EIT Heartbeat Signal Segmentation by Fourier Based Synchrosqueeezing

Major challenges in Electro Impedance Tomography (EIT) based monitoring of cardiovascular activity are the relatively low signal-to-noise ratio of perfusion related signals and the non-stationarity of major signal components. This contribution proposes an algorithm for separating ventilation and cardiovascular activity from EIT signals without reliance on side information like ECG. Our approach operates on raw electrode pair measurements without image construction. First, the ventilation rate is estimated from the short-time power-spectral density over all measurement pairs in order to attenuate the respiratory signal components. The linear combination of signal sources leading to the attenuation of ventilation is estimated from the input spectrum, enabling the algorithm for applicability to varying electrode belt configuration and orientation. Estimates of the instantaneous heart rate are obtained from the resulting data set by the Fourier Synchrosqueezing Transform (FSST) in our approach, where the local phase of the time-frequency representation provides the means for beat-by-beat segmentation. Apart from the assumption of two major non-sinusodial signal sources in the dataset, our proposed algorithm does not require explicit prior models. It is suitable for cases where the harmonics of the ventilation signal reach the range of the fundamental corresponding to the heart rate signal, as well as cases where ventilation and perfusion are coherent.