Model Based Enhancement of Bioimpedance Spectroscopy Analysis : Towards Textile Enabled Applications

Sammanfattning: Several signal processing approaches have been developed to overcome the effect of stray capacitances in Electrical Bioimpedance Spectroscopy (EBIS) measurements. EBIS measurements obtained with textile-enabled instrumentation are more vulnerable to stray capacitances. Currently, the most widespread approach for correcting the effect of stray capacitances in EBIS is the time delay ( Td) compensation method, which also has several drawbacks. In this study, the Td method is revisited and its limitations and its lack of a scientific basis are demonstrated. To determine better ways to overcome the effect of stray capacitances, a simplified measurement model is proposed that is based on previous models of artefacts in EBIS measurements described in the literature. The model consists of a current divider with a parasitic capacitance (Cpar) in parallel with the load. Cpar creates a pathway for the measurement current to leak away from the load, provoking a capacitive leakage effect. In this thesis, three approaches with different limitations are proposed to overcome the capacitive leakage effect. The first approach estimates Cpar and subtracts it from the measurements, thus finding the load. Cpar can be estimated because the susceptance of biological tissue is null at infinite frequency. Therefore, at high frequencies, the susceptance of the tissue can be neglected, and the slope of the susceptance of the measurement is Cpar. The accuracy of Cpar depends on the maximum frequency measured and the value of Cpar. Therefore, it may not be possible to accurately estimate small values of Cpar in the typical frequency ranges used in EBIS. The second and third approaches use the Cole fitting process to estimate the Cole parameters, which form the basis for most EBIS applications. Because the conductance of the measurement is free from the effect of Cpar, performing Cole fitting on the conductance avoids the effect of Cpar in the fitting process. With a poor skin-electrode contact, this approach may not be sufficiently accurate. The third approach would be to perform the Cole fitting on the modulus with a reduced upper frequency limit because the modulus and the low-medium frequencies are very robust against the effect of artefacts. In this approach, a slight capacitive leakage effect is unavoidable. Since it is common to find tainted measurements, especially among those obtained with textile-enabled instrumentation, it is important to find viable methods to avoid their effect. The three methods studied showed that they could reduce the effect of tainted measurements.