Advancing Privacy-Preserving Health Care Analytics and Implementation of the Personal Health Train: Federated Deep...

Area of Science:

• Artificial Intelligence in Healthcare
• Deep Learning
• Medical Imaging Analysis

Background:

• Deep learning offers significant potential for healthcare automation but faces challenges with data privacy and the need for large, diverse datasets.
• Federated learning (FL) enables collaborative AI model development without sharing patient data, addressing privacy concerns.
• Robust and secure federated deep learning infrastructures are crucial for widespread AI adoption in healthcare.

Purpose of the Study:

• Introduce the Personal Health Train (PHT), an innovative federated learning infrastructure for real-world healthcare data.
• Detail the procedural, technical, and governance components of the PHT for implementing federated deep learning.
• Apply the PHT to a proof-of-concept for gross tumor volume segmentation in lung cancer patients using chest CT images.

Main Methods:

• The PHT framework keeps data at its source, bringing analysis to the data to ensure privacy.
• Key technological components include 'tracks' (secure channels), 'trains' (containerized apps), and 'stations' (data repositories), supported by Vantage6 software.
• A secure aggregation server was introduced for trusted model averaging in the lung cancer segmentation use case.

Main Results:

• Demonstrated the feasibility of federated deep learning execution using the PHT infrastructure.
• The proof-of-concept study successfully linked 12 hospitals across 8 nations on 4 continents, showcasing scalability and global reach.
• No patient data was shared outside of participating hospitals during algorithm execution and training.

Conclusions:

• The PHT framework and Vantage6 platform represent a significant advancement for federated deep learning on medical imaging data.
• The infrastructure effectively addresses data privacy challenges, enabling collaborative model development for AI tools in medicine.
• The secure aggregation server design helps prevent data leakage issues in federated learning.