Alzheimer Diagnosis CNN Model

Project Description

This project uses Convolutional Neural Networks (CNN) to diagnose Alzheimer's disease from brain MRI images. The model is trained to classify different stages of Alzheimer's disease, including:

• Mild Impairment: Mild cognitive impairment
• Moderate Impairment: Moderate cognitive impairment
• Very Mild Impairment: Very mild cognitive impairment
• No Impairment: No signs of cognitive impairment

Project Structure

├── images/                              # Image for README description
├── cnn_custom_model_v3.ipynb            # Custom CNN model training notebook
├── transfer_learning_models.ipynb       # Transfer learning experiments notebook
├── test_saved_model.ipynb               # Saved model testing notebook
├── Alzheimer_diagnosis_cnn_custom_model(99.33%).h5  # Trained model (99.33% accuracy)
└── README.md                            # Project description file

System Requirements

• Python 3.7+
• TensorFlow 2.x
• Keras
• NumPy
• Pandas
• Matplotlib
• Seaborn
• OpenCV
• Scikit-learn

Installation

pip install tensorflow keras numpy pandas matplotlib seaborn opencv-python scikit-learn

Usage

1. Model Training

Open and run the cnn_custom_model_v3.ipynb notebook to train the custom CNN model.

2. Model Testing

Use the test_saved_model.ipynb notebook to test the saved model.

3. Transfer Learning

Experiment with pre-trained models through the transfer_learning_models.ipynb notebook.

Model Architecture

The custom CNN model includes:

• Convolutional Layers: Extract features from MRI images
• Pooling Layers: Reduce size and increase invariance
• Dropout Layers: Prevent overfitting
• Dense Layers: Final classification
• Activation Functions: ReLU for hidden layers, Softmax for output layer

Training Results

Model Accuracy

The custom CNN model achieved 99.33% accuracy on the test set.

Training Results Visualization

1. Loss and Accuracy Plot by Epoch

Plot showing the changes in loss and accuracy during training

2. Confusion Matrix

Confusion matrix showing the model's classification performance for each class

3. Classification Report

Detailed report on precision, recall, and f1-score for each class

Model Evaluation

Key Metrics:

• Accuracy: ~99.33%
• Precision: ~99% (average)
• Recall: ~99% (average)
• F1-Score: ~99% (average)

Dataset

• Total images: 16,650 MRI images
• Split:
  • Training: 60% (10,204 images)
  • Validation: 25% (4166 images)
  • Test: 15% (2244 images)
• Image size: 128x128 pixels
• Format: Grayscale

License

This project is distributed under the MIT License. See the LICENSE file for more details.

Note: This model is for research and educational purposes only. Do not use for actual medical diagnosis without supervision from medical specialists.