Smartware electrodes for ECG measurements : Design, evaluation and signal processing

Abstract: The aim of this thesis work has been to study textile and screen printed smartware electrodes for electrocardiographic (ECG) measurements both in terms of their electrode properties and possibility to further improve their robustness to movement induced noise by using signal processing. Smartware electrodes for ECG measurements have previously been used in various applications but basic electrical electrode properties have not sufficiently been looked into. Furthermore, we believe that there is a possibility to reduce disturbances in the smartware ECG by adding redundant sensors and applying sensor fusion signal processing.Electrical properties of conductive textiles have been evaluated in terms of stability and electrode impedance. Three yarns and textile electrode surfaces were tested. The electrodes made from pure stainless steel and 50\% stainless steel/ 50\% polyester showed acceptable stability of electrode potentials. All electrode measurements were performed on skin.Furthermore, we produced six screen printed electrodes and their electrical performance was investigated in an electrochemical cell. The tested inks contained carbon or silver particles in the conduction lines, and Ag/AgCl particles in the electrode surface. Results show that all electrodes were stable in time, with a maximum drift of a few mV during 30 minutes. The silver ink is superior to the carbon based in terms of electrode impedance at the higher frequencies.To extract viable physiological information from noisy signals, canonical correlation analysis (CCA) was applied on multi-channel ECG signals recorded with textile electrodes. Using CCA to solve the blind source separation (BSS) problem, we intended to separate the ECG signal from the various noise sources. The method (CCABSS) was compared to averaging of the ECG channels and to the independent component analysis method (ICA). In the dataset consisting of noisy ECG recordings, the signal was uninterpretable in 7% after CCABSS. Corresponding values for averaging and ICA were 33% and 17%, respectively.Smartware applications often include heartbeat detection while moving, a measurement situation which is prone to produce noise corrupted ECG signals. To compensate for this, we used an event detector based on a multi-channel input, a model of the event and weighted correlation. For measurements at rest and static muscle tension, the sensitivity of the event detector was 97% and 77% respectively. Corresponding values for the golden standard detector Pan-Tompkins were 96% and 52%, respectively.