Alzheimer's Disease Classification using Deep Learning

Project Overview

This project explores the application of deep learning models for classifying Alzheimer's disease stages using brain imaging data. We compare the performance of multiple state-of-the-art convolutional neural network architectures across two datasets: a standard dataset and a pseudo-RGB augmented dataset.

Datasets

1. Standard Dataset: Original brain imaging data without augmentation
2. Pseudo-RGB Dataset: Augmented version of the brain imaging data

The pseudo-RGB dataset preparation process is detailed in a linked repository: Pseudo-RGB Dataset Preparation.

Models Evaluated

We benchmarked several cutting-edge convolutional neural network architectures, including:

• VGG16
• ResNet50
• InceptionV3
• MobileNetV2
• Xception

Methodology

1. Data Preparation:

   • Split the datasets into train, validation, and test sets (80%, 10%, 10%)
   • Applied data augmentation techniques including rotation, width/height shifts, shear, zoom, and horizontal flip

2. Model Architecture:

   • Used transfer learning with pre-trained weights from ImageNet
   • Customized the top layers for our specific classification task
   • Implemented fine-tuning by unfreezing and training the last few layers of the base model

3. Training:

   • Utilized categorical cross-entropy loss and various optimizers
   • Implemented callbacks for early stopping, model checkpointing, and learning rate reduction
   • Trained for a maximum of 200 epochs with a batch size of 32

4. Evaluation:

   • Computed comprehensive metrics including accuracy, precision, recall, F1-score, Cohen's Kappa, log loss, and Brier score
   • Generated confusion matrices and ROC curves for multi-class classification
   • Analyzed per-class performance and misclassification rates

Repository Structure

For each model, the following files are generated:

• [ModelName].ipynb: Jupyter notebook containing all the code for data processing, model training, and evaluation
• model_summary.txt: Architecture summary of the model
• [Optimizer]_metrics_[ModelName].csv: Detailed metrics for the model trained with specific optimizer
• best_optimizer_metrics_[Optimizer]_[ModelName].csv: Best metrics achieved by the model with the best optimizer
• hyperparameter.csv: Hyperparameters used for model training
• segmented_multiclass_metrics.csv: Detailed metrics for all evaluated models
• summary_metrics_[ModelName].csv: Summary of metrics for the specific model
• training_history.csv: Epoch-wise training and validation metrics
• multiclass_confusion_matrix.png: Visualization of the confusion matrix
• multiclass_roc_curves.png: ROC curves for multi-class classification

Usage

To replicate this study:

1. Clone this repository
2. Install the required dependencies (list them here or include a requirements.txt file)
3. Run the desired [ModelName].ipynb notebook
4. Explore the generated CSV files for detailed metrics and PNG files for visualizations

Detailed File Descriptions

• [Optimizer]_metrics_[ModelName].csv: Contains epoch-by-epoch metrics for the model trained with a specific optimizer.
• best_optimizer_metrics_[Optimizer]_[ModelName].csv: Presents the best metrics achieved by the model using the optimal optimizer.
• hyperparameter.csv: Lists all hyperparameters used across different model trainings.
• segmented_multiclass_metrics.csv: Provides a comprehensive breakdown of metrics for each class and overall model performance.
• summary_metrics_[ModelName].csv: Offers a condensed view of key performance indicators for the specific model.
• training_history.csv: Records the training and validation metrics for each epoch during model training.

Future Work

• Explore ensemble methods combining multiple model architectures
• Investigate the impact of different data augmentation techniques
• Extend the study to include additional neuroimaging modalities

Acknowledgements

We acknowledge the contributions of various datasets, libraries, and tools used in this project. Special thanks to the creators of the pseudo-RGB dataset preparation process detailed in the linked repository.