Multi-layer encrypted learning for distributed healthcare analytics.

Area of Science:

• Health Informatics
• Data Security
• Machine Learning

Background:

• The Internet of Medical Things (IoMT) facilitates continuous healthcare data collection, creating data-rich environments.
• Widespread data distribution across systems raises significant privacy concerns, especially during centralized data aggregation for analysis.
• Existing methods struggle to balance robust data privacy with the need for accurate analytical insights.

Purpose of the Study:

• To propose a novel privacy-preserving framework for distributed healthcare analytics on encrypted data.
• To mitigate data breach risks and privacy violations inherent in large-scale health data processing.
• To maintain analytical model accuracy while ensuring stringent data privacy.

Main Methods:

• Implementation of a three-layer protection mechanism for distributed healthcare analytics.
• Utilization of Fully Homomorphic Encryption (FHE) to encrypt data, enabling computations on ciphertext.
• Development of a distributed FHE framework supporting iterative learning and a distributed ensemble learning architecture for parallel processing.

Main Results:

• The proposed framework effectively protects data privacy in real-world intensive care unit (ICU) case studies.
• Analytical model performance is maintained without compromising data security.
• The privacy-preserving framework achieved the highest accuracy of 84.6%, outperforming individual departmental models.

Conclusions:

• The novel framework successfully addresses privacy concerns associated with IoMT data.
• Distributed FHE and ensemble learning offer a viable solution for secure, collaborative healthcare analytics.
• This approach avoids centralized data storage, enhancing security and enabling continuous model updates in data-rich healthcare settings.