FL+2: Multi-Layered Privacy Protection for Federated Learning-Based Medical Diagnostic

To build trust between patients and health institutions, implementing a privacy-preserving mechanism to handle personal health information is crucial. Therefore, federated learning (FL) is gaining popularity in health data analysis, due to its ability to enhance trustworthiness by sharing model-related information rather than data. Blockchain, with its decentralization, immutability, and traceability features, has also been integrated with FL to address existing privacy concerns. In this paper, we introduce FL+2, which is designed to address the infrastructural security and model-focused traceability challenges associated with FL-based medical analysis. FL+2 incorporates a first layer that utilizes a strict data transmission control mechanism using a P4 switch for infrastructural security. The second layer ensures model traceability (for better model privacy) and data access (including data delegation) control via blockchain. Here, we present the early results of the proposed FL+2 prototype, which was implemented in our in-house cloud/fog testbed. We discuss how a P4 switch as well as blockchain-based controlled data access and traceability mechanism offer specific security benefits and the associated development overhead.