FeSEC: A Secure and Efficient Federated Learning Framework for Medical Imaging.

Area of Science:

• Medical Imaging
• Machine Learning
• Data Privacy

Background:

• Federated learning (FL) leverages distributed data for improved medical imaging model generalization.
• FL offers partial privacy but risks compromise during model parameter exchange.
• Communication overhead is a significant challenge in FL, especially with complex models and distributed collaborators.

Purpose of the Study:

• To propose FeSEC, a secure and efficient FL framework.
• To address privacy concerns and communication bottlenecks in medical imaging FL.
• To enhance the performance of FL models in distributed healthcare settings.

Main Methods:

• Implemented a sparse compression algorithm for efficient inter-hospital communication.
• Integrated homomorphic encryption with differential privacy for secure model exchange.
• Evaluated the framework on a COVID-19 detection task.

Main Results:

• FeSEC substantially improves accuracy and privacy preservation compared to FedAvg.
• Achieved significant reductions in communication costs (less than 10% of FedAvg).
• Demonstrated enhanced FL model performance in a real-world medical imaging scenario.

Conclusions:

• FeSEC offers a viable solution for secure and efficient federated learning in medical imaging.
• The framework effectively balances privacy, communication efficiency, and model accuracy.
• FeSEC shows promise for global health applications requiring collaborative medical data analysis.