Privacy preserving classification of ECG signals in mobile e-health applications

Abstract

Privacy protection is an emerging problem in mobile Health applications. On one hand, cloud services enable to store personal medical data, making them always available, and providing preliminary analysis on them, on the other hand, storing personal health data entails serious threats to users privacy. Privacy preserving solutions, such as Secure Multi-Party Computation techniques, give to nontrusted parties the opportunity of processing biomedical signals while encrypted. This chapter focuses on the development of a privacy preserving automatic diagnosis system whereby a remote server classifies an ElectroCardioGram (ECG) signal provided by the client without obtaining neither any information about the signal itself, nor the final result of the classification. Specifically, we present and compare three secure implementations of ECG classifiers: Linear Branching Programs (a particular kind of decision tree) with Quadratic Discriminant Functions, Linear Branching Programs with Linear Discriminant Functions and Neural Networks. Moreover we describe a protocol that permits to evaluate the quality of an encrypted ECG. The chapter provides a signal processing analysis aiming at satisfying both accuracy and complexity requirements. The described systems prove that carrying out complex tasks like ECG classification in the encrypted domain is indeed possible in the semi-honest model, paving the way to interesting future applications wherein privacy of signal owners is protected by applying high security standards.