Alzheimer's Disease Stage Detection Using Deep Learning

This repository contains the implementation of a deep learning model designed to detect various stages of Alzheimer's disease using MRI brain scans. The project employs a transfer learning approach, utilizing the EfficientNetB0 architecture, adapted to classify images into four distinct stages: nondemented, very mild, mild, and moderate dementia. Built with PyTorch, the model is trained on a structured dataset of MRI images and provides a robust framework for early and accurate diagnosis of Alzheimer’s progression.

Key Features:

• Model Architecture: Utilizes the pre-trained EfficientNetB0 model, fine-tuned for the specific task of classifying Alzheimer's disease stages from brain MRIs.
• Data Processing: Includes comprehensive preprocessing and augmentation techniques to optimize model training and performance.
• API Integration: Features a Flask-based REST API that enables users to upload MRI images and receive diagnostic predictions, facilitating easy interaction with the model.
• Containerization: Docker support for easy setup and deployment, ensuring consistency across different environments.
• Evaluation Metrics: Implements various metrics for thorough evaluation and validation of the model, ensuring high accuracy and reliability.

Project Structure:

• /model: Contains the model definitions and state.
• /data: Dataset directory with train and test splits.
• /utils: Utility functions for data loading and transformations.
• /api: Contains the Flask application for the REST API.
• /tests/: Includes test scripts
• train.py: Script for model training.
• evaluate.py: Script for model evaluation and performance metrics.

Goals:

• To provide a reliable tool for early detection of Alzheimer's disease stages, aiding in better management and treatment planning.
• To contribute to the ongoing research in medical AI by demonstrating the application of advanced machine learning techniques in healthcare diagnostics.
• To offer an accessible platform for further development and validation by the research community, healthcare professionals, and technology enthusiasts.

Usage:

1. Train the model using train.py.
2. Evaluate the model's performance with evaluate.py.
3. Start the API server by running python api/app.py.
4. Test the API using python tests/test_alzheimers_api.py.

This project is intended for educational and research purposes, aiming to bridge the gap between medical imaging and machine learning technologies.

Detailed Module Descriptions

API (app.py)

Located in the /api directory, app.py sets up a Flask-based REST API for the Alzheimer's Disease Detection model:

• /predict endpoint for MRI image upload and prediction
• Image preprocessing using PyTorch transformations
• Integration with the trained AlzheimerNet model

Testing (test_alzheimers_api.py)

Found in the /tests directory, test_alzheimers_api.py provides API testing functionality:

• Random selection of test images
• API endpoint testing with selected images
• Comparison of true labels with model predictions

Training (train.py)

Located in the root directory, train.py contains the main training logic for the Alzheimer's detection model:

• Defines the train_model function for model training
• Implements the main execution flow in the main function
• Utilizes PyTorch for model training and optimization
• Loads data using custom data loaders from utils/data_utils.py
• Initializes the AlzheimerNet model, optimizer, and loss function
• Trains the model for a specified number of epochs
• Saves the trained model to the saved_models directory

Evaluation (evaluate.py)

Located in the root directory, evaluate.py contains the evaluation logic for the Alzheimer's detection model:

• Defines the evaluate_model function for model evaluation
• Implements the main execution flow in the main function
• Loads the trained model from the saved_models directory
• Utilizes the test data loader to evaluate model performance
• Computes and prints the accuracy of the model on the test set
• Handles device-agnostic evaluation (CPU or CUDA)