Secure Federated Learning Integrated Statistical Modeling for Healthcare Data
Xiaoqian Jiang, Jihoon Kim, Tsung-Ting Kuo, Lucila Ohno‐Machado · 2024
Several biomedical informatics applications depend on the analysis of large data sets. Utilizing data from different sources results in higher statistical power and provides data diversity from which to construct statistical (and/or machine learning) models that are more generalizable than those built with data from a single source. The most common form of analysis in healthcare today involves centralization of data at a given institution. However, privacy concerns have been a major hurdle to centralizing healthcare data, due to the risks of personal information and intellectual property breaches. Additionally, a myriad of regulatory barriers may prevent data from being physically transferred outside an institution or country. To address this problem, many researchers have investigated secure and privacy-preserving federated learning solutions (i.e., solutions that bring computation to the data that are distributed in various institutions) that involve advanced cryptographic and privacy technology. The wide range of applications in clinical data analysis and genomics research inspired the development of different mechanisms to balance the privacy of individuals and institutions and the utility of shared data. In this chapter, we summarize recent work in biomedical models with a focus on federated learning and integration of security and privacy technology.