Efficient Server-Aided Personalized Treatment Recommendation with Privacy Preservation.

Dan Zhu,Hui Zhu,Cheng Huang,Rongxing Lu,Xuemin (Sherman) Shen,Dengguo Feng
DOI: https://doi.org/10.1109/globecom48099.2022.10000647
2022-01-01
Abstract:With AI-derived knowledge graph (KG), medical centers can recommend appropriate treatment options to physicians as references based on their patients' personal healthcare information (PHI). However, the treatment recommendation services may also cause serious privacy concerns. In this paper, we propose an efficient and privacy-preserving personalized treatment recommendation scheme with the aid of a third-party server. Specifically, a medical center denotes the KG of each disease by a directed graph with conditional edges and vertices that describe the treatment options and costs in different states. To prevent privacy leakage while reducing computational and management cost, the graphs are encrypted and then delegated to an honest-but-curious server. With the assistance of the server, physicians can set proper illness states and cost requirements according to patients' PHI, and correspondingly generate personalized ciphertexts to retrieve appropriate treatment options. The key component of the proposed scheme is a novel designed secure and flexible path comparison protocol, by tailoring a symmetric homomorphic encryption algorithm and combining it with a secure hash function. The protocol can enable the server to compare the uploaded ciphertexts with encrypted graphs in a secure and efficient way. Comprehensive security analysis indicates that the proposed scheme can meet desirable privacy requirements, and extensive experimental results demonstrate its practicality.
What problem does this paper attempt to address?