Hemorrhage is the loss of blood from the circulatory system and a major cause of death, particularly in car accident trauma and on the battlefield (6). The early application of lifesaving interventions, before the development of circulatory shock, is a priority in dealing with hemorrhage. Mortality can be significantly reduced when patients needing immediate care can be identified early. Traditionally, patients are diagnosed based on their mental status, pulse, and systolic blood pressure (dropping below 90 mmHg). However, it may be impossible to rescue patients when significant hypotension has developed. Thus, pulse pressure (PP; difference between systolic and diastolic blood pressure) has been proposed for early noninvasive identification of hemorrhage, when the systolic blood pressure is greater than 90 mmHg (6). However, the association of PP to stroke volume has been recently challenged, and the equivalent sensitivity of PP to blood loss has not been validated (11).
In all relevant laboratory studies, hemorrhage is simulated by creating graded lower-body negative pressure (LBNP) to replicate the hemodynamic and autonomous effects of hemorrhage (8, 11, 15, 19, 23). Similar to hemorrhage, LBNP reduces venous return and thus preload, resulting in decreased stroke volume and cardiac output, which is explained by Frank-Starling law. The response to hemorrhage and LBNP is divided in three phases of mild, moderate, and severe. In the first phase, sympathetic activity increases resulting in maintenance of, or slight increase in, blood pressure, when less than 10% of total blood has been lost. In the second phase, when there is a blood loss of up to 20%, the sympathetic nervous system fails to compensate for the loss, and finally in the third and last phase, which is severe hemorrhage, tachycardia is present with a failure to maintain blood flow to the vital organs resulting in death. The last phase starts when more than 20% of blood has been lost (approximately 1 L). For an effective diagnosis and treatment, it would be ideal to detect the progression to phase 2 before phase 3 occurs (8). Lower-body negative pressure, during which the blood is translocated to the lower portions of the body, provides a safe tool to reproduce conditions similar to hemorrhage, while maintaining total blood volume.
Recently, more sophisticated systems have been proposed for accurate detection of hemorrhage, which include simultaneous recording of several biosignals from a pulse oximeter, impedance plethysmogram, blood pressure, and blood flow (7, 21) or using Doppler ultrasonography (9). Machine learning algorithms have been used to estimate the level of blood loss as produced by LBNP (7). However, such a system, although more accurate, might not be a feasible solution in places where hemorrhage occurs the most, such as in automobile and industrial accidents or battlefield. It can take a considerable amount of time to apply the sensors and to ensure that they are calibrated and functioning properly. Thus, there is still a need for a less complicated solution.
The seismocardiogram (SCG) is a precordial vibration signal, a manifestation of low-frequency acceleration signals from the chest (3, 24, 31). The SCG signal can be recorded from an accelerometer placed on the sternum or with a wearable device (4). The electrocardiogram (ECG) signal has been used in the past to segment and annotate SCG peaks. The recorded ECG signal does not even need to be a standard lead ECG (5). However, there are now methodologies for segmentation of SCG without the use of other cardiac signal (2). Pulse plethysmogram signal is also a potential for segmenting SCG signal if ECG is difficult to record. Several studies in the past have proposed the relation of the SCG signal to myocardial contractility and stroke volume (4, 14, 22, 27). With this background, in this research, SCG is proposed as a simple low-cost alternative for diagnosis of hemorrhage. Similar to previous studies, LBNP was implemented as a surrogate of hemorrhage.
MATERIALS AND METHODS
Lower-body negative pressure
Each participant’s lower body was placed in a negative-pressure chamber and sealed at the iliac crest (Fig. 1). Vacuum was applied to the chamber to drop the box pressure to −20, −30, −40, and −50 mmHg progressively. Two sets of test were conducted. In the first, 18 participants were kept at each stage for 5 min and were returned to normal pressure at the end of the −50 mmHg stage with recording continued for another 5 min of rest. In the second, with 30 participants, following −50 mmHg the pressure was gradually increased to normal pressure in reversed steps. If a participant exhibited a sudden decrease in heart rate or blood pressure or if they expressed any discomfort and wanted to stop, the negative pressure was immediately terminated.
Stroke volume measurement and estimation
In the first set, which included 18 participants, stroke volume was recorded using an echocardiogram device (Vividi; GE, Milwaukee, Wisc) by an experienced echocardiographer. For every participant, a minimum of three consecutive cardiac cycles were chosen, and Doppler ultrasound was used to measure the stroke distance. This was then multiplied by their left ventricular output track area to yield the stroke volume. However, we were not able to record Doppler from many cycles in the participants as the movement of the tip of the echo probe distorted the SCG morphology. Therefore, we recorded a sufficient number as to be able to approximate the stroke volume at each level of LBNP and to validate the stroke volume estimation methodology used with the Portapres (Finapres Medical Systems, Amsterdam, the Netherlands).
In all participants, continuous noninvasive finger arterial pressure was measured using a Portapres device (Finapres Medical Systems, Amsterdam, the Netherlands), which is based on a photoplethysmographic technique. The finger cuff was applied to the midphalanx of the middle finger of the left hand. Pulse pressure was measured (systolic blood pressure minus diastolic blood pressure), and stroke volume was estimated from the recorded finger arterial pressure with the Beatscope software (Finapres), which uses the Modelflow technique to estimate stroke volume from PP (18). This technique has been validated against thermodilution (29) and Doppler ultrasound (18) for estimation of stroke volume and cardiac output. In this study, we also compared the stroke volume, measured using echocardiography with the Portapres to further verify that the Portapres is indeed capable of estimating stroke volume. It has been demonstrated, in a phlebotomy study, that within-subject changes in stroke volume closely mirrored the blood withdrawal (17). The blood withdrawal (17) is similar to the LBNP effect, as both simulate a mild hemorrhage (8, 10).
The SCG signal was measured with a high-sensitivity accelerometer as used by Castiglioni et al. (4) (Brüel and Kjær model 4381, Nærum, Denmark). The participants were in the supine position, and the signals were recorded in back-to-front direction, perpendicular to the body surface. The ECG signal was also acquired and used to segment the cardiac cycles. All signals were recorded using an NI 9205 analog input module (National Instruments, Austin, Tex). A snapshot of the recorded signals can be seen in Figure 2 with the SCG signal annotated as proposed by Salerno and Zanetti (24).
A total of 48 participants took part in this study including 16 female (aged 29.9 ± 4.1 years, weight, 60.1 ± 10 kg; height, 166.3 ± 6 cm) and 32 male subjects (aged 28.5 ± 4.3 years; weight, 78 ± 10 kg; height, 177.5 ± 5.3 cm). None had any documented cardiac abnormality. Biosignals were recorded at the Aerospace Physiology Laboratory under an ethics approval from the Simon Fraser University Research Ethics Board. The participants followed the informed consent procedure and signed their consent form. Participants were compensated by a fee for their time.
SCG signal annotation and feature extraction
Automatic annotation software was developed in MATLAB (Mathworks USA) to extract the R wave of ECG, and MC, IM, AO, and AC points of the SCG signal as shown in Figure 1 and defined by Leonetti et al. (17). The software initially detected the ECG R-wave using an algorithm based on filter banks (1). The Q wave was detected by finding the local minima before the detected R-wave peak. Based on the location of the Q wave, the MC point was assigned as the first maximum on the SCG after the Q wave. The IM and AO points of the SCG were assigned consecutively as the minimum and maximum after MC point. The AC point was assigned as the local minimum in a window (100 ms long) after about 300 to 400 ms of the ECG Q wave. The software annotations were imported to the HFM Waveform Analysis software (Heart Force Medical, Vancouver, British Columbia, Canada) and were manually corrected. From 48 participants, a total of 41,617 heartbeats were annotated.
Using the corrected SCG annotation, 34 different features were extracted in two categories: timing and amplitude. The timing features included Q-MC, Q-AO, Q-MI, Q-AC, and AO-AC. The amplitude features were in three subgroups of amplitude values (MC, AO, MI, MI-AO), slopes (MI to AO, MC to MI, MA to RE), and root mean squares (150 ms after R wave, during isovolumic contraction period).
Systolic time interval extraction from SCG
The timing features mentioned in the previous section correspond to the systolic time intervals as defined in the noninvasive cardiology literature (28). Q-AO is pre-ejection period (PEP), Q-AC corresponds to QS2, AO-AC is left ventricular ejection time (LVET), and Q-MC is electromechanical delay or EMD.
The initial annotation of AO and AC points, where the major systolic time intervals are derived from, was proposed by Zanetti and Salerno (30). As they used older echocardiograph systems for these annotations, a new study was performed, by the first and last author of the article, further confirming the original annotations for AO and AC points using Doppler echocardiography and impedance cardiography (25).
The JMP10 software (SAS Institute, Arlington, Va) was used for statistical analysis. A one-way repeated-measures analysis of variance was used to evaluate if there were statistically significant changes in SCG-extracted parameters caused by LBNP.
LBNP effect quantification
As can be seen in Figure 2, the echocardiogram measurements show a gradual drop in the stroke volume through the LBNP stages for the first 18 subjects, corresponding to an overall drop of 32.4% in stroke volume from the resting pre-LBNP stage to the peak negative pressure of the LBNP. The LBNP setup used in these experiments was capable of inducing a hemodynamic effect similar to mild, moderate, and severe hemorrhage (early stage) as defined in the literature (6).
As can be seen in the last two columns of Table 1, the Portapres device also detected the same scale of reduction in stroke volume compared with the echocardiogram. The correlation coefficient between the two devices was 0.94 (P < 0.05). The capability of the Modelflow technique in the estimation of stroke volume changes has also been previously investigated as mentioned in Materials and Methods, and the current results reaffirm these findings and justify the use of the Portapres as an estimate of stroke volume changes.
SCG parameter selection
Thirty-four different morphological features were extracted from every SCG cycle. The correlation coefficients between these features and the stroke volume values, estimated from the Portapres signal, were calculated. Except for four subjects, there was at least one SCG feature with a correlation coefficient of more than 0.9 (P < 0.05) with stroke volume.
Forty-two of 48 participants had correlated features in the timing category, in particular the PEP, QS2, LVET, and PEP/LVET parameters. The averages of the correlation coefficients over all 48 participants for these features were −0.84 ± 0.15, 0.82 ± 0.17, 0.89 ± 0.12, and 0.90 ± 0.11, respectively (P < 0.05). Pre-ejection period/LVET is the dominant feature that is in line with previous preliminary results and is confirmed by a larger number of subjects (26, 27). The remaining six subjects had their maximum correlated features with stroke volume in the amplitude category.
Changes of SCG features with LBNP stages
To evaluate the effects of LBNP stages on the SCG features, it was important to know the exact times where stages were switched. In 16 participants from the second set of experiments, these timings were not available, sometimes because of the discomfort expressed by the subjects during recordings, and thus only a total of 32 participants were used to derive the results presented in Table 1 and Figure 3. Table 1 lists the percentage changes of different SCG parameters and also the change in heart rate and PP with LBNP. The last row of the table lists the correlation coefficient of each of these parameters with the LBNP levels. Among the SCG features, LVET and PEP/LVET had the highest correlation coefficients. These are plotted as a function of LBNP together with heart rate and PP (Fig. 3). As expected, the stroke volume was highly correlated with the LBNP stages.
Although amplitude components of SCG were not the dominant features, the overall results for two of them (peak-to-peak and slope of SCG during isovolumic contraction period) are also presented in Table 1. Although the overall results for amplitude showed a gradual decline with LBNP, in 19 subjects the trend fluctuated, and there were increases in the amplitude of the signal, as LBNP progressed.
Two SCG-derived features showed a strong relationship with LBNP. These were the systolic time interval of LVET and systolic index of PEP/LVET with correlation coefficients of 0.9 and 0.88, respectively. The PEP/LVET feature had the greatest correlation with the stroke volume values estimated from the Portapres device (R = 0.90). This finding is in line with previous findings that correlate the PEP/LVET index with ejection fraction and myocardial contractility (12).
It is of paramount importance to diagnose hemorrhage as it progresses to phase 2 and before it reaches phase 3. With respect to the LBNP stages in our study, this is equivalent to the sensitivity of detecting the change between LBNP levels −20 and −30 mmHg as they are very important precursors to the imminent stages resembling severe hemorrhage. Based on Table 1, PEP/LVET had an overall increase of 15.7% and 26% in these two first stages of LBNP, whereas neither PP nor heart rate showed any significant change in these two stages. In the first stage of LBNP, there was an increase in the overall PP, which may be due to sympathetic activity. After this stage, there is a gradual decrease in PP peaking to 7.5% at the maximum of LBNP negative pressure (Table 1 and Fig. 3). This further confirms the findings of Cote et al. (11) versus Convertino et al. (6), suggesting that PP alone is not sensitive enough to changes in stroke volume induced by LBNP.
The effect of LBNP on the SCG-derived features, heart rate, and PP was analyzed using one-way repeated-measures analysis of variance. If a main effect of LBNP was detected, subsequent post hoc analyses using the Tukey honestly significant difference test were performed (P ≤ 0.05). These showed that LVET was significantly different in all LBNP stages. Pre-ejection period/LVET was different between rest, −20 mmHg, and −30 mmHg but not significantly different between −30 mmHg and −40 mmHg. As well, between −40 and −50 mmHg was not different from each other but was different from all other levels of LBNP. Heart rate was not significantly different between rest and −20 and −30 mmHg, and PP did not change significantly over all stages.
These data suggest that parameters such as LVET or PEP/LVET could be better candidates for the early detection of mild and moderate hemorrhage (corresponding to −20 and −30 mmHg stages of LBNP) compared with PP and heart rate. However, PEP/LVET was not as sensitive in higher levels of LBNP as LVET, which was significantly different in all stages. Left ventricular ejection time also showed a high correlation with the stages of LBNP. The right side of Figure 2 shows the changes of LVET for one of the subjects during the LBNP. Left ventricular ejection time has been previously proposed for the detection of hypovolemia (13).
A variety of extracted SCG features were compared in this study with the intent of selecting the best features for every subject. It was observed that the features extracted from the amplitude of the SCG signal were not as correlated in comparison with the timing features. As mentioned in Results, for 19 subjects the amplitude actually increased, which was against an overall expected decrease, with LBNP. Based on these data, the systolic time intervals were the best correlate with stroke volume and appear to be more sensitive to the hemodynamic changes induced by LBNP.
In summary, these results suggest the potential for a patient-specific method, based on seismocardiography, to monitor progressive and worsening hemorrhage. Using this method, a baseline recording of an SCG-derived parameter, such as LVET, would be obtained upon arrival of a patient at the emergency room, provided the patient is in a reasonably stable condition. This baseline measurement will be variable, as in healthy populations. However, any significant change in the SCG-derived parameter, relative to the baseline assessment, can be used to trigger reassessment as a result of potential hemorrhage, which would not be noticed by current measurements using blood pressure, as demonstrated in this article. Finally, although the SCG signal in this experiment was collected from the sternum, there are a number of other placement areas on the chest that can be used if the sternum is not accessible (3).
LIMITATIONS AND FUTURE WORK
In this work, we used Doppler echocardiography for assessing the effects of LBNP on the stroke volume. We were not able to record Doppler signals for the complete length of time the participants were in the LBNP chamber. As well, measurements were not made in the ideal lateral decubitus body position for acquisition of a quality signal. Thus, the echocardiographer had difficulty securing a proper window for measuring the stroke distance. Furthermore, the movement of the echo probe induced artifacts on the SCG signal. As a result, we verified the use of the Portapres with echocardiography for long-time beat-to-beat estimation of stroke volume. The use of the Portapres-derived stroke volume was limited to the feature selection, and the results presented in Data Acquisition and Table 1 and Figure 3 are all independent of the stroke volume and represent the changes with regard to LBNP stages alone.
The ethics approval for this study was limited to −50 mmHg of LBNP. This prevented us from observing presyncopal conditions in any of the subjects. However, in a future study, we intend to decompensate the subjects by keeping them at −60 mmHg for a longer period (10). This would enable us to evaluate the capability of SCG in predicting severe hemorrhage. It should be noted that although LBNP is widely accepted by researchers in this field as a surrogate for hemorrhage, SCG technology needs to be clinically verified on actual emergency room patients. Before reaching that stage, an invasive animal study could also be useful (16).
As mentioned in Materials and Methods, manual annotation was also used to check for the possible misannotation of the software developed in MATLAB. Further work is required to make this into a reliable automated system; however, because of the nature of SCG morphology and between-subject variability, the current software requires manual observation. On this note, we are developing algorithms to reduce such errors by the incorporation of more advanced signal processing techniques and the inclusion of higher-frequency components of the same acceleration signal.
Noncardiac or background vibrations can affect SCG morphology, the accuracy of annotation, and subsequent parameter estimations. For measurement of the LVET, only a few properly recorded cycles are sufficient to provide the required information; however, in presence of motion artifact, the processing of SCG to provide these values becomes more challenging. Nevertheless, we need to have in mind that there is a spectrum of patients affected by hemorrhage. On one end of the spectrum, we have awake, mobile patients, and on the other end, we have unconscious intubated and ventilated patients. In the middle of the spectrum, we have patients who are sedated, thus drowsy and mechanically stable. In the category of awake patients, we may have those who are agitated and noncooperative as well as those who are in control and cooperating. There will be challenges for recording SCG from those who are agitated, but for the remaining patients, we expect that SCG should be able to be recorded properly. However, we should point out that the primary objective of this monitoring technology is to provide an alarm for those patients who are most probably unable to communicate their worsening conditions. Such patients are left alone because either they have received a quick treatment or they have not received priority. In all likelihood, the agitated, in-pain patients will attract attention and will receive current standard of care without the need for this monitoring device.
In the past, we have recorded usable SCG in subjects with heart rates up to 150 beats/min on treadmills and also in this article during LBNP. For heartbeats around 200 beats/min, extreme tachycardia, we do not expect to have good-quality SCG signals, because the heart mechanical function is significantly compromised by the short time available for myocardial relaxation. This will lead to unreliable parameters extracted from SCG, which by itself is a warning, not considering the fact that the goal of using this monitoring system is to avoid reaching a situation like this. Upon reaching extreme tachycardia, ECG, pulse plethysmogram, or any other signal that extracts the heart rate is sensitive enough to trigger an alarm, and there is no particular need for SCG.
The authors thank the members of Aerospace Physiology Laboratory at Simon Fraser University for their assistance during data acquisition. They also thank F. Khosrow-khavar, G. Jahns, and J. Fitz-Clarke for their assistance.
1. Afonso VX, Tompkins WJ, Nguyen TQ, Luo S: ECG beat detection using filter banks. IEEE Trans Biomed Eng
46: 192–202, 1999.
2. Akhbardeh A, Kaminska B, Tavakolian K: BSeg++: a modified blind segmentation method for Ballistocardiogram cycle extraction. In: Institute of Electrical and Electronics Engineers, Engineering in Medicine and Biology Conference, August 2007, Lyon, France. pp. 1896–1899.
3. Baevskii RM, Egorov AD, Kazarian LA: The method of seismocardiography. Kardiologiia
18: 87–89, 1964.
4. 4. Castiglioni P, Faini A, Parati G, Di Rienzo M: Wearable seismocardiography. Presented at the Institute of Electrical and Electronics Engineers, Engineering in Medicine and Biology Conference, August 2007, Lyon, France. pp. 3954–3957.
5. Chuo Y, Tavakolian K, Kaminska B: Evaluation of a novel integrated sensor system for synchronous measurement of cardiac vibrations and cardiac potentials. J Med Syst
35: 445–455, 2011.
6. Convertino VA, Cooke WH, Holcomb JB: Arterial pulse pressure and its association with reduced stroke volume during progressive central hypovolemia. J Trauma
61: 629–634, 2006.
7. Convertino VA, Moulton SL, Grudic GZ, Rickards CA, Hinojosa-Laborde C, Gerhardt RT, Blackbourne LH, Ryan KL: Use of advanced machine-learning techniques for noninvasive monitoring of hemorrhage. J Trauma
71: S25–S32, 2011.
8. Cooke WH, Ryan KL, Convertino VA: Lower body negative pressure as a model to study progression to acute hemorrhagic shock in humans. J Appl Physiol
96: 1249–1261, 2004.
9. Corradi F, Brusasco C, Garlaschi A, Santori G, Vezzani A, Moscatelli P, Pelosi P: Splenic Doppler resistive index for early detection of occult hemorrhagic shock after polytrauma in adult patients. Shock
38: 466–473, 2012.
10. Cote AT, Bredin SSD, Phillips AA, Warburton DER: Predictors of orthostatic intolerance in healthy young women. Clin Investig Med
35: E65–E74, 2012.
11. Cote AT, Phillips AA, Bredin SSD, Warburton DER: A questionable association of stroke volume and arterial pulse pressure under gravitational stress. J Trauma Acute Care Surg
72: 708–712, 2012.
12. Garrard CL, Weissler AM, Dodge HT: The relationship of alterations in systolic time intervals to ejection fraction in patients with cardiac disease. Circulation
42: 455–462, 1970.
13. Geeraerts T, Albaladejo P, Declère AD, Duranteau J, Sales J-P, Benhamou D: Decrease in left ventricular ejection time on digital arterial waveform during simulated hypovolemia in normal humans. J Trauma Inj Infect Crit Care
56: 845–849, 2004.
14. Gemignani V, Bianchini E, Faita F, Lionetti V, Campan M, Recchia FA, Picano E, Bombardini T: Transthoracic sensor for noninvasive assessment of left ventricular contractility: validation in a minipig model of chronic heart failure. PACE
33: 795–803, 2010.
15. Goswami N, Roessler A, Lackner HK, Schneditz D, Grasser E, Hinghofer-Szalkay HG: Heart rate and stroke volume response patterns to augmented orthostatic stress. Clin Auton Res
19: 157–165, 2009.
16. Guzman JA, Lacoma FJ, Najar A, Kruse JA: End-tidal partial pressure of carbon dioxide as a noninvasive indicator of systemic oxygen supply dependency during hemorrhagic shock and resuscitation. Shock
8: 427–431, 1997.
17. Leonetti P, Audat F, Girard A, Laude D, Lefrère F, Elghozi J-L: Stroke volume monitored by modeling flow from finger arterial pressure waves mirrors blood volume withdrawn by phlebotomy. Clin Auton Res
14: 176–181, 2004.
18. Van Lieshout JJ, Toska K, van Lieshout EJ, Eriksen M, Walløe L, Wesseling KH: Beat-to-beat noninvasive stroke volume from arterial pressure and Doppler ultrasound. Eur J Appl Physiol
90: 131–137, 2003.
19. Lindenberger M, Länne T: Reduced defense of central blood volume during acute lower body negative pressure-induced hypovolemic circulatory stress in aging women. Shock
37: 579–585, 2012.
20. Marcus FI, Sorrell V, Zanetti J, Bosnos M, Baweja G, Perlick D, Ott P, Indik J, He DS, Gear K: Accelerometer-derived time intervals during various pacing modes in patients with biventricular pacemakers: comparison with normals. PACE
30: 1476–1481, 2007.
21. McGrath SP, Ryan KL, Wendelken SM, Rickards CA, Convertino VA: Pulse oximeter plethysmographic waveform changes in awake, spontaneously breathing, hypovolemic volunteers. Anesth Analg
112: 368–374, 2011.
22. McKay WP, Gregson PH, McKay BW, Militzer J: Sternal acceleration ballistocardiography and arterial pressure wave analysis to determine stroke volume. Clin Investig Med
22: 4–14, 1999.
23. Ryan KL, Rickards CA, Ludwig DA, Convertino VA: Tracking central hypovolemia with ECG in humans: cautions for the use of heart period variability in patient monitoring. Shock
33: 583–589, 2010.
24. Salerno D, Zanetti J: Seismocardiography for monitoring changes in left ventricular function during ischemia. Chest
100: 991–993, 1991.
25. Tavakolian K, Blaber AP, Ngai B, Kaminska B, Science E, Columbia B, Physiology B: Estimation of hemodynamic parameters from seismocardiogram. In: Computing in Cardiology
2010: 1055–1058, 2010.
26. Tavakolian K, Dumont GA, Blaber AP, Physiology B: Analysis of seismocardiogram capability for trending stroke volume changes: a lower body negative pressure study. Computing in Cardiology
39: 733–736, 2012.
27. Tavakolian K, Portacio G, Ngai B, Jahns G, Blaber AP: Myocardial contractility: a seismocardiography approach. Presented at Institute of Electrical and Electronics Engineers, Engineering in Medicine and Biology Conference, August 2012, San Diego, California. 2012, pp. 3801–3804.
28. Weissler AM, Harris WS, Schoenfeld CD: Systolic time intervals in heart failure in man. Circulation
37: 149–159, 1968.
29. Wesseling KH, Jansen JR, Settels JJ, Schreuder JJ: Computation of aortic flow from pressure in humans using a nonlinear, three-element model. J Appl Physiol
74: 2566–2573, 1993.
30. Zanetti J, Salerno D: Seismocardiography: a new technique for recording cardiac vibrations. Concept, method, and initial observations. J Cardiovasc Technol
9: 111–120, 1990.
31. Zanetti JM, Tavakolian K: Seismocardiography: past, present and future. Presented at Institute of Electrical and Electronics Engineers, Engineering in Medicine and Biology Conference, Osaka, Japan July 2013. pp. 7004–7007.