Explainable federated learning scheme for secure healthcare data sharing.

Area of Science:

• Artificial Intelligence
• Smart Healthcare
• Data Security

Background:

• Vast amounts of medical data are generated by wearable/implantable devices in Body Area Networks.
• Leveraging this data with AI is crucial for advancing smart healthcare applications.
• Existing methods face challenges in ensuring explainability and security for AI in healthcare.

Purpose of the Study:

• To propose an innovative Federated Learning (FL) scheme for smart healthcare.
• To address the critical challenges of explainability and security in AI-driven healthcare.
• To enable the utilization of dispersed medical data from Body Area Networks.

Main Methods:

• Implemented an FL scheme with independent federated modeling and explainability analysis.
• Utilized post-hoc explanation techniques for global model analysis.
• Introduced a client private gradient evaluation for fair contribution assessment.
• Proposed a multi-server model with homomorphic secret sharing and hashing for secure aggregation.

Main Results:

• Achieved explainability comparable to centralized training without performance degradation.
• Demonstrated effective filtering of low-quality data through gradient contribution evaluation.
• Ensured robust data privacy and aggregation correctness against malicious servers.
• Showcased competitive security and efficiency in experimental results.

Conclusions:

• The proposed FL scheme effectively balances explainability and security in smart healthcare.
• It provides a robust solution for analyzing sensitive medical data from wireless Body Area Networks.
• This approach unlocks AI's potential in smart healthcare while safeguarding patient privacy.