This project explores and implements Federated Learning (FL) techniques in the clinical environment, particularly for medical applications where data privacy is crucial. The focus is on using FL to enhance machine learning models that classify MRI images of brains with various tumor types, while keeping patient data decentralized and secure.

Our project is developed in Python 3. For required packages, refer to requirements.txt.

pip install -r requirements.txt

Note: This code requires a Linux environment (either a VM or WSL).

1. Start the central server in a terminal:

   python3 server.py

2. Open four additional terminals for the clients. In each, start a client using:

   python3 client.py <index>

   Replace <index> with 0, 1, 2, or 3. The federated learning process begins once all clients are connected.

The machine learning models are in the mri_classification.ipynb Jupyter Notebook, detailing individual model configurations and evaluations.

• server.py: Manages the central server for federated learning, handling connections and aggregations.
• client.py: Each client runs this script, processing local data and contributing to model updates.
• utils.py: Includes functions for data loading and preprocessing, adapted to handle MRI image formats.
• mri_classification.ipynb: Notebook for training and evaluating machine learning models.
• requirements.txt: Lists all Python packages needed.

• Use server.py to launch the federated server.
• Run client.py for each federated client.
• mri_classification.ipynb contains the code for training and evaluation.

Recent updates have optimized model architectures and refined data handling, improving both the robustness and accuracy of tumor classifications.

Ensure the ./data/ directory contains the necessary MRI images for training and testing.

Coral, G., Hait, J., Isaak, K., & Watkins, A. (n.d.). KyleIsaak/Federated_Medical_Machine_Learning: A project at Simon Fraser University under CMPT 340, focusing on a federated approach to medical machine learning using MRI data.