Proceedings of the 29th Annual International

Conference of the IEEE EMBS

FrD06.2
Cit Internationale, Lyon, France
August 23-26, 2007.

Wearable Seismocardiography
Paolo Castiglioni, Andrea Faini, Gianfranco Parati, and Marco Di Rienzo, Member, IEEE

Abstract Seismocardiogram (SCG) is the recording of body
vibrations induced by the heart beat. SCG contains
information on cardiac mechanics, in particular heart sounds
and cardiac output. In this paper we present a new wearable
device for SCG recordings during long term monitorings, and
the results of a validation test in 4 subjects. The system is based
on the integration of the MagIC smart shirt (i.e., a textile-based
wearable system for the assessment of ECG and respiratory
movements), and an external triaxial MEMS accelerometer
positioned on the left clavicle. SCG was estimated as the
average of accelerations occurred in each heart beat. The SCG
components due to the valve closure and to recoil forces
following the heart contraction (ballistocardiogram) were
extracted by high-pass (>18 Hz) and band-pass (0.6-20 Hz)
filters respectively. Then the difference between the I and J
waves of the ballistocardiogram ( |I-J| index, possibly related to
the cardiac output) was identified by an ad-hoc procedure and
compared with the model flow indirect estimation of cardiac
output.
Validation on 4 volunteers showed that: 1) our wearable
system provides statistically consistent estimates of both heart-
sound related vibrations and recoil movements; 2) reliable
estimates of the |I-J| index can be obtained by considering
about 1 minute of SCG recording in stationary conditions; and
3) changes of the |I-J| index during exercise correlate well with
changes of cardiac output estimated by the model flow.
I. INTRODUCTION
contraction of the heart muscle ejecting blood into the
vessels. These forces are responsible for body accelerations
in the order of 5 mg. Since most studies on
ballistocardiogram (BCG) were aimed at measuring these
movements and their correlation with the cardiac output, in
this paper BCG will refer to the SCG low-frequency
component reflecting recoil movements, while SCG will
indicate the whole spectrum of cardiac vibrations.
Despite initial enthusiasms for its possible clinical
applications, however, the SCG technique has remained a
tool for physiologists, failing to be widely used in clinical
practice. Possibly this is due to the need of using relatively
complex recording devices, which restricted its applicability
to laboratory tests, where similar clinical information can be
provided by other systems (echocardiographers and
phonocardiographers). In recent years, however, MEMS
technology has provided small and cheap accelerometers
that may became part of a new class of wearable systems for
long term SCG monitoring. Although such a class of
systems will be inevitably more prone to noise and artefacts
than laboratory devices, they might allow the assessment of
SCG in a wider range of conditions, including sleep and
daily life activities.
In this paper we describe a new wearable system we have
recently developed for SCG assessment out of the laboratory

S EISMOCARDIOGRAPHY (SCG) is the study of body

vibrations induced by the heart beat. This term was
popularized by the works of Salerno and Zanetti at the
setting, through MEMS accelerometers. We also illustrate
the results obtained from the evaluation of the capability of
the system to monitor changes in cardiac output.
beginning of the 90s [1-3], but techniques for measuring
body movements associated to the cardiac activity are much
older. According to the measured quantity (displacement,
velocity or acceleration) and sensors position, this method
has been called apex-cardiography [4], kineto-cardiography
[5], cardio-kymocardiography [6] or ballistocardiography
[7], this latter being probably the more widely used term.
Vibrations induced by the cardiac activity are mainly
correlated to valve closure (heart sounds) and cardiac
systole. In particular, SCG movements at relatively lower
frequencies reflect the recoil forces that results from the

Manuscript received March 30, 2007.

P. Castiglioni is with Polo Tecnologico (Biomedical Technology
Department), IRCCS S.Maria Nascente, Fondazione Don Gnocchi
ONLUS, Milan, Italy (fax: +39-024048919; e-mail: pcastiglioni@
dongnocchi.it).
A. Faini is with DIMET, Universit degli Studi di Milano Bicocca, Italy
G. Parati is with Istituto Auxologico Italiano, Milan, Italy
M. Di Rienzo is with Polo Tecnologico (Biomedical Technology Figure. 1. Scheme of the wearable SCG system: A= 3D accelerometer;
Department), IRCCS S.Maria Nascente, Fondazione Don Gnocchi B=textile ECG electrodes; C=piezoelectric textile sensor for thoracic
ONLUS, Milan, Italy. movements; D= electronic board for signals storing and transmission;
BD= textile paths of conductive fibers; AD= external cable.

1-4244-0788-5/07/$20.00 2007 IEEE 3954

 II. SYSTEM DESCRIPTION
A. Sensors and Recording Device
Our SCG system is based on the integration of a textile-
based smart garment previously developed (the MagIC
system) and an external MEMS 3D accelerometer. In short,
the MagIC system is composed of a vest of cotton and lycra
with embedded textile ECG electrodes, an embedded textile
piezoresistive transducer for respiratory movements and a
portable electronic module. Electrical connections are made
of textile paths of conductive fibers. The electronic module
digitizes signals at 150 Hz, locally stores them on a memory
card and, possibly, transmit data via bluetooth connection to
a remote computer or PDA. Details on the system have been
provided elsewhere [8].
To the above smart garment we added an external triaxial
MEMS accelerometer (LIS3L02AL, STMicroelectronics,
size= 4.85 x 4.85 x 1.46mm; weight=0.08 grams; full scale=
2g, resolution= 0.5mg). The accelerometer was stuck on
the left clavicle. We selected this site because vibrations
related to heart sounds are transmitted with low damping,
and pulse movements of arterial walls do not affect
importantly recoil movements. The accelerometer was
connected to the MagIC electronic module through an
external cable (fig.1).
B. Assessment of SCG components
Each component (X, Y and Z) of the acceleration vector
was filtered twice to separate the BCG signal (band-pass
filter, from 0.6 to 20 Hz) from heart-sounds related
vibrations (high-pass filter, from 18 Hz). The R-peak of
ECG was identified through a threshold algorithm. Due to
Figure 2. Upper panels: average ECG wave over 600 heart beats in a
subject sitting at rest for 10; lower panels: average vertical
acceleration after band-pass (left) and high-band (right) filtering; thin
lines are the 95% confidence intervals of the mean. The band-pass
filtered SCG shows typical BCG waves, while the high-pass filtered
SCG shows vibrations synchronous with heart sounds.
the low signal-to-noise ratio, filtered accelerations were split
into individual beats, synchronised with the R-peak, and
averaged. Figure 2 shows the averaged vertical accelerations
after band-pass and high-pass filtering in a representative
subject. Typical BCG waves [9] are clearly visible in the
band-pass filtered SCG, while vibrations synchronous to
heart sounds appear in the high-pass filtered SCG .
C. Assessment of I-J amplitude from BCG
The BCG feature more correlated to the cardiac
hemodynamics is the difference between the I and J peaks.
This feature can be correctly quantified on the acceleration
component oriented as the mechanical axis of the heart. The
orientation of the cardiac axis may change with changes of
posture [9]. This makes it difficult to identify an
accelerometer axis (even after rotations of the reference
frame) for calculating the I-J amplitude in a wearable
system. For this reason, we computed the I-J amplitude
independently from the reference frame as follows. First, we
plot the band-pass filtered acceleration components, after the
R-peak synchronized average, in the X-Y-Z frame of the tri-
axial accelerometer, obtaining the 3D trajectory of the
acceleration vector. Then we estimated the modulus of the
difference between the acceleration vectors in I and J, |I-J|,
as the maximum distance between any couple of points of
the acceleration trajectory.
Only the part of the trajectory within 50 ms and 250 ms
from the R peak was considered: in fact this is the time
period where I and J peaks are expected to occur. The
distance calculated in this way does not dependent on
rotations of the reference frame (fig. 3).
Figure 3. Trajectory of the acceleration vectors obtained by plotting
the X, Y and Z components of BCG, for 450 ms after the R peak; the
thick line falls in the time window between 50 ms and 250 ms, i.e.,
where I and J waves are expected to occur; the modulus of the I-J
distance is estimated as the maximum distance between two points of
the curve
3955
 III. SYSTEM VALIDATION
First, the system has been validated by assessing the
statistical consistency of the estimates of low and high
frequency components of SCG, i.e., BCG and heart-sounds
vibrations, during a 10 minutes recording in a subject sitting
at rest.
Then we focused on the validation of the |I-J| index. This
was done by evaluating the minimum number of heart beats
required for the calculation of |I-J|; and by quantifying the
relation between changes of cardiac output and changes of
|I-J|. For this validation, we recorded the SCG in four
healthy volunteers during two conditions of 5 minutes
duration each: sitting at rest (REST), and while performing
physical exercise at 25W on cycloergometer (EXE). We
derived an indirect measure of cardiac output, to be
compared with the |I-J| index, by simultaneously measuring
the finger arterial pressure by the Finometer device
(Finapres Medical Systems, Amsterdam, the Netherlands) in Figure 5. |I-J| estimated in REST condition in a typical subject, as a
function of the employed number of heart beats N. The dashed line is
both REST and EXE conditions. It was possible to indirectly the value calculated from the whole 5 minutes recording (N=317).
calculate changes of cardiac output by analyzing the pulse
contour of the blood pressure wave with the model-flow derived from the whole REST recording with values
method [10]. obtained by progressively increasing the number of heart
beats N used for the estimate, starting from N=10 beats. A
A. SCG components: BCG and high frequency vibrations
typical example is shown in fig.5: after an initial transient,
We verified the statistical consistency of the average the estimate quickly converges to the final value. In practice,
synchronized with the R-peak by computing again the band- the |I-J| estimate reaches a sufficiently stable value when N
pass and high-pass filtered components of SCG previously is greater than 60 heart beats, i.e., after about 1 minute.
shown in fig.2, but this time by means of a randomly Similar results were obtained in all the four subjects.
synchronized average. This was done by synchronizing the We also evaluated whether the increase of cardiac output
average of the acceleration components with the series of R- which is expected to occur from REST to EXE, produces
peaks after having added a random delay to the ECG. The comparable changes of the |I-J| index. For this aim, firstly
added delay removed the beat-by-beat synchronization we expressed the increase of cardiac output from REST to
between ECG and SCG. Results are shown in fig. 4. The EXE in percentage; then we similarly evaluated the percent
absence of the SCG features observed previously (fig.2) increase of the |I-J| index. Finally we computed the linear
indicates that they are not due to random fluctuations or to correlation between the two percent increases among the
methodological artifacts. four subjects.
Individual values of percent increases of cardiac output
and |I-J| index are shown in Table 1. Changes of the two
quantities are similar in all the subjects. The regression line
between cardiac output percent changes (CO) and |I-J|
percent changes is given by:
CO = 0.61 I J + 16.5 (1)
with correlation coefficient r2=0.89.

TABLE I
CHANGES OF CARDIAC OUTPUT, (CO), FROM REST TO EXE AS
ESTIMATED BY THE MODEL FLOW METHOD, AND CORRESPONDING
Figure 4. Average vertical acceleration after band-pass (left) and high- CHANGES OF |I-J|, IN PERCENTAGE
band (right) filtering of the same subject of fig.2, but in this case the
average was de-synchronized from the R-peaks (see text); thin lines Subject CO Model Flow |I-J|
show the 95% confidence intervals of the mean.
1 28.8% 26.5%
2 46.5% 43.3%
B. |I-J| index 3 55.5% 68.0%
To evaluate the minimum number of heart beats needed 4 43.6% 40.7%
for a valid estimate of the |I-J| index, we compared the value

3956
 IV. CONCLUSION
This study showed the feasibility of monitoring the SCG
by a wearable system based on textile technology for ECG
recordings plus an external MEMS triaxial accelerometer.
The system we propose does not require the accelerometer
axes to be oriented precisely with respect to the cardiac
mechanical axis, simplifying the subjects instrumentation.
Moreover we showed that it can provide reliable estimates
of the SCG components, in particular the main waves of
BCG, just from one minute of recording during a steady-
state condition. Since we also showed that it is possible to
derive an index related to the changes of cardiac output from
the I and J waves, our system could potentially be used to
assess long term changes of cardiac hemodynamics, for
instance in sleep studies, in subjects during daily-life
activities, or in cardiac patients under therapy or following
rehabilitation.
REFERENCES
[1] D. M. Salerno and J. Zanetti, "Seismocardiography: A New
Technique for Recording Cardiac Vibrations. Concept, Method,
and Initial Observations," J Cardiovasc Technol, vol. 9, pp. 111-
118, 1990.
[2] D. M. Salerno and J. Zanetti, "Seismocardiography for monitoring
changes in left ventricular function during ischemia," Chest, vol.
100, no. 4, pp. 991-993, Oct.1991.
[3] J. Zanetti, M. O. Poliac, and R. S. Crow, "Seismocardiography:
waveform identification and noise analysis," 1991, pp. 49-52.
3957
[4] A. Silvestre, G. Sandhu, K. B. Desser, and A. Benchimol, "Slow
filling period/rapid filling period ratio in the apexcardiogram:
relation to the diagnosis of coronary artery disease," Am. J Cardiol.,
vol. 42, no. 3, pp. 377-382, Sept.1978.
[5] A. Grandi, F. Barzizza, L. Bernardi, A. Venco, and G. Finardi,
"Kinetocardiographic detection of ventricular dyssynergy after
myocardial infarction. Correlations with two-dimensional
echocardiography," Acta Cardiol., vol. 39, no. 1, pp. 19-27, 1984.
[6] R. A. Silverberg, G. A. Diamond, R. Vas, D. Tzivoni, H. J. Swan,
and J. S. Forrester, "Noninvasive diagnosis of coronary artery
disease: the cardiokymographic stress test," Circulation, vol. 61,
no. 3, pp. 579-589, Mar.1980.
[7] I. Starr, J. Rawson, H. A. Schroeder, and N. R. Joseph, "Studies on
the estimation of cardiac ouptut in man, and of abnormalities in
cardiac function, from the heart's recoil and the blood's impacts; the
ballistocardiogram," Am. J. Physiol, vol. 127, pp. 1-28, July1939.
[8] M. Di Rienzo, F. Rizzo, P. Meriggi, B. Bordoni, G. Brambilla, M.
Ferratini, and P. Castiglioni, " Applications of a Textile-Based
Wearable System for Vital Signs Monitoring,", 28th Annual
International Conference of the IEEE ed Engineering in Medicine
and Biology Society, 2006, pp. 2223-2226.
[9] R. W. Soames and J. Atha, "Three-dimensional
ballistocardiographic responses to changes of posture," Clin. Phys.
Physiol Meas., vol. 3, no. 3, pp. 169-177, Aug.1982.
[10] K. H. Wesseling, J. R. Jansen, J. J. Settels, and J. J. Schreuder,
"Computation of aortic flow from pressure in humans using a
nonlinear, three-element model," J Appl. Physiol, vol. 74, no. 5, pp.
2566-2573, May1993.