Privacy preserving ECG quality evaluation

Remote health-care applications are gaining popularity as an alternative for patients who do not require hospitalization. In this setting, privacy preserving protocols are useful to enable the offering of personalized online services, thus preventing the unnecessary disclosure of personal data. A problem often neglected in privacy-preserving protocols is the need to ensure that processed signals, which are often recorded by non-expert consumers, are of sufficient quality, hence raising the need for solutions that assess the quality of the recorded signals to guarantee correct (medical) decisions. In this paper, we propose a privacy preserving protocol that assesses signal quality and combines this with a linear classifier used to decide whether the measured signal is of high enough quality or not. In particular, the protocol computes a frame based Signal-To-Noise Ratio (SNR) from the original signal and a filtered version of the signal itself; evaluates the mean and the variance of the SNRs obtained and computes the overall signal SNR. Finally these measures are combined with a linear classifier used to assess the quality of the signal. The proposed scheme relies on a hybrid multi-party computation protocol based on Homomorphic Encryption and Yao's Garbled Circuits. The analysis of the protocol indicates that it needs the transmission of less than 4~MBytes of data to analyze 30 seconds of ECG signals providing a classification accuracy close to 85%.