Effect of Respiration and Exercise on Seismocardiographic Signals

Abstract

Background:

Seismocardiographic signals (SCG) are chest wall vibrations induced by mechanical cardiac activities. This study investigated the morphological changes in the SCG signal due to respiration and exercise.

Methods:

Fifteen healthy subjects were recruited, and SCG was acquired before and after exercise. The changes in the SCG signal were quantified using time and amplitude features.

Results:

The amplitudes of the two main SCG events (SCG1 and SCG2) tended to increase after exercise. The absolute cardiac intervals (pre-ejection period (PEP), left ventricular ejection time (LVET), and diastolic time) decreased; the diastolic time relative to cardiac cycle duration (i.e., the R-R interval) also decreased, while the relative PEP and LVET increased for the majority of the subjects. Amplitude modulations were observed in both SCG1 and SCG2 and increased with exercise. Additionally, respiratory influences on the SCG features were observed in both the pre- and post-exercise states. SCG2 amplitude was higher during inspiration (p < 0.01), but SCG1 amplitude didn’t exhibit consistent changes with respiration in the study subjects (p > 0.05). For cardiac intervals, PEP decreased during inspiration, while LVET and diastolic time increased (p < 0.01). All the cardiac intervals (both absolute and as a percentage of cardiac cycle duration) showed reduced respiratory variability post-exercise.

Conclusions:

These results document SCG signal variabilities that were not reported before and provide a link between cardiac activity, respiration, and exercise, which may help increase the clinical utility of SCG in the diagnosis and management of cardiopulmonary conditions. More studies are required to validate the study findings in more normal subjects and in those with cardiopulmonary pathology.

1. Introduction

Seismocardiography (SCG) measures low-frequency vibrations at the chest wall surface, which in turn are induced by cardiac phenomena such as valve closure, blood flow turbulence, myocardial contraction, and flow momentum changes [1]. Given the global rise in cardiovascular disease (CVD) cases, there has been a growing interest in exploring noninvasive and cost-effective methods for CVD diagnosis and monitoring. Several studies have suggested that SCG may prove to be such a method [2]-[4].

SCG signals can be separated into two main components, SCG1 and SCG2, each of which occurs during different cardiac mechanical processes (Fig. 1). For example, SCG1 occurs during the closure of the mitral and tricuspid valves, isovolumetric contraction of ventricles, aortic valve opening, and ventricular ejection, marking the onset of systole [5]. Conversely, SCG2 occurs during aortic and pulmonary valve closure, isovolumetric relaxation of ventricles, mitral valve opening, and ventricular filling, indicating the shift from systole to diastole [5].

Fig. 1.

An SCG beat with SCG1 and SCG2 locations specified.

Some previous studies were conducted to monitor heart activity by combining electrocardiography (ECG) with the SCG signal. Sahoo et al. proposed a cardiac health monitoring approach based on the simultaneous acquisition of ECG and SCG signals [6]. The study employed analysis of ECG and SCG signals to distinguish normal and abnormal heart behavior. Results showed that the combined analysis of ECG and SCG signals provided more reliable cardiac health monitoring than using either signal separately.

Inan et al. investigated the feasibility of a wearable ECG and SCG sensing patch in analyzing cardiac response to submaximal exercise in HF patients to identify compensated and decompensated HF states [3]. In that study, HF patients were fitted with the wearable patch, stood at rest for an initial recording, performed a 6-minute walk test, and then stood at rest for 5 minutes of recovery. Results showed that the graph similarity score is significantly higher in decompensated patients than compensated patients, suggesting a reduced cardiovascular reserve in the decompensated patients.

Physical exercise induces reversible morphological changes in SCG and ECG signals. The study of these changes may provide useful information about cardiac conditions. In a study of 824 women with suspected coronary artery disease (CAD), it was suggested that the ECG exercise tredmill test should be considered as the initial diagnostic strategy in symptomatic women [7]. In [8] and [9], the presence of ST segment change in the ECG after treadmill stress test has been found to be indicative of obstructive CAD. But using only the ECG parameters may limit the diagnostic accuracy. For instance, horizontal or downward ST-segment depression alone has a sensitivity of 68% for the detection of coronary obstruction [10]. Hence, combining other clinical data with ECG parameters has been suggested in [11]. In the light of this, integrating SCG features with ECG may be a logical approach. Because together the signals offer a more comprehensive understanding of both the electrical (ECG) and mechanical (SCG) activities of the heart. In their respective studies, Salerno et al. [12] and Dehkordi et al. [13] showed that combining exercise SCG with exercise ECG indeed increased the sensitivity of CAD detection. Although the clinical potential of exercise SCG has been demonstrated, the morphological changes in SCG induced by exercise have not been adequately investigated. The proper understanding of the exercise-induced changes in SCG may help to identify the prominent SCG features, which may further improve its utility for cardiac monitoring. Therefore, one objective of this study is to investigate the relationship between the SCG signal and exercise-induced physiological changes.

While SCG signal holds promise for diagnostic applications, SCG variability has been a consistent utility limitation [14]. This is primarily because SCG signal morphology is influenced by several factors, including respiration, subject posture, and different underlying cardiac pathologies [15], [16]. The variability in SCG may hamper the process of accurate feature extraction and lead to misdiagnosis [17], [18]. Previously, studies have been conducted that reported the variability of SCG induced by respiration; however, the sources of the variability and related physiological mechanisms remain inadequately understood. Peshala et al. performed unsupervised clustering to reduce SCG variability and observed that SCG morphology can be optimally separated into two clusters based on respiration [19]. Pandia et al. also observed respiratory effects on SCG in their study [20]. Taebi et al. suggested the change in intrathoracic pressure during respiration may contribute to the SCG variability [21]. In this study, SCG morphological variability due to exercise and respiration are analyzed, and the underlying physiological mechanisms are described. Understanding the sources of variability and separating the SCG features according to their respiratory phases can reduce the effects of respiratory patterns and allow extracting more meaningful information from the SCG signal. This may also offer insights into complex cardiopulmonary interactions and can enhance the diagnostic utility of SCG.

2. Method

The SCG signal was acquired from 15 healthy subjects with no known history of medical conditions after their informed consent and IRB approval by the University of Central Florida (protocol number: BIO-16–12783; the date of approval: February 19, 2024). The demographic information of the subjects is given in Table 1.

Table 1.

Demographic data of the study participants. Age is reported as mean ± standard deviation (SD). BMI = Body Mass Index (kg/m²;). N (%) represents the number and percentage of participants.

Subject Characteristics (N=15)
Gender, N (%) Male Female	8 (53%) 7 (47%)
Age [years], mean ± SD	25 ± 3
BMI [kg/m2], mean ± SD	22.4 ± 2.05

A triaxial accelerometer, placed on the 4th intercostal space (ICS) at the left sternal border (LSB), was used to acquire the SCG signal. This is consistent with a previous study that reported that this is an optimum sensor location where maximum signal strength and low spatial variation of SCG features were found [22]. The accelerometer was attached to the chest surface using double-sided medical-grade tape (B205–1, 3M, Minneapolis, MN). A signal conditioner (Model: 482C, PCB Piezotronics, Depew NY) was used to amplify the signals from the accelerometer. The amplified signal was acquired on a computer using a data acquisition system (IX-TA-220, iWorx Systems Inc., Dover, NH 03820, USA) and associated software (LabScribe Data Acquisition and Analysis Software, version 3.611700, iWorx Systems Inc., Dover, NH 03820, USA) at a 10,000 Hz sampling frequency. ECG and galvanic skin response (GSR) signals were also acquired simultaneously (IX-B3G Biopotential & GSR Recorder recording module). In a previous study, the GSR signal was found to best match the lung volume signal when one electrode was attached in the midclavicular line just below the right clavicle and another in the middle of the lower left abdominal quadrant [23]. The experimental setup and sensor placements are shown in Fig. 2.

Fig. 2.

(a) Block diagram of the experiment setup (b) Sensor locations. SCG, ECG, and GSR represent seismocardiogram, electrocardiogram, and galvanic skin response, respectively.

A schematic of the main steps of the study protocol is shown in Fig. 3. The protocol consisted of the following steps: (a) Subjects sat on a chair and waited for 10 minutes before data acquisition. This helps to reach a stable resting state. (b) Sensors were placed on the chest of the subject. (c) Data was acquired for 10 minutes at rest (Fig. 3(i)). (d) Subjects exercised by cycling on a stationary bike until the target heart rate for moderate exercise level was achieved (Equation 1) (Fig. 3(ii)). Typically, it took 10–15 minutes for subjects to reach their respective target heart rate. The equation of the target heart rate was obtained from [24].

Fig. 3.

Schematic representation of study protocol

Target heart rate = (220 − age) ∗ 0.7 (1)

(e) After achieving the target heart rate, subjects moved back to the chair (within 5 to 10 seconds) and data was acquired for another 10 minutes at the sitting position (Fig. 3(iii)).

Post-exercise data of one subject (subject #14) wasn’t recorded due to equipment malfunction.

3. Data Analysis

Data analysis was performed using a commercial software package (Matlab, Mathworks, Natick, MA). The analysis in this study is limited to the dorsal-ventral direction of the SCG signals. The signals were downsampled to 1000 Hz for faster processing time. Raw SCG and ECG signals were filtered to remove environmental noises (bandpass filter with a 0.05–100 Hz passband). GSR signals were filtered using a bandpass filter with a passband of 0.1–8 Hz. ECG R-waves were detected using the Pan-Tompkin algorithm [25]. ECG and SCG signals were segmented into ECG and SCG beats that started 0.1 seconds before the R peak of the corresponding ECG signal and ended at 0.1 seconds before the next R peak (Fig. 4).

Fig. 4.

Segmentation of electrocardiogram (ECG) and seismocardiogram (SCG) signals.

The SCG and ECG beats were assigned into eight respiratory groups [26]. The groups were based on the locations of systole (R to SCG2) and diastole (SCG2 to the beat end) relative to the first and second portions of inspiration and expiration (Fig. 5).

Fig. 5.

Representation of the 8 respiratory groups determined by the GSR and its derivative (airflow rate signal). SCG and ECG events were categorized into the 8 groups based on the location of systole and diastole in the respiratory phases. The transition from inspiration to expiration was shown by the solid black vertical line in the middle, where the GSR derivative shifted from positive to negative. The 1^st and 2^nd halves of inspiration and expiration were separated by the blue lines. Group 1: systole in the 1st half of the inspiration; Group 2: diastole in the 2^nd half of the inspiration; Group 3: transition from inspiration to expiration in diastole; Group 4: transition from inspiration to expiration in systole; Group 5: systole in the 1^st half of the expiration; Group 6: diastole in the 2^nd half of the expiration; Group 7: transition from expiration to inspiration in diastole; and Group 8: transition from expiration to inspiration in systole. The grouping was based on [26].

The respiratory phases were determined from the derivative of the GSR signal (which is related to airflow rate), where inspiration and expiration corresponded to positive and negative GSR derivatives, respectively. The border between the first and second portions of each repiratory phase (i.e., inspiration and expiration) was determined from the GSR signal zero crossing (Fig. 5).

After grouping ECG and SCG beats in the groups shown in Fig. 5, certain time-amplitude features were calculated from SCG and ECG beats as shown in Figs. 6 and 7.

Fig. 6.

Calculation of SCG RMS Amplitudes. A window is chosen for each event, then the RMS amplitude is calculated.

Fig. 7.

The cardiac timing intervals. ECG and SCG peaks were first detected (white triangles), then the time intervals were calculated.

A. Amplitude Calculation

The highest peak in the SCG1 region was selected, and a window of size 120 ms (20 ms before the peak and 100 ms after the peak) was chosen (window 1 in Fig. 6). The root mean square (RMS) value of this window was defined as the SCG1 amplitude. Similarly, the SCG2 amplitude was calculated. The window size for SCG2 was chosen as 100 ms (20 ms before the peak and 80 ms after the peak, window 2 in Fig. 6).

B. Cardiac intervals calculation

Pre-ejection period (PEP) is typically measured as the interval between the ECG Q-wave and the SCG1 peak. However, the Q-wave is often hard to determine with accuracy, and Pilz et al. demonstrated that using the R-wave as the starting point for calculating PEP may not significantly alter analysis [27]. In the current study, PEP was calculated as the interval between the R-wave and SCG1 peak, as it is easy to detect the R-wave with accuracy. The left ventricular ejection time (LVET) was defined as the interval between SCG1 and SCG2 peaks in the same cycle [28]. The time interval between the SCG2 peak and the R-wave of the subsequent beat was used to define diastole [29].

The features were calculated for each cycle in the eight respiratory groups during pre- and post-exercise states. The results are shown in the following section.

4. Results

A. Effect of exercise on SCG signal:

The effect of exercise on the amplitude of SCG1 and SCG2 was examined. Both amplitudes increased with exercise (Fig. 8(a)), except for subject 1 SCG2 (where a small amplitude decrease was seen). The amplitude increase was more noticeable in SCG1 than SCG2 in most of the subjects. Subject 12 showed an opposite trend (i.e., a higher SCG2 amplitude increase than SCG1), whereas changes were approximately similar for subjects 3 and 8. All the cardiac intervals (PEP, LVET, and diastole) decreased after exercise (Fig. 8(b)). When the intervals were calculated relative to cardiac cycle duration, diastolic time decreased, but PEP and LVET increased for the majority of the subjects (Fig. 8(c)).

Fig. 8.

(a) The percent change in the SCG1 (blue) and SCG2 (orange) amplitudes with exercise. The amplitudes increased after exercise except for subject 1’s SCG2 amplitude. (b) The percent change of the pre-ejection period-PEP (blue), left ventricular ejection time-LVET (orange), and diastolic period (yellow) post-exercise. Negative changes imply that all the cardiac intervals decreased post-exercise. (c) The percent change of cardiac intervals relative to cardiac cycle duration. Diastolic time (yellow) relative to cardiac cycle duration decreased, while relative PEP (blue) and LVET (orange) increased post-exercise for most of the subjects. To assess the immediate effect of exercise, the SCG features in this figure were computed using the entire pre-exercise data and the first ten SCG beats of the post-exercise data.

B. Effect of exercise on the amplitude modulation

Previous studies showed that respiration induces amplitude modulation of the SCG signal, including SCG1 and SCG2 waves [30]. The RMS amplitude modulation of SCG1 and SCG2 of a representative subject is shown in Fig. 9.

Fig. 9.

The RMS amplitude modulations of a representative participant (subject 4) for SCG1 (blue) and SCG2 (yellow) during a 5-minute pre-exercise window. Each data point (marked with a solid circle) indicates the timing of the ECG R-wave. SCG1 amplitude is higher than SCG2, and the modulation is higher for SCG2 in this subject. The inset boxes display amplitude modulations more clearly in a zoomed 10-second window.

Fig. 10(a) shows the post-exercise SCG1 amplitude, where a decaying amplitude can be seen during recovery from exercise effects. To quantify the amplitude modulation during recovery, a moving average trendline was generated (red line in the figure) and subtracted from the SCG1 amplitude to remove the slow amplitude decay. This will help focus on the fast oscillatory changes due to respiration.

Fig. 10.

(a) SCG1 RMS amplitudes (blue circles) during recovery and the associated trendline (red). The trendline was subtracted from the amplitudes to remove the decaying pattern. (b) The percentage changes of amplitude modulations for SCG1 (top) and SCG2 (bottom) relative to pre-exercise for early (circle), mid (triangle), and late (square) recovery phases. Modulations increased for most of the subjects right after exercise, except subjects 5 (SCG2) and 6 (SCG1 and SCG2). The changes in modulations decreased with recovery. This trend is more consistently observed in SCG1. (c) ECG R-wave amplitude change during recovery. The R-wave increased during the first 150 seconds of recovery and then decreased.

To compare respiratory amplitude modulation pre- and post-exercise (i.e., during recovery), the RMS of the amplitude was calculated at different time intervals: pre-exercise and post-exercise during early, mid-, and late-recovery. Here, the early recovery was represented by the first 50 post-exercise beats. The mid- and late-recovery were taken as the mid-50 beats (~40–45% of beats) and the last 50 beats (~90–95% of total beats), respectively. For the resting period, the standard deviations of all pre-exercise beats were calculated. The percentage change of standard deviations relative to the pre-exercise value is plotted in Fig. 10(b) for SCG1 and SCG2. This data shows that the amplitude modulations increased in most of the subjects right after exercise and decreased with recovery (Fig. 10(b)).

To compare with the SCG1 amplitude, the post-exercise trend in the ECG R-wave is shown in Fig. 10(c). Unlike SCG1, the R-wave amplitude increased at the beginning of the recovery and then decreased with time.

C. Effect of respiration on SCG features:

SCG features and heart rate were divided into the eight respiratory phases (described in the data analysis section) to observe and understand the effect of respiration on the SCG signal. Interval features and the heart rate varied with respiration but also varied in a much slower pattern, which is not related to respiration. This slow variation was extracted by finding trendlines (that are determined from the moving average) similar to what was done in Fig. 10(a). To reduce the effects of these slow phenomena, features were transformed by subtracting the trendline. Amplitude features were normalized by dividing by the respective trendlines after subtracting the trendlines to address the inter-subject variability issue. 10 minutes of pre- and post-exercise data were considered. More signal features are shown for all subjects in boxplot figures (Figs. 11–19). All statistical analysis was performed using one-way ANOVA to demonstrate the significant differences in features among the respiratory groups.

Fig. 11.

Heart rate (after subtracting the moving average heart rate) distribution among 8 respiratory groups for pre- (left) and post-exercise (right). Here, zero lines represent the trendlines.

Fig. 19.

SCG2 RMS amplitude distribution among the respiratory groups for pre- (left) and post-exercise (right).

1). Heart rate

Fig. 11 shows the heart rate for the eight respiratory groups described in the methods section. This data shows that in the pre-exercise phase, heart rates peaked in Group 2 during the second half of inspiration and then steadily declined, reaching their lowest point in Groups 6 and 7 between the mid- and second half of expiration.

For post-exercise, the heart rates of Groups 2 and 6 were the highest and lowest, respectively (similar to pre-exercise), while the group differences were significantly less than the pre-exercise period (p_pre=8.82e-26, p_post=4.11e-10).

2). Pre-ejection period

The pre-ejection period (Fig. 12) was longest at the second half of inspiration (Groups 2 and 3) and shortest at the second half of expiration (Groups 6 and 7). This trend was seen in both pre- and post-exercise states. The differences among the groups appeared smaller after exercise (p_pre=2.69e-9, p_post=1.20e-4), but more pronounced than heart rate differences (Fig. 11).

Fig. 12.

Distribution of pre-ejection period (after subtracting the mean) among the respiratory groups for the pre- (left) and post-exercise (right) states.

The change in PEP relative to the R-R interval is shown in Fig. 13. The trend is similar to that seen in Fig. 12, but group differences appear more pronounced for both the pre- and post-exercise phases (p_pre=1.87e-12, p_post=8.36e-6).

Fig. 13.

Distribution of pre-ejection period/R-R interval ratio among the respiratory groups for pre- (left) and post-exercise (right). The trend is similar to Fig. 12, but the group differences appear more pronounced for both the pre- and post-exercise phases (p_pre =1.87e-12, p_post=8.36e-6).

3). Left ventricular ejection time:

The results of LVET are shown in Fig. 14. This data shows that the LVET interval was lower for inspiration (Groups 1–3) and increased with expiration (Groups 5–7). The trend is more significant in the pre-exercise than post-exercise state (p_pre=7.47e-16, p_post=.005).

Fig. 14.

Distribution of left ventricular ejection time LVET among the respiratory groups for pre- (left) and post-exercise (right).

However, the trend in LVET variation among the respiratory groups changed when presented in relation to the R-R interval.

4). Diastolic time

The diastolic time distribution is shown in Fig. 16 for the eight respiratory groups. These results resembled those of LVET, with lower values at inspiration (Groups 1–3) and higher values at the second half of expiration (Groups 6–7). The differences among the groups were noticeably lower in post-exercise (p_pre=8.11e-17, p_post=5.17e-5).

Fig. 16.

Distribution of diastolic time intervals for pre- (left) and post-exercise (right).

When diastolic time relative to the R-R interval is plotted (Fig. 17), diastolic time showed a similar trend (p_pre=3.12e-18, p_post=6.19e-4).

Fig. 17.

Diastolic time change in relation to R-R interval for pre- (left) and post-exercise (right).

5). Amplitude of SCG1

The amplitude of SCG1 is shown in Fig. 18 for the different respiratory groups. This data showed lower amplitudes in groups 8, 1, and 2 compared to groups 4, 5, and 6 in the pre-exercise state but may not suggest clear trends for the post-exercise state as suggested by respective p-values (p_pre=.03, p_post=.76).

Fig. 18.

SCG1 RMS amplitude distribution among the respiratory groups for pre- (left) and post-exercise (right).

6). Amplitude of SCG2

The SCG2 amplitude is shown in Fig. 19 for the different respiratory groups. Unlike SCG1, SCG2 amplitude showed a trend for both the pre- and post-exercise states (p_pre=1.53e-12, p_post=1.33e-4). The amplitudes were higher in inspiration (Groups 1–3) and lower in expiration (Groups 5–7) for pre-exercise values. At post-exercise, a leftward shift in trends was observed, with the highest SCG2 amplitude found in Groups 1–2 and the lowest in Groups 5–6.

5. Discussion

This study investigates the changes in SCG waveforms with respiration and exercise. The changes are quantified by the time interval and amplitude features of the SCG beats. By classifying the beats into eight respiratory phases, the effect of respiration on SCG was assessed in both the pre- and post-exercise states. The effect of exercise was observed by calculating the features of the first ten SCG beats immediately following the exercise and comparing those to those of pre-exercise data. The standard deviations of the SCG1 and SCG2 amplitude plots (after removing trendlines) at rest and during the early, mid, and late recovery phases are used to measure the effect of exercise on the SCG1 and SCG2 amplitude modulations.

A. Effect of exercise on SCG signal

Both SCG1 and SCG2 amplitudes increased after exercise (Fig. 8(a)). The higher amplitudes after exercise can be explained by the physiological changes due to exercise. These include enhanced cardiac output owing to higher heart rate and stroke volume, increased contractility, and increased blood pressure [31], [32]. Interestingly, the increase in SCG1 was noticeably higher than SCG2 in 11 out of 14 subjects (Fig. 8(a)). She et al. also found similar results in their study, where the intensity of the first heart sound increased more significantly than the second heart sound after exercise [33]. A higher SCG1 magnitude could indicate stronger ventricular contractility. The small change in SCG2 could be a sign of small alterations to the pressure gradient caused by aortic rebound.

All the cardiac intervals decreased after exercise (Fig. 8(b)), which is in line with the previous findings [26]-[28]. The increased heart rate with exercise results in shorter cardiac beats, which may be responsible for decreased intervals, particularly in left ventricular ejection time (LVET) and diastolic time. PEP decreased due to increased contractility and was less affected by heart rate [17]. Diastolic time decreased more than LVET, but the changes were consistent among subjects (Fig. 8(b)), suggesting that the reductions may be mediated by the same factor(s). On the other hand, reductions in pre-ejection period (PEP) were not as consistent as LVET and diastolic time.

PEP and LVET relative to cardiac cycle duration exhibited opposite trends compared to the absolute values; for most of the subjects, their relative values increased after exercise (Fig. 8(c)). However, the relative diastolic time decreased post-exercise. This supports the findings of a previous study by Fridericia et al. [37].

B. Effect of exercise on amplitude modulation

The respiratory amplitude modulations of SCG1 and SCG2 increased for most of the subjects after exercise (Fig. 10(b)), except subjects 5 (SCG2) and 6 (SCG1 and SCG2). The modulations decreased and shifted toward the baseline with exercise recovery.

Previous studies found that SCG1 and SCG2 amplitude modulations are caused by respiration [30]-[32]. Exercise causes an increase in breathing rate and tidal volume. This increase in respiration activity may result in increased modulations of SCG1 and SCG2 amplitudes. As respiratory parameters returned to baseline with recovery, the changes in modulations also decreased. This trend is more clearly observed in SCG1 than in SCG2 amplitude.

C. Effect of respiration on SCG features

1). Heart Rate

Differences in heart rate among the respiratory phases can be explained by respiratory sinus arrhythmia (RSA) [41]. During inspiration, the thoracic cavity expands as the diaphragm moves downward. This causes the intrathoracic pressure to fall below the intra-abdominal pressure, which compresses the inferior vena cava. This, in turn, squeezes more blood into the right atrium of the heart. Sympathetic signals are issued to the heart after atrial stretch receptors detect increased blood volume. This increases the heart rate during inspiration.

Although RSA can be detected in both pre- and post-exercise states, its magnitude was noticeably higher in pre-exercise. Previous studies found that RSA decreased during and after exercise, which supports the result of the current study [42]. In this study, RSA was calculated from the spectral power of heart rate in the frequency bands centered at the breathing rate [43]. For pre-exercise data, the entire recording was selected to calculate RSA. Recovery was divided into 3 phases (early, mid, and late) following the same procedure described in the amplitude modulation section. The results are shown in Fig. 20.

Fig. 20.

Pre-exercise RSA (blue) was significantly higher in all 14 subjects than post-exercise RSAs. The difference between pre- and post-exercise RSAs was statistically significant, as shown by the p-value (p<0.01). RSA increased with recovery, which can be observed by comparing early (orange), mid (yellow), and late (purple) recovery RSAs.

2). Pre-ejection period (PEP) and left ventricular ejection time (LVET)

PEP was higher in inspiration than expiration in both pre- and post-exercise states, confirming the results of previous studies [44]. During inspiration, the pleural pressure decreases, resulting in a relative aortic pressure increase. This causes the afterload to increase, and consequently, a longer contraction time is required, delaying both mitral closure and aortic opening. The increase in PEP at inspiration is consistent with this mechanism [26]. The opposite happens during expiration.

The decrease in LVET during inspiration can be explained with the same reasoning. As aortic pressure is relatively higher at inspiration, the aortic valve is expected to close sooner, shortening LVET. Post-exercise data follows similar changes with respiration.

3). Diastolic time

Diastolic time was found to be strongly correlated with heart rate in previous studies. This interval was significantly more impacted by the heart rate compared to ejection time [37]-[39]. As the heart rate was faster (i.e., smaller R-R interval) at inspiration (Groups 1–3), the diastolic time of associated beats also became shorter. The reverse pattern was observed in Groups 6–7 during expiration (lower heart rate, longer diastolic time). In post-exercise, as respiratory sinus arrhythmia decreased, the differences in diastolic time among the groups also decreased accordingly. This further confirms the strong dependency of diastolic time on heart rate.

Greater group overlap was observed in LVET and diastolic time (Figs. 14 and 16) following exercise compared to PEP. This may be explained by the dominant effect of heart rate on LVET and diastolic time [45], [48]. Mertens et al. showed that heart rate doesn’t affect PEP; rather, other factors such as conduction velocity, LV end-diastolic pressure, and volume have greater influences [48]. Decreased RSA caused by an increase in heart rate impacted post-exercise LVET and diastolic time by reducing the differences among the respiratory groups (Figs. 14 and 16), which were less observed in PEP (Fig. 12). This also supports that the heart dynamic states, rather than an increase in heart rate, were primarily responsible for the decrease in PEP with exercise (Fig. 8(b)). LVET and diastolic time both decreased due to heart rate increase; hence, a synchronous pattern was observed in these two intervals (Fig. 8(b)).

PEP and diastolic time followed similar respiratory variability when represented relative to the R-R interval (Figs. 13 and 15). On the contrary, LVET exhibited a change in trend when standardized by R-R interval (Fig. 15). Although the absolute LVET decreases with increased heart rate (inspiration), its relative value (with respect to the R-R interval) increases, meaning the drop in the R-R interval is larger than the drop in LVET [37]. The opposite effect is seen at expiration. The post-exercise LVET/R-R interval ratio didn’t show significant group differences, which implies that the variation in postexercise LVET may have been balanced by corresponding heart rate changes.

Fig. 15.

Distribution of left ventricular ejection time relative to R-R interval among the respiratory groups for pre- (left) and post-exercise (right) phases. The trend is opposite in pre-exercise, exhibiting higher values at inspiration and lower values at expiration (p_pre =9.63e-10). The features didn’t show any significant group differences post-exercise. (p_post=.87).

4). SCG1 amplitude:

The SCG1 amplitude didn’t show a consistent trend among the subjects. Tang et al. found similar results when they studied the effect of respiration on the first heart sound (S1) [49]. But Xiao et al. reported an increase in S1 intensity at inspiration [50], while Ishikawa et al. found the opposite (louder S1 at expiration). In the current study, during the pre-exercise state, SCG1 amplitude was higher at mid-inspiration in 8 of 15 subjects. In 4 subjects, SCG1 was higher at mid-expiration, and in the remaining 3 subjects, SCG1 amplitude was higher at late-inspiration and early-expiration. In the post-exercise state, 7 subjects showed increased SCG1 amplitude at mid-inspiration, 4 subjects at mid-expiration, and the remaining subjects demonstrated higher SCG1 in the transition. Since the SCG1 amplitude may be influenced by multiple factors, this may have resulted in variations in behavior among subjects. Data from more subjects is needed to understand this variability.

5). SCG2 Amplitude

SCG2 amplitude was found to be higher in inspiration, consistent with the findings reported by Tang et al. [49]. Due to the relative increase in aortic pressure during inspiration, the aortic valve may close with greater force, which can increase SCG2 amplitude. Additionally, the lower intrathoracic pressure during inspiration may aid in the relaxation of the left ventricle (LV), causing the pressure gradient between the LV and aorta to increase. This can also increase the SCG2 amplitude.

One limitation of this study is the subjects’ narrow age range and BMI. The study population was relatively small, and it only included healthy subjects. Future research should investigate larger cohorts of both normal subjects and those with different cardiopulmonary conditions, comprising wider ranges of age and BMI. In this study, a target heart rate was achieved by a moderate level of exercise, which may not be feasible for the patients to perform. The future protocol should involve low-level exercise performed at a comfortable pace for the subject. This can include walking, step-ups, or seated exercises if the subject is unable to stand up. Data were acquired and analyzed before and after the exercise in this study. In future studies, data collection can be continually carried out during the exercise. This may provide additional insights into the dynamic physiological responses occurring throughout the exercise, enhancing understanding of cardiopulmonary interactions. Wearable devices or patches would be more appropriate for data collection during exercise because they may be less sensitive to motion-induced noises. Future studies may also include automated real-time data analysis and artificial intelligence-based calculations for clustering SCG events and feature extraction [51–54]. Potential clinical applications of SCG include monitoring and diagnosis of cardiopulmonary conditions such as heart failure.

6. Conclusion

This study investigated the respiration-induced variability in the SCG signal and the effect of exercise on this signal in healthy subjects. SCG features were calculated in pre- and post-exercise states to observe the changes in the SCG waveform features with exercise. The features were also investigated during different respiratory phases to see how respiration affects the variability of the SCG signal. The results showed that alterations in the SCG signal may be reflective of cardiopulmonary physiological changes. To our best knowledge, exercise and respiratory effects of SCG signal have not been described in detail previously. This will provide insights on SCG variability and help in the feature extraction procedure for clinical diagnosis. Further characterization and understanding of these SCG feature changes may increase the clinical utility of this method in the non-invasive detection and monitoring of patients with cardiopulmonary conditions.

• Respiratory variability of seismocardiogram (SCG) signal has been explored.

• Effect of exercise on the SCG features has been investigated.

• Study results can potentially enhance the clinical utility of SCG signal.