INSTITUTE OF PHYSICS PUBLISHING
MEASUREMENT SCIENCE AND TECHNOLOGY
Meas. Sci. Technol. 15 (2004) 1614–1620
PII: S0957-0233(04)74430-6
A ‘smart’ bed for non-intrusive
monitoring of patient physiological factors
W B Spillman Jr1,2, M Mayer1, J Bennett1, J Gong1,
K E Meissner1, B Davis1, R O Claus1, A A Muelenaer Jr1
and X Xu1
1
Virginia Tech Applied Biosciences Center, Virginia Tech (0356), Blacksburg,
VA 24061, USA
2
Physics Department, Virginia Tech (0356), Blacksburg, VA 24061, USA
E-mail: wspillma@vt.edu
Received 24 December 2003, in final form 24 February 2004
Published 19 July 2004
Online at stacks.iop.org/MST/15/1614
doi:10.1088/0957-0233/15/8/032
Abstract
In this paper we present the results of research aimed at the development of
a ‘smart’ bed to non-intrusively monitor patient respiration, heart rate and
movement using spatially distributed integrating multimode fibre optic
sensors. The research is focused upon allowing more automation of patient
care, an especially important matter for the elder population, which is a
rapidly growing fraction of much of the world population today. Two
spatially integrating fibre optic sensors were investigated, one of which was
based on inter-modal interference and the other on mode conversion. The
sensing fibre was integrated into a bed and test subjects were monitored in
different positions. The sensor outputs were then correlated with subject
movement, respiration rate and heart rate. The results indicated that the
inter-modal sensor could detect patient movement and respiration rate while
the mode conversion sensor could detect patient movement, respiration rate
and heart rate. Results and analysis of the research are presented and future
research activities discussed.
Keywords: motion, deformation, perturbation, spatially integrating fibre
optic, mode modulation sensing, non-intrusive physiological factor
monitoring
1. Introduction
The continuing shortage of medical staff and the increase
in the elder population due to the baby boom after World
War II make the automation of health care an ever-increasing
priority. In particular, patient monitoring is very intrusive
and labour intensive. In this paper we discuss research
aimed at the development of a ‘smart’ bed to non-intrusively
monitor patient respiration, heart rate and movement using
spatially distributed integrating fibre optic sensors. These
three parameters are extremely important in determining
patient condition and preventing future problems for patients in
nursing homes and extended care facilities. The measurement
of the respiration rate and heart rate provides an immediate
indication of whether a patient is in any distress, while the
0957-0233/04/081614+07$30.00
measurement of patient movement can be used to determine
whether that movement has been so limited over a period of
time that the patient must be turned to a new position by a health
care professional to prevent the occurrence or exacerbation of
pressure or bed sores. In clinical settings such as hospitals,
outpatient surgery centres or nursing homes, vital signs such as
pulse and respiratory rates are measured by direct observation
by skilled medical personnel. Continuous monitoring of vital
signs requires attachment of sensors to the body in a number
of ways [1]. Monitoring of essential vital signs is an integral
part of medical care. The pulse rate can be determined by
placement of electrodes on the skin and monitoring of the
electrocardiogram. The output of a fibre optic sensor placed
on a finger, toe or ear lobe and attached to a pulse oximeter can
be used to determine the pulse rate. The respiratory rate can be
© 2004 IOP Publishing Ltd Printed in the UK
1614
‘Smart’ bed for non-intrusive patient monitoring
Figure 1. STM sensor schematic diagram.
Figure 2. HOME sensor schematic diagram.
determined by chest movement as detected by changes in the
chest wall electrical impedance or inductance. Each method
for detecting pulse or respiration requires an interface between
the sensor and the patient’s skin and the sensor must be held in
place with an adhesive or by mechanical means such as Velcro.
Any of these sensors can cause skin irritation or breakdown
and may contribute to patient discomfort. Pressure sores are
a major cause of morbidity and mortality in the healthcare
setting. As many as 1.5 million individuals are affected by
pressure sores, at a total cost of 5 billion dollars annually
[2]. The prevalence of pressure sores in one US teaching
hospital was 8% [3]. Repositioning schedules are utilized
as part of most preventive measures in healthcare facilities.
Recommendations are for repositioning bed-ridden patients
every 2 h and individuals in chairs at least once per hour
[4]. Two different types of sensor were investigated both of
which were based on modulation of the modal distribution in
multimode optical fibres. Experimental results are presented
and the relative merits and drawbacks of the two sensor types
are discussed. Finally, future planned research activity is
described.
2. Theory
In order to develop a non-intrusive method of detecting
respiration, the use of point sensors was ruled out due to the
possibility of continual shifting of the patient position. Instead,
a spatially distributed integrating approach was chosen so that
if a patient were present anywhere within a specific localized
area, sensing could be carried out [5]. The basic concept
is that any patient movement that also moved an optical
fibre within the specified area would produce a change in
optical signal that would indicate patient movement. The
physical repetitive movement caused by respiration or heart
pumping would be contained within the signal as well and
could be extracted via appropriate signal processing. To test
this concept, two different modal modulation approaches were
used with multimode optical fibre excited by a coherent laser
source. In the first technique (statistical mode sensing (STM)),
all the guided modes of the fibre are excited and then detected
by a low cost digital camera. This is shown schematically
in figure 1. The sum of the absolute values of the change
in light intensity on each of the pixels between each time
frame is then calculated. This technique [6] then provides
a measure of the absolute value of the first time derivative
of a perturbation integrated along the fibre length. In the
second approach (high order mode excitation (HOME)), only
the higher order modes of the fibre are excited so that the output
from the unperturbed fibre results in a bright annulus when
projected on a screen. A large area circular photodetector
is positioned so that its diameter fits within the annulus but
does not intercept it. A schematic diagram of this technique is
shown in figure 2. When the fibre is perturbed, the perturbation
couples light from the higher order modes to lower order
modes where it is intercepted by the large area detector,
converted into an electrical current and measured. This
technique [7] provides a signal that is directly proportional
to the perturbation integrated along the fibre length.
We analysed the applicability of these two techniques for
simultaneously detecting patient movement, respiration and
heart rate. The perturbation due to respiration and heart rate
was modelled as the sum of two cosine functions with the
second cosine (representing the perturbation due to the heart)
having an amplitude of 0.1 relative to the amplitude of the first
cosine (representing the perturbation due to respiration). The
frequency of the second cosine was 60 cycles min−1 while
the frequency of the first cosine was 9 cycles min−1. The
frequencies were chosen to have roughly the same values
as average respiration and heart rates, while the difference
in amplitudes was chosen simply to represent the fact that
physical body movement due to respiration is much greater
than that due to heart action. This model is clearly a very crude
approximation, since the integrated perturbations due to the
two sources (heart and lungs) would be periodic signals whose
shapes would not be uniform in the same way the mathematical
cosine signals would be. Nonetheless, implementing the
model is instructive as a way to contrast the two sensing
techniques and the information they might be able to provide.
The discrete Fourier transform of the modelled signal (sum
of the two cosines, i.e. representing the HOME sensor), and
the discrete Fourier transform of the absolute value of the
first derivative of the modelled signal (representing the STM
sensor) are shown in figure 3. One would expect that the power
spectrum of the HOME sensor would show two clear peaks
at the frequencies of the two cosine functions, since its output
should be directly proportional to the modelled integrated
perturbation. The power spectrum of the STM sensor,
however, should be more complex. The fact that the processing
takes the absolute value of the first time derivative of the
integrated perturbation should produce signals with maximum
power components at twice the fundamental frequencies and
a distorted power spectrum (e.g. if one takes the absolute
value of a cosine function, the frequency doubles and a
discontinuity in the slope is introduced). This implies that
the large signals seen by the STM sensor at low frequencies
will produce power spectra that mask the power spectra of
signals at higher frequencies.
1615
W B Spillman Jr et al
104
102
100
sensor output power -2
(arbitrary units) 10
10-4
HOME sensor
STM sensor
10-6
10-8
0
20
40
60
cycles/minute
80
100
120
Figure 3. Modelled power spectra of two perturbation detection by the STM and HOME sensors.
Figure 4. ‘Smart’ bed experimental set-up.
As can be seen from figure 3, the modelled STM power
spectra (dashed line) clearly show the first cosine function (at
twice its frequency due to the taking of absolute value) but
does not clearly indicate the second cosine function due to
the complications introduced by the sensor signal processing.
The HOME signal, on the other hand, clearly shows the peaks
due to both perturbations at their correct frequencies. This
suggests that the STM approach should allow detection of
respiration and perhaps heart rate, but the HOME approach
should introduce less signal distortion due to processing
and allow better signal discrimination. It should also be
noted that the HOME sensor output would saturate when
sufficiently large levels of perturbation are present and the
modal volume is uniformly populated, while the STM sensor
should be relatively insensitive to saturation since it is based
on interference.
3. Experiment
In order to develop a non-intrusive method of detecting
respiration, heart rate and patient movement, the top surface
of a mattress was covered with a 200 µm core step index silica
multimode optical fibre arranged in two sinusoidal overlapping
patterns arranged orthogonal to each other so that the fibre in
each pattern crossed the fibre in the other pattern at an angle
of 90◦ . Light from a laser pointer with output at 670 nm was
1616
used to excite the fibre and the output light was detected either
by a digital camera or a large area photodetector depending
upon the sensing technique used. The experimental setup in
the lab is shown in figure 4 with a test subject in position on
the fibre instrumented mattress.
4. Results
A number of experimental runs were conducted using both
the STM and HOME sensors. Since the natural time intervals
for measurements of physiological parameters are typically
fractions of minutes (respiration rates are of the order of
10 min−1 and heart rates are of the order of 70 min−1), most
data are plotted against a time scale of minutes or cycles min−1.
For perturbations due to a female test subject (height 1.6 m,
mass 50 kg), a typical time trace from the STM sensor taken
while the subject was lying on her stomach on the bed is
shown in figure 5 while its Fourier transform is shown in
figure 6. Figure 7 shows how the results vary for the same
test subject in different typical sleep positions: on back, on
stomach, left fetal and right fetal. The signal peaks are at twice
the respiration rate as expected due to the absolute value taken
during signal processing so that plotting the power spectra
versus 0.5 times the measured frequency provides the actual
perturbation frequency values.
‘Smart’ bed for non-intrusive patient monitoring
0.50
0.0
-0.50
STM sensor output
(arbitrary units)
-1.0
-1.5
-2.0
0
10
20
30
40
time (s)
50
60
70
80
Figure 5. Typical time trace from the STM sensor.
50
40
respiration rate
30
SMS sensor power
(arbitrary units)
20
10
0
0
20
40
60
80
100
120
0.5 x cycles/minute
Figure 6. Power spectrum of the time trace shown in figure 5.
50
on back
on stomach
left fetal
right fetal
40
30
STM power output
(arbitrary units)
20
10
0
10
20
30
cycles/minute
40
50
Figure 7. STM sensor power spectra for different test subject positions.
For perturbations due to a male test subject (height 1.75 m,
mass 80 kg) lying on his stomach on the bed, a typical
time trace using the HOME sensor is shown in figure 8,
and its power spectrum is shown in figure 9. Figure 10
shows the power spectrum from the HOME sensor when
the test subject held his breath for 0.5 of a measurement
period. This allowed the heart rate signal to be clearly
discerned although the respiration rate signal was distorted.
Finally, figure 11 displays the measured (via the peak
in the power spectrum) versus the actual (as determined
by patient counting) respiration rates using the HOME
sensor.
The results indicate that both the respiration rate and
heart rate are represented in the signal although unusual
measures needed to be taken (suppression of the respiration
signal for half of a measurement period) to clearly show the
heart rate signal in the power spectrum (figure 10). The
signal is not clearly evident in figure 9 when the perturbation
due to respiration is not suppressed. It is believed that the
introduction of a discontinuity in the respiration perturbation
1617
W B Spillman Jr et al
0.43
0.42
0.41
HOME sensor output
(arbitrary units)
0.40
0.39
0.38
0.37
0
10
20
30
40
50
time (s)
Figure 8. Typical time trace from the HOME sensor.
0.060
0.050
0.040
HOME sensor power
(arbitrary units) 0.030
0.020
0.010
0.0
0
20
40
60
80
100
120
cycles/minute
Figure 9. Power spectrum of the time trace shown in figure 8.
0.0020
respiration rate
0.0015
HOME sensor power
(arbitrary units)
0.0010
heart rate
0.0005
0.0000
0
20
40
60
cycles/minute
80
100
120
Figure 10. HOME sensor power spectrum showing breathing and heart rates.
40
35
30
25
cycles/minute
(inferred)
20
15
slope = 1, or
inferred = actual
10
5
0
0
5
10
15
20
25
30
cycles/minute (actual)
Figure 11. Measured versus actual breathing rates using the HOME sensor.
1618
35
‘Smart’ bed for non-intrusive patient monitoring
Figure 12. Off-the-shelf components used in the prototype STM sensor.
also resulted in an accentuation of its first harmonic in the
power spectrum (the large signal at 24 cycles min−1 which
is clearly twice the frequency of the fundamental respiration
rate, 12 cycles min−1) as shown in figure 10.
5. Discussion
As can be seen from these results, both the STM and HOME
sensors can be used to detect patient movement and respiration.
Only the HOME sensor, however, demonstrated the ability
to clearly detect heart rate. These two sensors might have
different applications. The STM sensor, by the nature of
its transduction process, will not become saturated, i.e. the
speckle pattern will always be present and will always change
in response to additional perturbation. The size of the
dc component does not affect the sensitivity of this signal
processing method. It can, therefore, detect patient movement
and give repeatable results that are somewhat independent of
patient weight. The HOME sensor, on the other hand, could be
saturated by perturbations large enough to cause the available
propagating mode volume to become completely filled. For
applications involving critical care, the HOME sensor could
find extensive application if a method can be found to make it
more sensitive to heart rate. Both sensors offer the potential
to be low cost, PC compatible and have the capability to be
integrated into larger wireless systems.
We believe that the smart bed technique is not a
replacement for standard physiological monitoring. The
outputs from the sensors integrated into the bed allow
continuous monitoring of indications of patient movement,
respiration rate and heart rate, but these signals are all
combined and have to be separated via signal processing. In
contrast, respiration rate signals and heart rate signals can be
monitored directly through various other sensing techniques
through the attachment of probes to the body and electronic
data recording of the individual signals themselves. This type
of monitoring, however, is intrusive and uncomfortable for
the patient, cannot be used for extended periods of time and
requires expensive equipment. The smart bed, on the other
hand, could be a cost effective way of automating long term
monitoring of patients that would enhance the productivity
of health care professionals and optimize their one-on-one
interaction time with those patients for whom the need for
such personal interaction has become critical.
Finally, it should be noted that a very large number of
advanced signal processing techniques have been developed
since the advent of the digital computer, the Kalman filter being
one example. In our case, no advanced signal processing
has yet been applied to the outputs of the sensors we have
been investigating for the smart bed application. Our research
has clearly shown, however, that components due to the
physiological parameters of interest are present in the sensor
output signals. We are confident that when we begin to apply
advanced signal processing techniques, we will be able to
extract and separate signals of interest in a robust manner and
the system performance will improve considerably as a result,
without any modification in hardware.
6. Future work and clinical trials
In order to develop prototypes suitable for clinical testing,
two approaches are being pursued. In the first, a cost effective
wireless version of the STM sensor has already been designed,
fabricated and tested. In this case, the optical detector is an
off-the-shelf digital camera with the capability of wireless
transmission to a remote PC. In addition, to enhance the
performance of both the STM and HOME sensors, a study
is being performed to analyse the distribution of motions
and forces produced by the human body on a bed due to
respiration and heart action. The results of this study will be
used to identify optimal spatial configurations of the fibre on
the bed and other parameters such as required stiffness of fibre
support, etc. The new configurations will then be validated
experimentally.
Work has already begun to make the STM sensor practical.
The prototype stand alone system that has been developed uses
primarily off-the-shelf components. As shown in figure 12,
the system consists of a laptop PC, a wireless transmitter, a
remote wireless unit containing a laser diode and a wireless
digital camera that served as the detector, and a sensing fibre.
The camera transmits at its maximum rate from the sensing
location to the laptop PC where the individual pixels from
1619
W B Spillman Jr et al
0.25
4
2
0.20
0.15
STM sensor output
(arbitrary units)
5
1
0.10
0.05
3
0.00
0.0
0.5
1.0
6
1.5
2.0
2.5
time (minutes)
3.0
3.5
4.0
Figure 13. STM sensor output corresponding to a patient getting out of bed and then returning.
0.25
0.20
0.15
STM sensor output
(arbitrary units)
coughing
0.10
0.05
0.00
0.0
0.5
1.0
1.5
minutes (coughing)
2.0
2.5
Figure 14. STM sensor output corresponding to a patient coughing.
sequential frames are processed to provide the appropriate
output. This particular configuration allows the sensor and
processing to be separated with the potential for a single PC to
be able to multiplex and process the outputs from a number of
spatially separated sensors simultaneously which should result
in a significant reduction in the cost/sensing location due to
the fact that the laptop PC is the most expensive component in
the whole system.
Plans are underway to test the STM sensor in a clinical
trial at the Carilion Health System Sleep Center in Roanoke,
Virginia in the near future. Following that, a clinical trial
of an advanced multiplexed STM system will be conducted
at a Medical Facilities of America nursing home, also in the
Roanoke, Virginia region. A one night validation of the STM
sensor has already been carried out at the Carilion Health
System Sleep Center with an individual patient prior to the
actual clinical trial. In figure 13, the output of the STM sensor
is shown for a period of time when the patient moved to the
edge of the bed (1), got up out of the bed (2) leaving the bed
empty (3), then sitting back down on the bed (4), settling to
a comfortable position (5) and then resting quietly (6). In
figure 14, the STM sensor response to a person coughing is
clearly shown. These results clearly indicate the potential
for non-intrusive patient measurement in the health care
environment.
1620
Acknowledgments
The authors gratefully acknowledge the financial support
provided for this project by the Carilion Biomedical Institute,
and assistance provided by the Carilion Health System Sleep
Center and ADMMicro, Incorporated.
References
[1] Ganong W F 2001 Review of Medical Physiology 20th edn
(Norwalk, CT: Appleton & Lange)
[2] Kaufman M W 2001 The WOC nurse: economic, quality of
life, and legal benefits Dermatol. Nurs. 13 215–21
[3] Granick M S et al 1996 Wound management and wound care
Adv. Plastic. Reconstr. Surg. 12 99–121
[4] Clinical Practice Guideline No 3. AHCPR Publication
No 92-0047 1992 (Rockville, MD: US Department of Health
and Human Services, Public Health Service, Agency for
Health Care Policy and Research)
[5] Spillman W B Jr and Huston D R 1995 Scaling and antenna
gain in integrating fiber optic sensors J. Lightw. Technol. 13
1222–30
[6] Spillman W B Jr et al 1989 Statistical mode sensor for fiber
optic vibration sensing applications Appl. Opt. 28 3166–76
[7] Herczfeld P R et al 1990 An embedded fiber optic sensor
utilizing the modal power distribution technique J. Opt. Lett.
15 1242–4