Privacy-Preserving Collaborative Medical Time Series Analysis Based on Dynamic Time Warping

Abstract

Evaluating medical time series (e.g., physiological sequences) under dynamic time warping (DTW) derives insights assisting biomedical research and clinical decision making. Due to the natural distribution of medical data, a collaboration among multiple healthcare institutes is required to carry out a reliable and quality medical judgment. Yet sharing medical data cross the boundaries of multiple institutions faces widespread privacy threats, along with increasingly stringent laws and privacy regulations nowadays. Addressing such demands, we propose a privacy-preserving system tailored for the DTW-based analysis over the decentralized medical time series sequences. Our system constructs a secure and scalable architecture to deliver comprehensive results from a joint data analytic task with privacy preservation. To accelerate complicated DTW query processing, our system adapts the advancement in secure multi-party computation (MPC) framework to realize encrypted DTW computation, decomposing complicated and iterative operations into atomic functions under suitable MPC primitives and optimized for DTW. Moreover, our system introduces a secure hybrid pruning strategy that diminishes the volume of time series sequences that are submitted before and processed within the encrypted DTW query. We implement a prototype and evaluate its performance on Amazon Cloud. The empirical evaluation demonstrates the feasibility of our system in practice.