Multi-layer Privacy Protection for Federated Learning and Encrypted Healthcare Analytics

Abstract

The Internet of Medical Things (IoMT) enables continuous collection and transmission of healthcare data through interconnected networks of patient wearables and other devices. This capability transforms traditional healthcare systems into data-rich environments. However, this data-rich environment also brings privacy concerns because of the widespread distribution of health data across multiple healthcare systems. Such concerns, including data breaches and privacy violations, become paramount when aggregating data into a centralized location for analytical purposes. To tackle these challenges, this paper presents a new privacy-preserving framework with three-layer protection for distributed analytics on encrypted data from different units across multiple independent healthcare systems. First, we propose encrypting the data for analytical computation to mitigate the likelihood of data breaches. Second, a distributed framework with federated learning is designed to circumvent the need for centralized data storage and enable iterative learning and model updates when new data become available. Finally, the proposed framework leverages an ensemble learning approach to enhance both computational efficiency and model performance. Experimental results from real-world intensive care unit (ICU) case studies show that the proposed framework effectively protects data privacy while maintaining the performance of analytical models. These findings highlight the potential of a federated privacy-preserving framework to avoid centralized data storage and support collaborative analytics in data-rich healthcare environments.