuration
PROJECT_NAME=Brain_Tumor_Classification
PROJECT_VERSION=1.0

# Model Parameters
IMAGE_SIZE=224
BATCH_SIZE=32
EPOCHS=20
LEARNING_RATE=0.0001

# Paths
TRAIN_PATH=/kaggle/input/brain-tumor-mri-dataset/Training
TEST_PATH=/kaggle/input/brain-tumor-mri-dataset/Testing
MODEL_SAVE_PATH=/kaggle/working/

# Training Configuration
EARLY_STOPPING_PATIENCE=5
REDUCE_LR_PATIENCE=3
REDUCE_LR_FACTOR=0.5
MIN_LEARNING_RATE=1e-7

# Callbacks
MONITOR_METRIC=val_loss
SAVE_BEST_ONLY=True

---

📚 Resources

Trained Models

All trained models (.h5 and .keras formats) are available in models.txt:

• Custom CNN Model
• VGG16 Model
• VGG19 Model
• ResNet50V2 Model
• MobileNetV2 Model

Kaggle Notebook

Complete interactive notebook: See kaggle_notebook.txt for link

Datasets

Download links available in dataset.txt

Documentation

• TensorFlow Documentation
• Keras Applications
• Brain Tumor Types

---

🤝 Contributing

Contributions are welcome! Please follow these steps:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/AmazingFeature)
3. Commit your changes (git commit -m 'Add some AmazingFeature')
4. Push to the branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

Development Guidelines

• Follow PEP 8 style guide for Python code
• Add docstrings to functions and classes
• Update README.md for significant changes
• Test code before submitting PR

---

📄 License

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

---

👥 Authors

• Abdelhady Ali - Initial work - MYGitHub

---

🙏 Acknowledgments

• Datasets: Thanks to Sartaj Bhuvaji and Masoud Nickparvar for providing the datasets
• Kaggle: For providing the platform and computational resources
• TensorFlow/Keras: For the deep learning framework
• Medical Community: For inspiring this work to assist in early diagnosis

---

📞 Contact

For questions, suggestions, or collaborations:

• Email: abdulhadi2322005@gmail.com
• LinkedIn: My LinkedIn
• GitHub: My GitHub

---

⚠️ Disclaimer

This project is for educational and research purposes only. The models should not be used as a substitute for professional medical diagnosis. Always consult qualified healthcare professionals for medical advice and diagnosis.

---

📊 Citations

If you use this project in your research, please cite:

@misc{brain_tumor_classification_2025,
  title={AI-Powered Brain Tumor Classification using Deep Learning},
  author={abdelhady ali},
  year={2025},
  publisher={GitHub},
  url={https://github.com/Abdelhady-22/brain_tumor_classification}
}

---

Made with ❤️ for advancing medical AI

⭐ Star this repo if you find it helpful!