2.2. Characteristics and Simulation of Metal Electrodes and Measurement Device
In order to implement sophisticated algorithms for assessing the signal quality and finally classifying for different cardiac arrhythmia, the signal morphology resulting from the environmental conditions and the measurement device itself needs to be understood in detail. Along with the metal electrodes of the utilized device and the specific measurement setup, some specific characteristics of the signal processing chain arise, which need further consideration. First, metal electrodes are not adhesively connected to the skin and, therefore, prone to an increased influence of motion artifacts. Additionally, gel-based wet-contact electrodes (usually Ag/AgCl) provide an optimal interface to the ionic conduction over the skin-electrode interface [
25]. The main difference between these two types of electrodes is the categorization into polarizable and non-polarizable electrodes, respectively. In polarizable electrodes, the metal electrode material cannot ionize into the salt solution and they are therefore prone to charge separation caused by the increased capacitive properties due to a possibly oxidized electrode surface [
14].
Additionally, metal electrodes highly depend on current characteristics of the connecting skin area and underlying tissue properties. Especially when being applied to the hands, the outer layer of the skin, the
stratum corneum (SC), adds a significant dynamically changing impact on the overall skin-electrode impedance characteristics. The complex impedance can be modeled by a capacitance and a parallel resistance. The resistance of the SC
varies between several
and
[
26]. The capacitance
of the SC was approximated by a cylindrical capacitance representation, which can be described by
where
denotes the vacuum permittivity,
and
are the length and the radius of the metal handle electrodes, respectively, and
is the thickness of the SC of the palm. The relative permittivity
of the SC was reported by Gabriel et al. [
27]. It can be regarded as approximately constant over the range of frequencies that are of interest to ECG analysis and varies in magnitude from
to
with the respective humidity conditions of the skin [
27,
28]. The thickness
of the SC on the hands strongly depends on individual patient characteristics and location and is typically in the range of several
up to
[
29,
30]. The capacity of the SC was calculated for an effective electrode length of
and an electrode radius of
according to the device manual. The results presented in
Figure 2 suggest a capacity range of
to
over the range of humidity conditions and SC thickness.
Figure 2.
Capacitance of the stratum corneum over the layer thickness. The relative permittivity is additionally dependent on the skin humidity.
Figure 2.
Capacitance of the stratum corneum over the layer thickness. The relative permittivity is additionally dependent on the skin humidity.
Besides the additional influences of the skin-electrode interface in this specific application and characteristics of the polarizable nature of stainless steel electrodes, systems like this incorporating metal electrodes often include actively-driven buffers with high input impedance
and coupling capacities
. In this case, the coupling capacity and the input resistance or bias resistance form a passive high-pass filter with cut-off frequencies in the lower
range to suppress baseline wandering. This methodology additionally affects the signal morphology but, on the other hand, provides a more stable measurement signal. The equivalent circuit diagram is presented in
Figure 3. To evaluate the influence of the previously described elements on the respective ECG signal morphology, the electrical model was simulated in MATLAB R2021b Simscape.
In our experiments, the input voltage
to the system was provided by an artificially generated ECG signal that was given by the synthesizer framework of Hoog Antink et al. [
31]. The ECG signal prototypes of the synthesizer are based on templates extracted from real Einthoven I-lead ECG signals provided by a publicly available database [
32]. Prior to each simulation run, an ECG signal of
length was generated, as would be expected when recording with a conventional patient monitor and standard ECG electrodes. Besides the correlation of the template morphology to a real ECG, the input signal additionally corresponded to the amplitude- and frequency-modulated characteristics of a realistic ECG signal to also capture dynamical inter-beat variabilities. The measured voltage
corresponds to the voltage at the input of the instrumentation amplifier. After each simulation run with a given parameterization, the resulting output signal was segmented by identifying the R-peaks using the Pan-Tompkins algorithm [
33] and the mean signal morphology was calculated with respect to the percentage of the heart cycle.
Figure 3.
Simplified equivalent circuit diagram of the ECG measurement chain provided by the device used in this study. The circuit consists of the internal voltage source , which equals a standard ECG signal measurement, the skin-electrode interface, and , of the metal electrodes, the input high pass-filter characteristic formed by the coupling capacity and the input impedance of the analog amplifiers.
Figure 3.
Simplified equivalent circuit diagram of the ECG measurement chain provided by the device used in this study. The circuit consists of the internal voltage source , which equals a standard ECG signal measurement, the skin-electrode interface, and , of the metal electrodes, the input high pass-filter characteristic formed by the coupling capacity and the input impedance of the analog amplifiers.
The simulations were conducted for a parameterization sweep through the range of SC resistance and capacity values of the hand
and
, respectively, and a coupling capacity
range from
to
, which corresponds to a cut-off frequency of
to
with
. Since the influence of the electrode-skin interface was found to be less significant with respect to the signal morphology (when assumed to be constant over time yet patient-individual), the simulation results were achieved with
and
.
Figure 4a shows the change of the signal morphology over a varying coupling capacity
and, thus, a varying cut-off frequency
of the 1st-order input high-pass filter. In
Figure 4b, the signal morphology of the original ECG recording and the resulting measured ECG signal with an input high-pass filter with
or
, respectively, are compared with an aligned Q-R-amplitude.
Figure 4.
Results of the measurement chain simulations. (a) Resulting changes in ECG signal morphology due to input high-pass filter characteristic over varying coupling capacity . Different coupling frequencies are represented by different shades of blue from to , the original ECG signal is shown in orange. (b) Comparison between original ECG signal morphology and amplitude-aligned filtered signal for the input high-pass characteristic with a cut-off frequency of .
Figure 4.
Results of the measurement chain simulations. (a) Resulting changes in ECG signal morphology due to input high-pass filter characteristic over varying coupling capacity . Different coupling frequencies are represented by different shades of blue from to , the original ECG signal is shown in orange. (b) Comparison between original ECG signal morphology and amplitude-aligned filtered signal for the input high-pass characteristic with a cut-off frequency of .
2.3. Consequences for ECG Signal Evaluation
Several observations can be obtained from
Figure 4 that provide information for the subsequent signal processing chain. First, the measurement signal is strongly affected by the input high-pass filter, which manifests both in amplitude as well as morphology (predominantly P-wave, S-wave, and T-wave). Especially for smaller P-wave amplitudes, the resulting waveform after high-pass filtering can be masked by noise artifacts with similar amplitude. The absence of P-waves, which is a necessary characteristic of AF, may therefore not be a reliable feature for identifying AF, especially for low signal quality-type ECG recordings. The QRS-complexes, on the other hand, still have a distinct characteristic that can be identified with standard peak detection methods, i.e., the Pan-Tompkins algorithm [
33]. The T-wave tends to appear as a fully periodic sinusoidal signal at first and then turns negative as the coupling capacitance decreases. The isoelectric line is reached much earlier due to the high-pass characteristic. Therefore, important parameters for other types of cardiac arrhythmias, such as ST segment length or ST deviations, can hardly be determined reliably, so that additional diagnostic capabilities are limited.
Additionally, a high-pass filter with a comparably high cut-off frequency of reduces baseline wandering and thus enables valid ECG measurements with low settling time but, on the other hand, lowers the energy in the signal band resulting in a decreased SNR for constant noise components in the upper frequency bands. With respect to the ECG morphology, this results in an overall decreased R-peak amplitude and relatively amplified noise components. As far as R-peaks can be clearly distinguished from (noise) artifacts, AF can still be detected by an irregular heart rhythm or the absence of a discernible pattern of the RR intervals, respectively. According to the manual of the MyDiagnostick device, the classification algorithm is based on RR-interval dispersion, which requires reliable R-peak detection but is not dependent on P-wave visibility.