Privacy-Preserving Distributed Healthcare Analytics through Personal Clouds

In this research, we propose to develop a system to overcome the existing problems and limitations related to privacy-preserving data analysis and distributed learning in healthcare. Focusing on privacy-preserving contact-tracing, personalized health, and distributed learning, we propose collecting and processing personal information through distributed data containers, we refer to as “personal clouds”. Personal cloud idea can be realized by utilizing existing cloud architecture: by storing data in centralized clouds in combination of encryption and granular and dynamic access control. Each personal cloud provides both storage and computational capabilities. Utilizing personal clouds, our proposed system provides high availability and computationally rich environment. In the future, personal clouds can be deployed on individuals' end-devices as well.

To develop the proposed privacy-preserving healthcare analytics system, we will (i) develop a novel privacy-preserving data analysis and inference tool that will rely on information propagation on an interaction graph in a distributed way. For this, we will, rather than relying on a centralized server, utilize the proposed distributed architecture to provide privacy-by-design. Focusing on contact-tracing (for tracking and control of contagious diseases), we will study and show the robustness, efficacy, and scalability of the proposed tool. (ii) develop algorithms to combine data from multiple data sources, run models on the combined data, and provide recommendations to individuals (patients) or information to the healthcare providers. And, (iii) integrate a robust distributed learning algorithm we developed in our preliminary work into the proposed distributed system, which will allow generation of new models (e.g., about the symptoms of a disease or personalized treatments), and hence improve healthcare. We will also study the privacy and robustness guarantees of the proposed distributed learning platform.

Beyond these features, our proposed solution can be used for early detection of acute and chronic conditions, integration of various sensors to provide on-going and remote care, and personalized self-care and wellness management. We provide a proof-of-concept implementation of our proposed research using personal data accounts (PDAs), a cloud-based technology for data rights, portability, and control for individuals. PDA technology allows users to be the legal and functional owner of their data. We plan to extend our implementation for other alternative personal cloud architectures and explore interoperability across different platforms. This proposed research effort will develop a one-of-a-kind framework for efficient and privacy-preserving processing of healthcare data in a wide-range of applications.