-
Abstract— We present an algorithm for obtaining the heart
rate from the signal of a single, contact-less sensor recording the
mechanical activity of the heart. This vital parameter is
required on a beat-to-beat basis for applications in sleep
analysis and heart failure disease management. Our approach
bundles information from various sources for first robust
estimates. These estimates are further refined in a second step.
An unambiguous comparison with the ECG RR-intervals taken
as reference is possible for 98.5% of the heart beats. In these
cases, a mean absolute error of 17 ms for the inter-beat interval
lengths has been achieved, over a test corpus of 20 whole nights.
I. INTRODUCTION
ome monitoring solutions where cables hinder the
patient during daily routines lack acceptance.
Unobtrusive monitoring can be realized with sensors which
are not in direct contact with the patient and are hidden in the
patient’s home. However, this comes at the cost of a more
sophisticated signal processing. In this paper we present a
novel algorithm for determining the heart rate of a person
during sleep. A contact-less mechanical sensor is integrated
into an ordinary bed. It records the ballistic forces due to
cardiac activity, the so-called ballistocardiogram (BCG). But
changes in the relative position of sensor and patient occur
and lead to a high signal variability. Limb movements,
coughing or snoring cause mechanical activity as well and
superimpose to the BCG in the recording. Therefore, our
former solution could only provide an average heart rate for
an epoch covering a few tens of seconds [1]. This is also the
case for most approaches presented so far in the literature
[2], [3]. However, applications in the field of sleep staging,
for the detection of sleep apnea or high risk patient
identification after a heart attack, require information about
the heart rate on a beat-to-beat basis. Since cardiovascular
diseases are the leading cause of death world wide and
quality of life is closely linked to quality of sleep, there is a
clear need to develop algorithms which estimate the length
of each cardiac cycle (beat-to-beat resolution) [4],[5].
Current systems with this resolution either require interaction
of a human expert [6], machine learning steps which are
Manuscript received on day month year (version of 21st March 2010).
D. Friedrich, Philips Research Europe, present address: Lehrstuhl für
Bildverarbeitung, Sommerfeldstr. 24, 52074 Aachen, Germany (email:
David.Friedrich@ RWTH-Aachen.de, phone +49 241 8027803)
X. L. Aubert and A. Brauers are with Philips Research Europe in
Aachen, Germany ({Xavier.Aubert, Andreas.Brauers}@philips.com).
H. Führ is with the Lehrstuhl A für Mathematik, RWTH Aachen
University of Technology, Germany.
sensitive to artifacts during the learning phase [4] or use
expensive multi-sensor equipment [7]. In this work we
describe an algorithm for computing the heart rate from a
BCG recorded with a single sensor. Human interaction is not
required. The algorithm has been especially tailored for the
ballistocardiogram and its underlying signal structure. By
exploiting knowledge about the physiology of the human
heart and bundling information from distinct aspects in a
hybrid approach, first robust estimates for heart beat interval
length are obtained. These estimates are then subjected to a
second refinement step, yielding the final result.
In this work, the performance of the algorithm has been
assessed on a test corpus comprising 20 whole night
recordings from ten different subjects. The RR intervals
obtained from an Electrocardiogram (ECG) served as a
reference to evaluate the accuracy of BCG derived intervals.
This paper is organized as follows. The second section
describes how we have collected the data used for the study.
The algorithm for heart rate estimation on a beat-to-beat
basis is presented in section III. Section IV describes the
evaluation of our approach and presents the main results.
These results are then discussed in section V. The last
section concludes our work and outlines perspectives for
future research.
II. BCG A
CQUISITION
Electromechanical film (Emfit) sensors are very flat sensors
which generate an electrical charge if they get mechanically
deformed. For BCG acquisition, a single Emfit sensor of
30cm × 60cm is placed in the thorax region under a thin
foam mattress on top of an ordinary bed (See Fig. 1). Due to
the flatness and flexibility of the sensor element, it is almost
unnoticed by a person lying on the bed. Yet, it provides
enough sensitivity to record the BCG of this person.
Several acquisition devices have been used for recording the
Emfit sensor signal, providing a sampling rate of at least 128
Hz. They submit the recorded signal to a standard PC where
DSP is performed offline: major body movement artifacts
which superimpose to the BCG are identified using low level
features related to the signal amplitude and energy. The
whole night recording is then split up into signal parts which
are free of significant artifacts. Remaining contributions in
the recorded signal mainly arise from breathing movements
of the thorax and ballistic forces of the beating heart. As they
both cover distinct frequency ranges, the ballistocardiogram
can be separated from the respiration component with linear
filter methods [1].
Heart Rate Estimation on a Beat-to-Beat Basis via
Ballistocardiography - A hybrid Approach.
David Friedrich, Xavier L. Aubert, Hartmut Führ and Andreas Brauers
H
32nd Annual International Conference of the IEEE EMBS
Buenos Aires, Argentina, August 31 - September 4, 2010
978-1-4244-4124-2/10/$25.00 ©2010 IEEE 4048
- 1
- 2
前往页