OIDPR: Optimized insulin dosage via privacy‐preserving reinforcement learning

• Main Authors: YING, Zuobin, ZHANG, Yun, CAO, Shuanglong, XU, Shengmin, MA, Maode
• Format: text
• Language: English
• Published: Institutional Knowledge at Singapore Management University 2021
• Subjects: additive secret sharing; individualization dosing delivery; privacy-preserving; Medicine and Health Sciences; Software Engineering
• Online Access: https://ink.library.smu.edu.sg/sis_research/5182; https://doi.org/10.1002/ett.3953
• Institution: Singapore Management University

The precision of insulin dosage is essential in the process of diabetes treatment. The fact is providing precise dosage is almost impossible for clinicians since blood sugar levels are dynamically affected by many factors. Even though some auxiliary dosing systems have been proposed, the required real‐time physical data about the health situation of diabetics is still hard to synchronize to the end‐devices instantly. The traditional personalized drug delivery frameworks for accurate dosing of insulin always collect and transmit medical data in cleartext, which raises privacy problems. In this article, we propose a framework for an optimized insulin dosage via privacy‐preserving reinforcement learning to diabetics (OIDPR). In OIDPR, both the additive secret sharing and edge computing are deployed to achieve data confidentiality and performance optimization. The medical data is divided into multiple secret shares uniformly at random for outsourcing and operating at the edge servers. During the computation task of reinforcement learning, data is encrypted and processed via our proposed additive secret sharing protocol, where the privacy is reserved by the efficient encryption mechanism and the secret sharing system only incurs little workload. We provide comprehensive theoretical analyses and experimental results that demonstrate the supervisor functionality and high performance of our framework.