Systematic variability in ICG recordings results in ICG complex subtypes – steps towards the enhancement of ICG characterization

The first derivative of the ICG signal impedance change (ΔZ), dZ/dt, is typically represented [6], as shown in Fig. 1, with the 7 characteristic points introduced in Table 1. These characteristic points are critical in calculating certain hemodynamic parameters. Some of these parameters are presented in Table 2[9], including the requirements for its estimation.

In this work, two different sets of impedance cardiographs were used. Both sets were recorded at the Laboratory for Medical Textile-Electronics at the University of Borås, Sweden, at two different times with two different recording devices of continuous thoracic bioimpedance and on two different group of volunteers. In both cases, the recordings were performed after participants signed an informed consent form according to the ethical approval, 274-11, granted by the Regional Committee for Ethical Vetting of Gothenburg.

One set of ICG data was obtained in 2013 as described in [25, 26] with a Respimon impedance recorder (Medical Electronics Lab, Chalmers University of Technology, Sweden).

A second set of data was acquired with the Z-RPI ICG recorder [27]. It is a custom-made measurement device that simultaneously records a 3-lead ECG and ICG, combining the following system-on-chip (SoC) solutions: the ADAS1000 and the AD5933 from Analog Devices with a Raspberry PI3 card. The Z-RPI ICG recorder is fully described elsewhere [27].

To ensure equivalence between ICG measurements performed with both devices, the measurement protocol for the Z-RPI device was adjusted to the measurement protocol executed in previous studies [25, 28]. Measurements were performed in 5 young, healthy volunteers in a seated position while resting using Ag/AgCl gel electrodes. There were a total of N=10 volunteers.

The only difference between the measurement protocols was that ICG recordings with the Z-RPI device were 5 min rather than 60 s. To avoid respiration artifacts, breathing was kept shallow at 10 cycles/min during the recording. Despite the availability of 500 s for each volunteer, the analyses of recordings were limited to 60 s windows.

Both datasets were obtained with Ag/AgCl gel electrodes placed according to a neck-to-chest electrode configuration, as proposed in a previous study [4]. A small shift of electrode placement was introduced (see Fig. 2) to determine whether electrode placement significantly influenced the variability of recordings.

Fig. 2

After all ICG and ECG measurements were stored, the recordings were preprocessed offline using MATLAB^TM, as shown in Fig. 3. Both detrending and filtering functions were applied to both ICG and ECG signals.

Fig. 3

Heart rate was obtained from the ECG signal by detecting R-waves and applying the Pan-Tompkins algorithm [28]. The first derivative of the ICG signal was obtained from ΔZ through derivation after smoothing with a low-pass filter with cut-off frequency of 18 Hz (see Fig. 3).

Analyses of the ICG data were performed using MATLAB™ software by first displaying the processed recorded signals and then evaluating the PQRST and ABEXYOZ complexes. ABEXYOZ complexes were detected by the R peaks of the ECG. Then, ABEXYOZ complexes were categorized according to the ICG points.

There were 5 ABEXYOZ complexes that were clearly defined. This categorization enabled the calculation of the percentage of different ABEXYOZ complexes in each group and their distribution among the volunteers. The percentage of each characteristic ICG point in each volunteer was also calculated.

All procedures performed in this study involving human participants were in accordance with the ethical approval, 274-11, granted by the Regional Committee for Ethical Vetting of Gothenburg.

Informed consent was obtained from all individual participants included in the study.

In this study, two different sets of ICG recordings were used. They were recorded at two different times with two different groups of volunteers using two different ICG devices. In both groups, the arrangement of electrodes was performed according to the configuration proposed by Kubicek [4], but the actual position of the electrodes was slightly different. The same measurement protocol was

used for each subject with minor changes, ensuring that valid comparisons could be made while broadening the scope of the study. These non-identical measurement sets enriched the study by incorporating different devices and electrode placements, and despite being not identical, they are equivalent. The measurement sets even presented similar heart-rate ranges despite being collected completely independently and years apart. The different ABEXYOZ complexes were clearly defined by corresponding specific waveforms, which are summarized in Table 6.

The presence of up to five different ABEXYOZ complexes in dZ/dt signals, in addition to the typical and undetermined waveforms, was confirmed. All ABEXYOZ types were present in each of the dZ/dt recordings from both measurement groups. Therefore, we confirmed that (1) all ABEXYOZ complex types are present in dZ/dt recorded signals from the two devices used and (2) all ABEXYOZ complex types are present in the dZ/dt recorded signals from both electrode configurations.

By analyzing Tables 3 and 4, we observed that no specific ABEXYOZ complex seems to be dominant, and the distribution of waveforms differs between groups. However, the distribution of ABEXYOZ complexes varied between different volunteers, with ABEXYOZ 4 dominating among volunteers of group 1 and ABEXYOZ 1 being dominant in those of group 2 (see bar plots in Fig. 7).

Given that the volunteers in each group were completely different, these results are expected because specific physiological and cardiodynamic events related to the duration of the opening and closure of the aortic valves vary between people [13]. Variation in the ICG signal between subjects has been previously observed and reported [29]. Therefore, five different ABEXYOZ types were observed in most of the ICG recordings, which were well distributed but not homogeneously distributed. In several ICG recordings, all five different ABEXYOZ types were observed, but in other cases, only two types were observed.

In the obtained recordings, the less dominant ABEXYOZ complex types were ABEXYOZ 5 and ABEXYOZ 2 in both groups.

Although ABEXYOZ 2 and 5 were not present as often as the rest of the types in the ICG recordings, the observed differences in the YOZ segment clearly indicated that they are both specifically different ABEXYOZ types. They are distinct because a distinguishable X point is missing, which is a remarkable feature given that the X point is critical for calculations of hemodynamic parameters.

The primary difference between the five types of ICG complexes and the typical complex, ABEXYOZ 0, was observed in the second part of the complex after the E wave (the XYOZ segment). Among all of the alterations observed in different ABEXYOZ complexes, the X point varied the most, with the lowest minimum after the E wave. This variability of the X point has led to different definitions of X points, which consequently have produced different detection algorithms, which are tailored to each definition [22]. Additionally, this alteration of the X point has been observed, reported [21] and studied previously [19, 20]. In previous studies [19], [20], the physiological sources that might contribute to building ICG signals (ΔZ and dZ/dt) were investigated.

While it is widely accepted that the main contribution of the ICG signal is produced by aortic expansion, the authors have also shown that the contribution from lung perfusion and erythrocyte orientation cannot be neglected.

Therefore, in the ABEX complex, there are three possibilities for point X:

(d1) X, which is a true X point and is defined as typical (see Fig. 1). This point is present as the lowest minimum after the E point. In a previous study [22], it was defined as the lowest minimum in the entire cycle after the E point. This definition can be misleading with the Y point in some cases. However, in some cases, the tailor-made detection algorithm fixes a percentage from the period of CC intervals to be scanned.

(d2) X1 is the minimum point on the descending slope of the E wave, and

(d3) X is the observed alternative of X2 as an unpronounced X in terms of amplitude (ampX2 < ampX1).

Finally, the YOZ complex has missing points that commonly appear or disappear in specific locations.

In this study, the typical X point is only present in the typical ABEXYOZ complex, ABEXYOZ 0, and ABEXYOZ 1, as defined in a previous study [6]. In other ABEXYOZ complexes, there are two other wave flexions near the typical X point (X1 and X2). These two flexions are not typical, and they do not refer to a true X point, where true X is considered according the criteria of the lowest minimum flexion after the E wave, as shown previously in Fig. 1 and defined in a previous study [6].

Usually, detecting an X point according to d1 or d2 is misleading; for d1, it may lose the detection of the real X point in terms of time, position and physiological events. For d2, it may detect the Y point and consider it as the actual X point, and in that case, X2 is often present when problems in detecting X arise.

According to physiological events from the ECG, X2 showed the possibility of being an actual X point; therefore, we devised an alternative definition for the X point (d3). In this study, we could account for the number of ABEXYOZ types that includes an X2, which led to approximately 20% of the ICG data representing up to three ABEXYOZ types. More studies in frequency and time analysis should be performed to validate the definition of d3 as a method to detect the X point.

As examples, in the case of ABEXYOZ 4, X1 is the only minimum without any flexion decrease, and in ABEXYOZ 2, -

3 and -5, X2 is not sufficiently pronounced to be the lowest minimum amplitude after the E wave.

In addition to the 5 ABEXYOZ types and the typical ABEXYOZ 0, ABEXYOZ u was present in almost all volunteers. This type of waveform was observed as weakly defined complexes when all typical characteristic points of the dZ/dt signal were missing.

When inspecting the waveforms reported from the simulation work in previous studies [19], [20], among the ABEXYOZ complexes, ABEXYOZ complex 2 improved the reassembling of the waveform and contained contributions from all of the physiological sources, including lung perfusion, erythrocyte orientation and heart rate. Furthermore, the typical ICG complex, labeled here as ABEXYOZ 0, closely agreed with the ICG waveform produced only by the volume changes in the aorta [20].

The ECG recordings were perfectly clear and repetitive, independent of the differences observed in the ICG waveforms. However, all ICG measurements varied between subjects’ activities, and repeatable measurements were not possible for all cases [29].

The absence of X has been considered as a pathological instance in a previous study [17] or refers to elderly people. In a previous study [30] with recordings from young healthy subjects, ICG complexes with absent X points were averaged to produce ICG complexes with a clear X point. However, the presence of the X point was achieved at the expense of losing YOZ points.

The observation that the X point is completely absent, i.e., the flexion is not present (X1) or not pronounced as a minimum (X2) in our relatively young and healthy volunteers, suggests that its absence cannot have a pathological source as proposed by other authors [17].

Unlike in a previous study [30], in our study, averaging the ICG produced ICG complexes with clear X points in 6 out of 10 volunteers. Consequently, we did not consider averaging to be a valid alternative to improve the detection of the X point in ICG recordings, especially when the YOZ points were lost in the process as well, as previously reported [30].

The relationship between the ABEXYOZ type and the absence of ICG characteristic points indicates that a difference in the frequency of ABEXYOZ types impacts the absence of ICG characteristic points. Although X is the most common missing ICG characteristic point among the presented ABEXYOZ types, and absent X occurs in 4 out of 5 cases, X is not the most commonly absent point in the obtained ICG recordings. Since both ABEXYOZ 0 (12.43%) and ABEXYOZ 1 (20.39%) contain X points, the absence of a clear X point is limited to 67.18%, while the combined occurrence of ABEXYOZ 1, -3 (13.70%), -4 (32.85), and u (14.19%) ensures the absence of a clear YOZ segment at 81.13%.

The most common set of ICG characteristic points is the ABE segment, and only the ABEXYOZ u case contains no ABE segments. The absence of characteristic points on the ABEXYOZ complexes influences the assessment of hemodynamic parameters from ICG analysis. The estimation of hemodynamic parameters requires the identification of different ICG characteristic points. Left ventricular ET, SV and CO require the detection of both X and B. Other ICG segments or parameters, such as the IVRT and FT, require the accurate detection of X-O and Y-Z, respectively. These last two parameters are important for the diagnosis of diastolic functions of the heart and cardiac malformations.

The currently evaluated recordings suggest that it is possible to directly estimate all hemodynamic parameters of interest (ET, SV, CO, IVRT and FT) in only 12.47% of cases, corresponding to ABEXYOZ 0. Overall, it is possible to estimate the ET, SV and CO in only 32.91% of cases. In only 12.47% and 18.83% of cases, it is possible to estimate IVRT and FT, respectively. Additionally, no parameters can be accurately determined, due to the missing X and YOZ points in ABEXYOZ u, ABEXYOZ 3 and ABEXYOZ 4, in up to 60.74% of cases.

Therefore, to calculate these hemodynamic parameters, it is important to accurately detect ICG characteristic points; therefore, it might be helpful to identify which ABEXYOZ complex types are present in the dZ/dt signal.