Design and Development of a Contactless Planar Capacitive Sensor by Thuvatahan Sivayogan A thesis submitted in conformity with the requirements for the degree of Master of Health Science in Clinical Engineering Institute of Biomaterials and Biomedical Engineering University of Toronto c Copyright 2013 by Thuvatahan Sivayogan
57
Embed
Design and Development of a Contactless Planar Capacitive ... · Design and Development of a Contactless Planar Capacitive Sensor Thuvatahan Sivayogan ... who lives in the hearts
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Design and Development of a Contactless Planar CapacitiveSensor
by
Thuvatahan Sivayogan
A thesis submitted in conformity with the requirementsfor the degree of Master of Health Science in Clinical Engineering
Institute of Biomaterials and Biomedical EngineeringUniversity of Toronto
• Sensitive to motion artefacts• Dependent on bed characteristics• Dependent on patient orientation, positioning, and size
Capacitive • Simple• Easy fabric integration
• Sensitive to motion artefacts• Signal corruption due to metal• Electromagnetic interference from environment• Potential signal drift• Dependent on bed, clothing, and patient characteristics
Video • Fully non-contact method • Privacy concerns• Limited by visibility of patient• Heavy signal processing and computation• Dependent on lighting conditions and image contrast
Doppler • Fully non-contact method • Not inexpensive• Limited by visibility of patient• Dependent on reflection characteristics of patient surface• Signal degrades with patient orientation, posture, and
position
Table 2.1: Summary of novel respiration sensors.
Table 2.1 presents a summary of the advantages and disadvantages of several novel sensor
technologies found in literature, particuarly acoustic [24], piezoelectric [25], capacitive [26],
video [27], and Doppler sensors [28]. Acoustic sensors work by measuring tracheal sounds
using a microphone placed around the neck or attached to a mouthpiece. Piezoelectric
sensors are the most common implementation which involve sensors placed underneath
the bed. Capacitive sensors embedded into clothing measure respiration through stretch.
Video sensors use cameras placed overhead and sometimes require a marker placed on
Chapter 2. Literature Review 9
the patient. Finally, Doppler sensors use an antenna placed overhead to measure the
doppler-shift between the transmitted and received signal reflected from the body surface.
Overall, these sensors suffer from a variety of shortcomings including low sensitivity, poor
reliability, and impracticality, which have limited their use in the clinic.
2.4.1 Original Respiration Pad
The first functional prototype of the Respiration Pad (PAD) was developed by Hart et al.
[15]. The original PAD, shown in figure 2.4, has 3 major components: the pad sensor,
development board, and iPhone application. The pad sensor consists of dry electrodes
silk-screened on a yoga mat placed underneath the patient. The development board
contains a capacitance chip (Analog Devices AD7746), which sends an excitation signal
to the electrodes to make capacitance measurements. The iPhone application implements
a user interface that captures, filters, and displays the respiration waveform.
Figure 2.4: The original Respiration Pad consisting of the pad sensor, development board,and iPhone application. Figure adapted from Hart et al. [15].
Hart et al. hypothesized that the electric field generated by the electrodes was able to
penetrate into the body to measure the dielectric properties of the human torso, which
change with respiration. However, this hypothesis has been shown to be false because the
generated electric field does not penetrate into the human body (see chapter 3).
The study investigated the accuracy of the PAD measurement against respiratory induc-
tance plethysmography (RIP) for respiratory rates (RR) and tidal volumes (TV) of 5 to
Chapter 2. Literature Review 10
55 breaths/min and 150 to 400 cubic-centimeters (cc), respectively. Results showed high
correlations of 0.882 (95% CI, 0.861 to 0.900) and 0.953 (95% CI, 0.945 to 0.960) for the
SimMan and healthy subject, respectively. However, correlation does not imply agreement
[29] and, therefore, the inappropriate use of correlation may lead to very misleading
conclusions. Furthermore, the study showed that the PAD had reduced sensitivity to
shallow breathing for tidal volumes below 250 cc.
The Hart work showed that it was possible to measure respiratory rate using an active
capacitive sensor. However, two issues remain unclear: (1) the actual sensing mechanism
and (2) the true performance of the sensor using correct statistical analysis. These issues,
as well as the clinical potential of the Respiration Pad, prompted further investigation
and development of the technology.
2.4.2 Capacitive differential sensor
A novel capacitive sensor was developed by Wartzek et al. [30] to measure respiration of
subjects lying supine on a mattress. The sensor, shown in figure 2.5, consists of an array
of 3 pairs of planar electrodes oriented orthogonal to the sagittal plane and positioned
underneath the mattress at either edge of the torso. The sensor works by using a 20 MHz
electric field Us with gain ±G to drive the differential capacitive electrode. The received
signal is amplified by a gain Gr, filtered, and measured as a phase change ∆φ. Phase
changes are caused by Maxwell-Wagner relaxation that occurs due to conductivity and
permittivity changes within biological tissue as a result of respiration [31].
Figure 2.5: Cross-sectional view of the torso and the capacitive differential sensor.
A small self-study was done by the authors to validate the technology against a mass flow
meter (TSI 4040), which was used as the gold standard. In the experimental setup, the
Chapter 2. Literature Review 11
sensor array was placed under the mattress with the subject lying supine. Results indicate
that the size of the phase changes are extremely small with an order of magnitude of 4
milli-degrees. Therefore, the authors admit that the signal quality is very poor resulting
in the wrong estimation of respiratory rate in some cases.
The authors claim that overall, the accuracy of the sensor was within ±0.8 breaths/min
of the the gold standard. However, the experiment conducted by the authors does not
evaluate the sensor performance at multiple tidal volumes or respiratory frequencies.
The authors also fail to use a diverse population of subjects and instead take repeated
measurements on the same subjects at a fixed tidal volume and respiratory rate. Therefore,
the results presented by the authors are very limited and may be misleading.
The capacitive differential sensor has several limitations that make it inadequate for use
as a non-invasive, continuous respiratory monitor. The sensor is sensitive to body size
(i.e. height and weight), patient positioning, and motion artefacts. Also, since the sensor
injects a high-frequency electrical field into the body, there is potential for interference
with other medical devices in the patient vicinity or implanted within the patient’s body
(e.g. pacemakers). These issues are a cause for concern and need to be examined further.
2.4.3 Angelcare baby monitor
The Angelcare AC401 baby monitor is a consumer device designed to provide parents with
wireless sound and movement monitoring of their infant in another room. The monitor is
composed of 3 components: the SensorPad, the nursery unit, and the parent unit. The
purpose of the device is to monitor sound and motion of the infant in the crib and to alert
the parent of the infant’s status. The device does not measure nor display respiratory
rate information on the unit although, as shown from the experiment below, the device
does have the capability to do so.
The SensorPad is actually a spring-loaded force-plate with a piezoelectric pressure trans-
ducer (piezo) placed inside the device. The piezo is made of a thin piezoelectric ceramic
round glued to a thin metal disc. When pressure is applied to the SensorPad, it compresses
the SensorPad resulting in contact between the piezo and a small rubber cone directly
underneath it. The piezo uses the piezoelectric effect to measure pressure by converting
applied pressure to an electrical charge, which is then measured by a circuit.
Chapter 2. Literature Review 12
a: SensorPad on top of mattress. b: SensorPad under mattress.
Figure 2.6: Qualitative testing of SensorPad using an adult subject.
To demonstrate the operation of the piezo transducer, two wires were soldered to the
transducer and connected to an oscilloscope. A qualitative experiment was performed
using 2 healthy subjects of different sizes breathing normally at a fixed respiratory rate
on a foam mattress. The signal for the thin subject was noisy but usable (see figure 2.6);
but the signal for the large-bodied subject was not visually distinguishable from noise.
Although the sensor is sensitive to patient orientation, position, posture, and body size,
these qualitative results indicate that the Angelcare baby monitor or any piezoelectric
pressure transducer has the potential for respiratory monitoring.
2.5 Summary
Respiratory rate is a clinically valuable vital sign; yet, clinical measurement of respiratory
rate remains intermittent, biased, and inaccurate. Research findings suggests that a
point-of-care device that allows for continuous, unbiased, and accurate measurement
of respiratory rate may improve respiratory rate documentation in hospitals. Several
respiration sensors found in literature were discussed; among these sensors, the original
Respiration Pad is novel and shows potential for clinical measurement of respiratory rate.
Therefore, this work aims to make a contribution through the design and development of
an improved Respiration Pad and its evaluation using a feasibility study.
Chapter 3
Computational Model
3.1 Introduction
The Hart work [15] hypothesized that the Respiration Pad works by capacitive volume
sensing as shown in figure 3.1(a). However, doubts about his reasoning and conclusion
arose from literature review [32] that suggested that planar capacitive sensors primarily
work by capacitive displacement sensing as shown in figure 3.1(b). But, intuitively, there
will not be any significant vertical displacement of the torso away from the mattress surface
during respiration since the mattress will contour itself to the shape of the torso. This
suggests that, perhaps, the Respiration Pad is actually measuring capacitance changes due
to changes in electrode geometry on the mattress surface caused by respiration as shown
in figure 3.1(c). Therefore, the new hypothesis is that the Respiration Pad measures
respiration through capacitive deformation sensing.
Although mathematical models describing planar capacitive sensors exist, they do not
provide insight into the interaction of the electric field with the human body nor do
they explain the actual sensing mechanism that is occurring in reality. Therefore, a
computational model is required. The goal of this theoretical study is to validate the
capacitive deformation sensing hypothesis using a computational model.
Figure 3.1: Computational models for capacitive (a) volume, (b) displacement, and (c)deformation sensing where the yellow square represents the human torso; the blue striprepresents clothing and linen; and the red and black strips represent the transmitting andreceiving electrodes, respectively.
3.2 Methods
The computational model was developed using a finite-element methods software, which
enables modelling, and simulating coupled physics phenomena to solve various physics
and engineering problems. By using this software to develop a computational model, it is
possible to theoretically demonstrate how capacitance measurement takes place under three
different underlying assumptions of capacitive volume, displacement, and deformation
sensing. This would otherwise be very difficult to demonstrate using experiment since
ensuring consistency throughout all tests will be challenging due to numerous confounding
factors arising from the environment and through random effects and disturbances that are
always present with capacitance measurement [33]. Therefore, by using a computational
model, the hypothesis can painlessly be verified in a repeatable and controlled way.
Development of the computational model involved several steps in the modelling process
including defining the geometry, specifying the physics phenomenon, meshing, solving
and post-processing of the results. To ensure that the computational model was built
correctly, the model setup was verified by comparing the capacitance values calculated
using known analytical formulas against those predicted by the model for both parallel
plate and planar capacitive sensors.
The parallel plate capacitor, shown in figure 3.2, has a simple analytical formula:
C = εrεoA
d(3.1)
where C is the measured capacitance; d is the separation distance between the two
Chapter 3. Computational Model 15
Figure 3.2: Parallel plate capacitive sensor opens up to become a planar capacitive sensor.
capacitor plates; A is the surface area; εr is the permittivity of the dielectric material
sandwiched by the two capacitor plates; and εo = 8.8541× 10−12 F/m is the permittivity
of free space. The dielectric permittivity, εr, is a measure of how easily a material is
polarized in response to an external electric field.
The two-electrode planar capacitive sensor, also shown in figure 3.2, has both electrodes
lying on the same plane and produces a fringing electric field that allows for one-sided
measurement of the dielectric material. The analytical formula for a two-electrode planar
capacitive sensor with electrodes of unit length and dielectric layer of unit height was
developed by Nassr et al. [34] using a conformal mapping technique [35]:
C = εoK(k
′o)
K(ko)(3.2)
where K() is a function of the complete elliptic integrals of the first kind with the elliptic
integral moduli ko = g/(s + g) and k′o =
√1− k2o . The parameters s and 2g represent
electrode width and electrode separation, respectively.
To test the hypothesis, three computational models were developed; one model for each
capacitive sensing mechanism shown in figure 3.1. A simulation was performed using each
model to generate a simulated respiratory waveform. By comparing the magnitude of the
capacitance change, ∆C, the hypothesis that the Respiration Pad works by capacitive
deformation sensing can be validated.
To standardize the simulations across all three models, the models were built using two
planar electrodes of unit width and length (1 cm x 1 cm) and negligible height (1 mm).
The electrodes are placed such that one electrode is underneath the sternum and the
other is spaced 30 cm apart towards the lumbar spine. A spacing of 30 cm between the
electrodes was used because this should be sufficient to cover the entire length of an
adult human lung. The models have two stationary electrodes (one transmitting and
one receiving) placed on top of a mattress and underneath cloth (bedsheets and patient
Table 3.2: Capacitance values derived from formulas and the models for parallel plateand planar capacitive sensors to verify correct computational model setup.
Chapter 3. Computational Model 17
Figure 3.3 shows the simulated respiratory waveforms generated by the computational
models for each of the sensing mechanisms; and figures 3.4, 3.5, and 3.6 show the
corresponding electric field distributions. The capacitance deformation sensing model
produced the largest capacitance change with ∆C = 0.085 pF, followed by the capacitive
displacement sensing model with ∆C = 0.015 pF. The capacitive volume sensing model
had ∆C = 0 pF indicating that changes in lung volume do not affect the measured
The goal of developing a computational model was to better understand the physical
principle behind capacitive sensing utilized by the Respiration Pad. Literature review cast
doubt on the original explanation of the sensing mechanism, thus, motivating this work.
Figure 3.4 shows that no change in the electric field occurs when the chest expands due to
inhalation. This is consistent with the simulated waveform for capacitive volume sensing
in figure 3.3(a), which shows a flat line indicating no change in capacitance. Furthermore,
figure 3.4 show that the electric field remains constant within the human torso but changes
outside the torso, which suggests that the electric field does not penetrate into the body.
Therefore, the results confirm that the electric field does not penetrate into the body and
that the Respiration Pad does not measure lung volume changes due to respiration.
Chapter 3. Computational Model 18
a: Resting. b: Chest expands.
Figure 3.4: Electric field distributions for capacitive volume sensing.
Figure 3.5 shows noticable changes in the electric field distribution when the chest lifts off
the mattress surface. This lift-off is consistent with the simulated respiratory waveform
in figure 3.3(b), which shows an exponential change in capacitance with a change of
∆C = 0.015 pF between the two extremes of resting and a unit displacement of the torso
above the electrode surface due to inhalation. However, the assumption of lift-off does not
apply when the patient is lying on a mattress because the mattress contours to the body
during respiration; therefore, no displacement can occur and no capacitance change is
possible on a mattress. Alternatively, if the Respiration Pad is placed on the floor, then,
in this case, there will be lift-off of the chest from the electrode surface due to the rigidity
of the floor.
a: Resting. b: Chest lifts.
Figure 3.5: Electric field distributions for capacitive displacement sensing.
Chapter 3. Computational Model 19
Figure 3.6 shows a significant change in the electric field distribution due to deformation
of the mattress surface, which corresponds to the simulated respiratory waveform in figure
3.3(c) showing a linear change in capacitance. The capacitance changes by ∆C = 0.085
pF between resting and a unit change in deformation of the mattress surface. This result
is consistent with formulas for parallel plate and planar capacitive sensors, which indicate
that as more deformation in the mattress surface occurs the more the electrodes rotate to
face each other to increase the effective surface area resulting in an increase in capacitance.
a: Resting. b: Mattress deformed.
Figure 3.6: Electric field distributions for capacitive deformation sensing.
There are some limitations with the computational model. First, three sensing mechanisms
were explored; however, it may possible that other physical phenomenon were unaccounted
for. Second, the model is simple and does not incorporate material properties of the
electrode and dielectrics aside from the human phantom, which may offset the capacitance
values but not the conclusions drawn from the results. The human phantom was also
simple, not of high resolution, and did not model all organs and tissues of the body.
Finally, stray capacitive effects and noise from the environment could not be modelled
and therefore was left out. Despite all these limitations, the computational model was
adequate enough to draw some clear conclusions about the sensing mechanism.
In summary, the results show that (1) capacitive volume sensing does not occur, (2) capac-
itive displacement sensing is not possible on a mattress, and (3) capacitive deformation
sensing results in the largest capacitance change in the simulated respiratory waveform.
Therefore, the Respiration Pad works by capacitive deformation sensing as hypothesized.
Having verified that the Respiration Pad works by capacitive deformation sensing, the
following text describes how electrode geometry optimization may be achieved using the
Chapter 3. Computational Model 20
computational model in future work. Figure 4.1 shows the interdigital capacitance sensor
(IDCS) design that was selected as the basis for the Respiration Pad. The IDCS utilizes
multiple electrode fingers to control the strength of the output signal by changing the
number (N), width (s), and separation (2g) of the electrode fingers. To optimize for the
electrode geometry, five models need to be developed; one model for each IDCS with
N = 2, 3, 4, 5, 6 electrode fingers. This leaves two parameters (s and g) for each model.
However, electrode separation (2g) is not an independent variable since the maximum
IDCS width was restricted to 30 cm (see chapter 4). Also, the IDCS length does not
affect the results of optimization so it can be taken as unity. Therefore, each model will
have only one independent variable, electrode width (s), to optimize for, with electrode
separation (2g) becoming a function of the geometry. Finding the optimal electrode
geometry means selecting a specific s, 2g, and N yielding the largest signal-to-noise ratio
(SNR), which is calculated as follows:
SNR = 10 log10
(Psignal
Pnoise
)∝ ∆C (3.3)
where Psignal and Pnoise are signal power and noise power, respectively. Since noise power
is independent of electrode geometry, the noise term drops out of the equation leaving
behind signal power as the only term. Furthermore, signal power can be reduced to
capacitance change ∆C. From this we understand that SNR is maximized when ∆C is
maximized. Therefore, optimization is possible by selecting the specific electrode geometry
that will yield the largest ∆C.
3.5 Summary
Three computational models were developed to verify the hypothesis that the Respiration
Pad works by capacitive deformation sensing. Results indicate that capacitive deformation
sensing had the strongest response, while capacitive volume sensing had no response.
Although, the model for capacitive displacement sensing produced a small response,
intuitively, we know that displacement sensing is not possible on a mattress because there
is no displacement of the torso from the electrode surface. Therefore, the results validate
the hypothesis and refute the original explanation of the sensing mechanism.
Chapter 4
Technology Redesign
4.1 Introduction
Hart et al. [15] made a significant contribution by developing a functional prototype of the
Respiration Pad to show that capacitance measurement of respiratory rate using a planar
capacitive sensor was possible. However, the initial prototype had several limitations:
• Non-durable: Electrodes made of conductive paint would flake off and deteriorate
after a few weeks of repeated use.
• Limited portability: Sensor head was a large and heavy yoga mat which is
difficult to seamlessly integrate with a hospital bed.
• Unsafe: Electronic circuits were breadboarded and not electrically isolated in an
enclosure, which exposes the electronics to damage and puts the patient at risk. No
electromagnetic or electrical safety assessment was done.
• No real-time operation: Software was buggy, slow, and requires manual user
interaction, which prevented the sensor from functioning automatically and inde-
pendently without user intervention.
A major component of this work was improving and refining the initial prototype by
addressing these limitations. By doing so, the current iteration of the technology is more
durable, safe, portable, and allows for real-time operation. The following section describes
the design choices that were made to achieve these improvements.
21
Chapter 4. Technology Redesign 22
4.2 Design choices
The interdigital capacitance sensor (IDCS) design, shown in figure 4.1, was selected
for the electrode configuration because it is versatile, extensively studied in literature,
and allows for electrode geometry optimization to improve signal quality by adjusting
the number (N), width (s), and separation (2g) of the electrode fingers. In this work,
electrode geometry optimization was not performed; instead, the selection of the number,
width, and separation of the fingers was chosen through trial and error. It was found that
N = 3 fingers, width of s = 2.5 cm, and separation of 2g = 10 cm yielded satisfactory
results and so was selected for the final design.
Figure 4.1: Interdigital capacitance sensor with electrode width (s), electrode separation(2g), number of electrode fingers (N), and applied voltage (±V ).
To improve the durability and portability of the sensor, four different sensor heads were
investigated. All together, four dielectric materials (plastic, cloth, thick fabric, and rubber)
and three conductive materials (conductive thread, conductive fabric, and conductive foil)
were used to build four distinct sensor heads shown in figure 4.2: Bedsheet Pad, Yoga
Pad, Fabric Pad, and Plastic Pad. Each sensor head was also qualitatively evaluated
using a human subject throughout development.
Chapter 4. Technology Redesign 23
a: Bedsheet Pad. b: Yoga Pad.
c: Fabric Pad. d: Plastic Pad.
Figure 4.2: Comparison of different sensor heads designed and constructed, includingsensor heads with (a) conductive fabric stitched onto a bed sheet, (b) conductive fabricglued and taped onto a yoga mat, (c) conductive thread stitched onto a thick fabric, and(d) conductive foil heat-sealed between two plastic sheets.
The best design was the Bedsheet Pad because it was durable, portable, and enabled
seamless integration with a hospital bed. Qualitative testing showed that this Pad
produced the best signal, which makes sense because the Bedsheet Pad fits very snugly
onto a hospital mattress enabling accurate measurement of mattress deformation due
to breathing. The Plastic Pad performed poorly because it would crumple up and
distort heavily when a body was present leading to inaccurate measurements of mattress
deformation. The Fabric Pad is similar to the Bedsheet Pad, but utilizes heavier cloth
and conductive thread as opposed to conductive fabric. However, its limitation was that
it was difficult to create wide electrodes using thin thread without leaving gaps or uneven
spacing leading to distortions in the electric field lines resulting in undesirable non-linear
effects. Finally, the Yoga Pad was created to directly improve upon the original design.
Chapter 4. Technology Redesign 24
Instead of silk-screened conductive paint, durable fabric electrodes were taped and glued
onto the mat surface. This solved the problem of the flaking paint; however, it was
difficult for the electrodes to remain firmly fixed on the mat surface after repeated use.
Furthermore, the yoga mat doesn’t address the problem of portability because it is still
relatively heavy and bulky. Ultimately, the Bedsheet Pad was chosen as the best of the
four designs.
The electronics of the Respiration Pad were custom designed from scratch, fabricated
in a foundry, and enclosed in a hard plastic enclosure, which improved durability of the
electronics and improved patient safety by reducing risk of electric discharge. Furthermore,
signal quality was improved by the addition of a hardware filter to increase noise immunity.
Real-time operation was made possible by developing a new integrated piece of software
from the ground up without relying on external applications. The new software was built
using Matlab and allowed for continuous and real-time capture, storage, filtering, and
processing of data from the Respiration Pad wirelessly via Bluetooth. The software works
by sending a command to the device to initiate transfer. The capture, storage, filtering,
and processing of data happens transparently in the background and, when complete,
plots of the respiratory signal showing the calculated respiratory rate is presented to
the user. Signal filtering was achieved by consecutively applying a lowpass and highpass
filter on the raw signal to produce a clean, filtered signal. The highpass filter has a
cut-off frequency of 0.05 Hz (corresponding to 3 breaths/min) and the lowpass filter has
a cut-off frequency of 1 Hz (corresponding to 60 breaths/min). The filters have a very
desirable sharp drop-off at the cut-off frequencies with minimal phase disturbance in the
frequency band of interest, which corresponds to respiratory rates in the range of 5 to 35
breaths/min. With the new software, real-time operation was made possible.
4.3 Final prototype
The Respiration Pad is a planar capacitive sensor that works by capacitive measurement
of small geometry fluctuations of the mattress surface due to mechanical deformation
induced by breathing. The Respiration Pad consists of three major components: sensor
head, electronics, and software.
The sensor head, shown in figure 4.3, is made from a fitted, elastic bed sheet containing
dry conductive fabric electrodes stitched onto the bed sheet. The bed sheet itself is fitted
Chapter 4. Technology Redesign 25
directly on top of a hospital twin-sized foam mattress found on a typical hospital stretcher.
The sensor head may be covered with additional linen to isolate the electrodes from
the patient who lies on top of the mattress and the linen. The sensor head is the only
component of the system in contact with the patient and it is isolated from the electronics
by 5 kilovolts of optical isolation so that any surge from the main power supply will not
be an issue. Furthermore, the electric field generated by the electrodes is very small,
does not penetrate into the body, and decays exponentially with distance. Therefore,
the sensor is safe for patient use from both an electrical and electromagnetic exposure
point-of-view (see chapter 5 for more details).
Figure 4.3: Sensor head consisting of electrodes made from conductive fabric; buttons tointerface with the coaxial cable; and a fitted bed sheet, which fits onto a foam mattress.
The electronics consist of two stackable boards enclosed by a small plastic case. The
two boards communicate with each other via a board interconnect. The first board
handles wireless transmission of data via the Bluetooth module and distributes power to
the entire circuit by receiving power via the micro-USB connector. The second board
handles capacitance measurement using an off-the-shelf capacitance chip (Analog Devices
AD7746) and ensures safety of the subject through 5 kilovolts of data/power isolation
using a micro-transformer in compliance with IEC 60601-1 and IEC 60950-1 standards.
The AD7746 capacitance chip is capable of taking measurements with an accuracy of
4 fF over a dynamic range of 4 pF by using a 32 kHz square-wave excitation signal at
a maximum of 1.65 volts and 21 microamps. The capacitance chip is protected from
electrostatic discharge by ESD protection.
Chapter 4. Technology Redesign 26
Figure 4.4: Sensor electronics consisting of micro-USB connector, Bluetooth module,board interconnect, micro-transformer, AD7746 capacitance chip, and ESD protection.
The electrical components, shown in figure 4.5, consists of a several components that
connect the separate pieces of the sensor together as one functional unit. The electronics
receive power through a USB cable, which connects to a medical grade wall mount power
adapter with USB charger. The power adapter complies with UL 60601-1 and IEC/EN
60601-1 electrical standards. A hardware filter, a front-end to the capacitance chip, is
used to improve signal quality by reducing the noise in capacitance measurement and
filtering out the frequencies not of interest. The coaxial cables connect the electronics to
the hardware filter and from the hardware filter to the button connectors, which snap
onto buttons found on the electrodes stitched on the sensor head. The device is powered
on automatically when the power adapter is plugged into a wall outlet.
Figure 4.5: Sensor electrical components consisting of power adapter, USB cable, hardwarefilter, coaxial cables, and button connectors.
Chapter 4. Technology Redesign 27
The software was developed in the Matlab scripting language and runs on a laptop with
the Matlab runtime environment installed. The software is an integrated application
that allows for automated and continuous capture, processing, and display of data. The
software achieves this by sending the command to the Respiration Pad to initiate and
transmit capacitance measurements. The software receives the wirelessly transmitted data
by interfacing with a Bluetooth receiver connected to the laptop. Since the transmission
is wireless, the laptop need not be present in the patient vicinity and, therefore, is not
medical grade. Once wireless data capture is complete, the user has a chance to select
the segment of data to be filtered and processed. Processed data is then displayed to the
user as plots of the respiratory waveforms before and after filtering, along with the power
spectrum, and calculated respiratory rate as shown in figure 4.6.
Figure 4.6: Sensor software showing plots of the raw and filtered signal with the corre-sponding power spectrum and calculated respiratory rate.
4.4 Summary
This chapter discusses the limitations with the original Respiration Pad and the improve-
ments that were made to create a more robust and capable sensor in preparation for the
feasibility study. Design choices were discussed that led to a significant improvement
in the durability, portability, safety, and real-time operation of the technology. Work
involved evaluating new materials for the sensor head, developing custom electronics,
and creating a new integrated software to handle capture, processing, and display of the
respiration signal. Finally, the final prototype of the technology was presented.
Chapter 5
Safety Assessment
5.1 Introduction
Safety is of paramount concern for a medical device when used on patients. Before a
medical device can be approved for use on patients it must meet several safety criteria,
including (1) electromagnetic exposure safety, (2) electrical safety, and (3) medical device
classification. This chapter deals with these safety considerations in detail.
5.2 Electromagnetic exposure safety
Electromagnetic fields are composed of orthogonal electric and magnetic field components.
A strong electric field may be dangerous to patient health because, although no direct
current is injected into the body, the body may experience displacement currents induced
by the electric field which have the potential to affect biological tissues. The type of
interaction experienced by the human body depends on the frequency and strength of the
electric field. At low frequencies (3 to 100 kHz), the interaction of the displacement currents
with nervous system tissue is of concern because of the sensitivity of nervous system tissue
to these currents [36]. Above 100 kHz, the nervous tissue becomes less sensitive to direct
stimulation by electromagnetic fields; but the heating effect due to electromagnetic fields
becomes the major mechanism of interaction [36]. Since the Respiration Pad operates
below 100 kHz, the heating effect is not of concern.
The Respiration Pad generates non-ionizing radiation in the form of a 32 kHz electric field
28
Chapter 5. Safety Assessment 29
to make measurements, which may have the potential to adversely affect human health.
To ensure that the Respiration Pad is safe for human use, it has been assessed against
Health Canada’s Safety Code 6 regulations [37]. Specifically, the maximum electric field
strength generated by the sensor is compared against the maximum permissible exposure
limit. Although, the Respiration Pad is a “contactless” device in that no direct skin
contact is required for operation, in order to account for accidental direct skin contact,
the contact current limits were also considered. As shown in table 5.1, the maximum
electric field strength of 1.65 V/m, for the Respiration Pad, is orders of magnitude smaller
than the maximum permissible exposure limits of 600 V/m and 280 V/m for occupational
and general public limits, respectively. Similarly, the maximum current of 0.021 mA,
that may accidentally be exposed to the patient, is also orders of magnitude below the
contact current limits of 32 mA and 14.4 mA for occupational and general public limits,
respectively. This shows that the Respiration Pad is well under the prescribed exposure
limits and, therefore, is safe for use.
Exposure Limits
Safety Limits Occupational General Public Respiration Pad Levels
MPE (V/m) 600 280 1.65
Contact Current (mA) 32 14.4 0.021
Table 5.1: Comparison of electromagnetic maximum exposure and contact current limitsfrom Health Canada’s Safety Code 6 against those of the Respiration Pad.
5.3 Electrical safety
The Respiration Pad has received equipment certification to the medical electrical standard
(ANSI/AAMI ES60601-1:2005 [38]) and was classified as a class 2 BF equipment. Class
2 equipment means that no single failure of this device can result in dangerous voltage
becoming exposed to the patient such that it may cause an electric shock. Furthermore,
this is achieved without relying on an earthed metal casing. BF means that the device is
a type B equipment with a type F applied part. Type B equipment is one that provides a
particular degree of protection against electrical discharge through grounding. Type F
applied part is one that extends from the patient into the equipment and is isolated from
all other parts of the equipment.
Chapter 5. Safety Assessment 30
5.4 Medical device classification
Medical devices in Canada are subject to varying degrees of regulation depending on
its risk category. Indicators of risk include degree of invasiveness, duration of contact,
body systems affected, and local versus systemic effects [39]. The Respiration Pad is a
medical device because it is a medical instrument intended for diagnosis of abnormal
respiratory states in patients. The Respiration Pad is a non-invasive, contactless, and
active device with a measuring function. Non-invasive means that no part of the body is
penetrated; no body orifice is contacted; it’s not for surgical use, and it’s not implantable.
Contactless means that there is no direct skin contact with the patient. Active means
that an external power source is used and energy conversion only happens within the
device itself. No energy conversion takes place between the device and human tissue or
within human tissue. Finally, a measuring function means that the device is intended to
quantitatively measure a physiological parameter such that the result of the measurement
is displayed in legal units.
The Respiration Pad received a preliminary medical device assessment as a class 2
medical device. The justification is that the intended purpose of the device plus software
implies accuracy where non-compliance could adversely effect patient safety. Furthermore,
predicate devices are class 2.
5.5 Summary
Electromagnetic exposure safety was assessed by ensuring that the generated electric
field and leakage current are within tolerable levels according to Health Canada’s Safety
Code 6. The assessment showed that the Respiration Pad operated within the prescribed
thresholds. Electrical safety was assessed by a qualified technician to ensure that the
Respiration Pad conformed to CSA standards for electrical equipment. The Respiration
Pad successfully passed certification and was classified as a class 2 BF equipment. Finally,
the Respiration Pad was classified as a class 2 medical device.
Chapter 6
Feasibility Study
6.1 Introduction
The purpose of this feasibility study was to determine whether the Respiration Pad can
produce measurements that agree with those from an established method sufficiently
well for clinical purposes. The hypothesis was that the Respiration Pad does statisti-
cally agree sufficiently well for clinical purposes. The actual clinical gold standard is
respiratory inductance plethysmograph (RIP). However, the Respiration Pad cannot be
used simultaneously with RIP because the metal in the RIP interferes with capacitance
measurement by the Respiration Pad. Therefore, an airflow meter (TSI 4040), which
allowed for continuous and synchronized data capture was used as a substitute. For the
purposes of this study, this substitution was considered satisfactory.
6.2 Methods
Figure 6.1 describes the experimental setup. The study simultaneously measured res-
piratory rates with the PAD and TSI using 1 healthy adult subject under 6 different
combinations of clothing (either t-shirt or coat) and added weights (0 lbs, 10 lbs, or 20 lbs
extra). Each measurement was taken for 15 seconds across 4 different respiratory rates
(RR) of 5, 15, 25, 35 breaths/min and 3 different tidal volumes (TV) of shallow, normal,
and deep breathing corresonding to clinical values of 250, 500, and 1000 mL/breath,
respectively [40]. Each simulated body type combination was taken as an independent
31
Chapter 6. Feasibility Study 32
measurement. A metronome was used to set breathing at the appropriate RR by beeping
at every inhale and exhale. A separate spirometer (by Datex) ensured that the subject
respired at a consist TV by visually displaying the tidal volume measurements on a
monitor in real-time.
a: Subject position b: Respiration readings
Figure 6.1: (a) shows the subject lying supine with the electrodes positioned between theshoulders and hip; (b) shows the readings from the PAD.
Many studies use the Pearson’s correlation coefficient (PCC) as an indicator of agreement
between two devices. However, this is statistically incorrect because the PCC measures
the strength of a relation between two variables, not the agreement between them
[29]. Therefore, to properly validate the hypothesis, several analyses were conducted to
statistically quantify the degree of agreement between the two devices which both measure
a continuous, numerical variable. The analyses used in this study were Bland-Altman
analysis, a single-factor analysis of variance (ANOVA), paired t-test, intraclass correlation
coefficient (ICC), and the British Standards Institute reproducibility coefficient (BSI
coefficient). The PCC is included for completeness.
6.3 Results
Boxplots of the results were generated in figure 6.2 where misread values, shown in figure
6.3, were removed and measurements were repeated until a proper reading was taken.
Figure 6.4 shows the spread of the data from the line of equality for each TV. The
Bland-Altman plots in figure 6.5 show the distribution of these average RR measures. Of
Chapter 6. Feasibility Study 33
Shallow Normal Deep0
5
10
15
20
25
30
35
40
45
Tidal Volume
Re
sp
ira
tory
Ra
te (
min
−1)
Figure 6.2: Boxplots of the results were generated for both the TSI and PAD measurementswith TV along the x-axis and RR along the y-axis. Measurements were grouped in pairsto facilitate visual comparison between the devices. Each pair has the TSI measurements(coloured black) on the left and PAD measurements (coloured blue) on the right.
the 72 pairs of RR measurements from the TSI and PAD, only 1.4% (95% CI, -3.0% to
5.8%) of the values measured by the PAD exceeded the clinically significant threshold of
4 breaths/min.
Slicing the results by RR, as shown in table 6.1, we see that the mean difference for any
RR does not exceed 1 breath/min. The mean difference across all RR is on average -0.02
(95% CI, -0.34 to 0.31), which includes 0 in the confidence interval and, therefore, this
difference is not statistically significant. This is also confirmed by using paired t-tests,
which failed to reject the null hypothesis of equivalence (P > 0.92 on average for 5%
significance, n− 1 = 5 degrees of freedom, and two-tails probability). The BSI coefficient
indicates the maximum likely difference between a pair of readings, which in our case
should not exceed the clinically significant threshold of 4 breaths/min. The average BSI
coefficient across all RR is 2σ = 2.76, which is below the clinical threshold. The largest
BSI coefficient occurs at the highest RR of 35 breaths/min with a value of 4.05 which is
reasonably close to the threshold.
Chapter 6. Feasibility Study 34
0
1
2
3
4
5
6
7
8
shallow normal deep
# o
f M
isre
ad
s
Tidal Volume
coat tshirt
Figure 6.3: Sometimes, during the study, the PAD would misread the RR and, therefore,repeated measurements were taken until a proper reading occurred.
RR (min−1) Mean Difference BSI Coefficient Paired T-Test (P-value)
5 0.20 (-0.05 to 0.45) 1.00 0.11
15 0.01 (-0.49 to 0.51) 2.01 0.98
25 -0.26 (-1.04 to 0.52) 3.15 0.49
35 -0.01 (-1.02 to 1.00) 4.05 0.99
Average -0.02 (-0.34 to 0.31) 2.76 0.92
Table 6.1: Results sliced by RR.
Slicing the results by TV, as shown in table 6.2, no significant difference was found between
the PAD and TSI (P > 0.96 for all TV) based on single-factor ANOVA. Measurements
from the PAD and TSI are in high agreement with an average ICC value of 0.9926 (95%
CI, 0.9882 to 0.9954). As a point of reference, the flash mob research paper [6] calculated
the average ICC for RR as 0.26 (95% CI, 0.16 to 0.35) and for pulse rate as 0.65 (95%
CI, 0.58 to 0.71). Although, it is statistically incorrect to use the Pearson’s correlation
coefficient (PCC), the PCC values are near unity and comparable to the ICC values.
Chapter 6. Feasibility Study 35
0 5 10 15 20 25 30 35 40 450
5
10
15
20
25
30
35
40
45
RR
fro
m P
AD
(m
in−
1)
RR from TSI (min−1
)
a: Shallow
0 5 10 15 20 25 30 35 40 450
5
10
15
20
25
30
35
40
45
RR
fro
m P
AD
(m
in−
1)
RR from TSI (min−1
)
b: Normal
0 5 10 15 20 25 30 35 40 450
5
10
15
20
25
30
35
40
45
RR
fro
m P
AD
(m
in−
1)
RR from TSI (min−1
)
c: Deep
0 5 10 15 20 25 30 35 40 450
5
10
15
20
25
30
35
40
45
RR
fro
m P
AD
(m
in−
1)
RR from TSI (min−1
)
d: All
Figure 6.4: A scatter plot was generated for each TV with the TSI measurements onthe x-axis and the corresponding PAD measurements on the y-axis for all RR. Theseplots show the spread of the data from the line of equality for each TV. The spread istechnically not a measure of agreement but of correlation. The smaller the spread thehigher the correlation between the measurements.
Chapter 6. Feasibility Study 36
4.7 4.8 4.9 5 5.1 5.2 5.3 5.4 5.5 5.6 5.7−5
−4
−3
−2
−1
0
1
2
3
4
5
Diffe
ren
ce
in
RR
(m
in−
1)
Average RR (min−1
)
a: 5 breaths/min
14 14.5 15 15.5 16 16.5−5
−4
−3
−2
−1
0
1
2
3
4
5
Diffe
ren
ce
in
RR
(m
in−
1)
Average RR (min−1
)
b: 15 breaths/min
22.5 23 23.5 24 24.5 25 25.5 26 26.5−5
−4
−3
−2
−1
0
1
2
3
4
5
Diffe
ren
ce
in
RR
(m
in−
1)
Average RR (min−1
)
c: 25 breaths/min
32.5 33 33.5 34 34.5 35 35.5 36 36.5 37 37.5−5
−4
−3
−2
−1
0
1
2
3
4
5
Diffe
ren
ce
in
RR
(m
in−
1)
Average RR (min−1
)
d: 35 breaths/min
Figure 6.5: Bland-Altman plots of the results were generated using individual measurementdifferences in RR between the TSI and PAD (y-axis, “Difference in RR”) and the bestestimate of the actual RR (x-axis, “Average RR”). The middle blue dashed line showsthe average RR difference (0.20, 0.01, -0.26, and -0.01 breaths/min for A,B,C, and D,respectively) while the upper and lower red dashed lines depict the clinically significantthreshold for the difference of ±4 breaths/min.
Chapter 6. Feasibility Study 37
TV ICC PCC ANOVA (P-value)
Shallow 0.9828 (0.9609 to 0.9925) 0.9824 (0.9590 to 0.9925) 0.963
Normal 0.9957 (0.9902 to 0.9982) 0.9957 (0.9900 to 0.9982) 0.975
Deep 0.9997 (0.9993 to 0.9999) 0.9997 (0.9993 to 0.9999) 0.975
Average 0.9926 (0.9882 to 0.9954) 0.9925 (0.9880 to 0.9953) 0.993
Table 6.2: Results sliced by TV.
6.4 Discussion
The aim of this study was to determine the level of agreement between the PAD and
TSI to validate the hypothesis that the PAD, when compared to the gold standard (TSI),
performs sufficiently well for simple measurement of respiratory rate. All the results
indicate this is so.
The boxplots and scatter plots visually show strong agreement between both devices. The
ICC and PCC values both quantify the strength of agreement as being very strong and
the BSI coefficient indicates that the maximum likely difference between a pair of readings
does not exceed the clinically significant threshold of 4 breaths/min. Furthermore, the
ANOVA and paired t-tests confirm that there is no statistical difference between the PAD
and TSI measurements. Therefore, it is safe to conclude that the Respiration Pad can
accurately measure respiratory rate for a wide range of RR from 5 to 35 breaths/min and
TV from 250 mL to 1000 mL.
In the study, different body types were simulated using clothing and additional weights
placed on top of a thin subject. From figure 6.3, we see that misreadings occur with
the PAD when the subject wears a large coat. Misreadings refer to instances when
the Respiration Pad is not able to properly measure respiration. Misreadings can be
identified when the power in the signal is overwhelmed by the power in the noise. This
may occur for numerous reasons including stray noise in the electrodes and circuitry
due to electromagnetic interference from the environment, static discharge from patient
clothing or mattress linen corrupting capacitance measurement, and motion and breathing
artefacts. Unfortunately, the actual capacitance values are extremely small in the 0.1
to 1 picofarad (pF) range and the actual size of the waveform measured as a change in
capacitance, ∆C, is even smaller in the 0.01 to 0.1 pF range. Therefore, the fact that
Chapter 6. Feasibility Study 38
misreadings occur is not very surprising and may even be worse outside the controlled
setup of this study.
In this study, measurements were repeated until no misreading occurred. Admittedly,
this may improperly bias the results in favour of the Respiration Pad. Misreadings occur
due to interference in the capacitive measurement due to environmental factors. The
difficulty here is that it may not be possible to differentiate a misreading from sleep
apnea experienced by the patient or from a sensor malfunction, for example, due to a
disconnected wire. Therefore, misreads will reduce the overall reliability of the Respiration
Pad.
Ideally, there shouldn’t be any misreadings, but there are several ways to mitigate the
potential for misreadings. The first approach is to reduce the potential for misreadings
by improving the sensitivity and robustness of the sensor by optimizing the electrode
geometry, investigating use of different materials that will reduce electrical discharge
causing capacitive interference, and by incorporating active shielding into the sensor design
to isolate the electrical field to a particular sensing region to limit external electromagnetic
interference. Furthermore, the Respiration Pad is intended for continuous use, therefore,
even if misreadings do occur once for every 15 seconds of capture, there will be 4 captures
ever minute and over the span of a few minutes, the misreadings can be removed by signal
analysis and the actual trend in the respiratory rate can be identified.
The amount of misreadings are highest for shallow breathing but decrease as tidal volume
increases. This inverse relationship makes sense since shallow breathing is most difficult
to detect whereas deep breathing is easier to detect. Misreadings did not occur when the
subject wore a t-shirt, which indicates that the size of the patient affects the capacitance
measurement by the PAD. Also, the TSI did not have any misreadings as expected, since
it is the clinical gold standard.
One of the original concerns that prompted this feasibility study was whether the Respi-
ration Pad was sensitive enough to accurately measuring RR during shallow breathing.
Despite the misreadings, results show that the new Respiration Pad is sensitive enough
even during shallow breathing with a mean difference in measured values of -0.20 (95%
CI, -0.05 to 0.45) breaths/min. Therefore, this feasibility study confirms that there has
been an improvement in the technology since the original Respiration Pad.
Currently, the Respiration Pad has been designed to exclusively measure respiratory
rate. However, from a clinical point of view, there is significant value in also measuring
Chapter 6. Feasibility Study 39
quantities such as minute volume (MV) and tidal volume (TV) in non-intubated patients.
This limits the clinical value of the technology in practice since patients in need of
monitoring typically have substantial cardiopulmonary co-morbidities such as asthma,
chronic obstructive pulmonary disease, obstructive sleep apnea, and congestive heart
failure [41]. For example, obese patients suffering from obstructive sleep apnea may
experience a situation where there is chest wall motion without air exchange due to
chronic airway obstruction. Such a situation may present itself as breathing to a health
care provider observing a patient, but would result in consistently low TV or MV readings,
which could be used as an early indicator of impending respiratory failure [42]. Therefore,
to improve the clinical utility of the Respiration Pad, the device needs the capability to
measure TV or MV.
It has been shown using the computational model that the Respiration Pad works through
capacitive deformation sensing, which means that the sensor detects capacitance changes
caused by changes in electrode geometry on the mattress surface as a person breaths.
Therefore, what is actually being measured is respiratory effort of the patient indirectly
through body motion. This is a very indirect method of respiratory measurement and is
sensitive to confounding environmental factors such as mattress elasticity and presence of
metal in the sensing region as well as patient body size and electromagnetic interference
from the environment.
This indirect method of respiration measurement coupled with the limitations of the
technology suggest that it may be difficult for this technology to accurately measure tidal
volume. Typically, what would be done to measure tidal volume, is to calibrate the device
against a spirometer for each patient so that the measured amplitude of the respiration
signal can be multiplied to get the tidal volume. However, with the Respiration Pad, it
may not be possible to ensure a consistent amplitude between use because the Respiration
Pad is sensitive to not only mattress elasticity, patient clothing, and patient size, but
potentially also to patient orientation, posture, position, and motion artefacts.
In contrast, the RIP technology uses two elastic bands placed around the chest and
abdomen to measure changes in cross-sectional area, which is linearly related to lung
volume. This allows the RIP to more directly measure respiratory effort while minimizing
environmental effects. Another advantage is that the RIP is less sensitive to distortion
of the chest and abdomen compartments during motion or posture change; therefore,
this reduces the effect of patient posture, orientation, position, and motion artefacts
on the measured signal. Furthermore, studies using RIP show that in normal subjects,
Chapter 6. Feasibility Study 40
the measured value remains within ±10% after calibration for 93% of the breaths [43],
thus, making RIP a more reliable method for monitoring tidal volume in spontaneously
breathing patients than the Respiration Pad.
The Respiration Pad aims to be a contactless alternative to the RIP technology. The
major difference being that the Respiration Pad does not require electrodes or bands
attached to the patient’s torso. However, this trade-off results in many limitations and
weaknesses in the Respiration Pad technology and may not be justifiable.
As mentioned earlier, the Respiration Pad indirectly measures respiratory effort. However,
respiratory effort alone cannot be used to detect more complex respiratory disorders
such as paradoxical breathing. During paradoxical breathing, the airflow is out of
phase with respiratory effort and chest movement is unsynchronized, or paradoxical,
to breathing efforts. Detecting paradoxical breathing requires airflow measurement or
at least measurement of the volume change of the abdomen and chest compartments
[44]. This is currently not possible with the Respiration Pad because it has only one
measurement channel, but it is possible with the RIP. Even if the capability to measure
two compartments was added to the Respiration Pad, the device will still suffer from
the limitations mentioned previously, especially the fact that the measured variable (i.e.
deformation of the mattress surface) is not linearly related to lung volume. Therefore,
accurately detecting lung volume may not be possible, which makes the technology unable
to detect complex respiratory disorders like paradoxical breathing.
This feasibility study has some major limitations. Only 1 healthy human subject was
used to simulate 6 different body types by using weights and clothing, which is not very
realistic. Furthermore, no diverse patient population was used. No long-term evaluation
was performed; ideally, an overnight study would help show that the Respiration Pad is
capable of continuous, long-term measurement of respiratory rate. Also, no evaluation
of the effect of motion and breathing artefacts on the measured signal was performed.
Finally, no evaluation of the effect of patient posture, orientation, or position was explored
in this study.
Some of these limitations can be addressed first by further improving the technology then
performing a proper feasibility study. The signal quality can be improved by optimizing the
electrode geometry through a computation model. Other improvements include, adding
shielding to allow the sensor to work on a spring mattress, as well as, adding multiple
channels to allow detecting paradoxical breathing. Finally, a proper feasibility study can
be performed by repeating the study with a diverse patient population; performing longer
Chapter 6. Feasibility Study 41
breathing captures; investigating the effect of motion and breathing artefacts on signal
quality; analyzing the effect of patient posture, position and orientation on measurement
accuracy; and investigating the sensitivity and specificity of detection of paradoxical
breathing using multiple channels.
6.5 Summary
This chapter describes a major aspect of this work, which was to validate the feasibility
of this technology. The aim was to validate the hypothesis that the Respiration Pad can
produce measurements that agree with those from an established method sufficiently well
for clinical purposes.
A feasibility study was conducted using a single human subject with an airflow meter as
the clinical reference method. The results showed that the Respiration Pad is an accurate
and unbiased method for respiratory rate measurement. Accurate means that there is
strong agreement with the clinical reference method with an average ICC value of 0.99
and an average BSI coefficient of 2.76 < 4 breaths/min clinical threshold. Unbiased means
that the mean difference between the two methods is -0.02 breaths/min on average.
However, the feasibility study also revealed many of the technology’s limitations. The
Respiration Pad cannot currently measure tidal volume or be used to detect complex
breathing disorders such as paradoxical breathing, and it may not be possible to accurately
do so. Furthermore, the measured signal is extremely small and is prone to misreadings.
Finally, the technology cannot currently be used on spring mattresses and may not work
in the presence of metal from other instruments, wires, or clothing. Therefore, even
though the feasibility study has shown that the Respiration Pad can accurately measure
respiratory rate even at shallow tidal volumes, the technology may not be useful for more
complex respiratory measurements.
Chapter 7
Conclusion
A contactless planar capacitive sensor was designed and developed. A feasibility study
was used to evaluate the performance of the sensor against an airflow meter using a single
person. Results indicate that the Respiration Pad is useful for simple measurement of
respiratory rate and that the measurements are unbiased and accurate. However, the
study also revealed that due to limitations of the technology, the Respiration Pad is not
suitable for measurement of tidal volume or complex breathing disorders. Ultimately, the
goal of this work was to create a technology that would potentially improve respiratory
rate documentation within hospitals. Despite the limitations, results indicate that the
Respiration Pad is adequate for this purpose.
7.1 Original contributions
This work has a substantial engineering component with a small clinical component in
the form of a feasibility study to evaluate the technology. The original contributions of
this work are described as follows:
1. Computational model: Showed that the Respiration Pad works by capacitive
deformation sensing using a computational model.
2. Technology redesign: Significantly improved the durability, portability, safety,
and enabled real-time operation of the technology by testing new materials for the
sensor head, and developing the electronics and software from scratch.
3. Safety assessment: Ensured Respiration Pad was safe for patient use by assessing
42
Chapter 7. Conclusion 43
the technology for both electromagnetic exposure safety using Health Canada’s Safety
Code 6 and electrical safety through certification to CSA standard. Furthermore,
the device received medical device assessment as a class 2 medical device, which is
required before the technology can be evaluated on patients.
4. Feasibility study: Showed that the Respiration Pad is accurate (i.e. strong
agreement with clinical reference method with an average ICC value of 0.99 and
an average BSI coefficient of 2.76 < 4 breaths/min clinical threshold) and unbiased
(i.e. average mean difference of -0.02 breaths/min) by conducting a feasibility study
using a human subject with an airflow meter as the clinical reference method.
7.2 Limitations of work
Due to large scope of this project and the difficult engineering challenges in the development
of the technology, several limitations in this work exist. These limitations include:
1. Sensor optimization: No optimization of the electrode geometry using the com-
putational model was performed.
2. Feasibility study: Only 1 health human subject was used to simulate 6 different
body types by using weights and clothing, which is not very realistic. No long-term
evaluation was performed; ideally, an overnight study would help to show that the
Respiration Pad is capable of continuous, long-term measurement of respiratory
rate. Finally, no evaluation of the effect of motion and breathing artefacts, patient
posture, orientation, and position on the measured signal was explored.
7.3 Limitations of technology
Although the results of this work are positive and show that the Respiration Pad is
capable of accurate and unbiased measurements of respiratory rate, several key limitations
of the technology have been identified. These limitations severely limit the potential of
this technology for clinical use. The limitations of the technology include:
1. Tidal volume: Measurement of tidal volume is required to characterize a patient’s
ventilation status, which is a better indicator of a patient’s physiological condition
Chapter 7. Conclusion 44
than simply respiratory rate alone. Since the Respiration Pad does not currently
have the capability to measure tidal volume, it’s clinical utility is limited.
2. Qualitative and non-linear: RIP is a quantitative method for respiratory mea-
surement because it measures changes in cross-sectional area of the chest, which is
linearly related to lung volume. In contrast, the Respiration Pad measures capacitive
deformation of the mattress surface, which is non-linearly related to lung volume
and therefore cannot be considered quantitative. Furthermore, the Respiration
Pad is sensitive to not only mattress elasticity, patient clothing, and patient size
but potentially also to patient orientation, posture, position, and motion artefacts.
Therefore, it will be difficult to maintain a calibrated state over time.
3. Misreadings: The measured capacitance values are extremely small and, therefore,
sometimes electromagnetic interference, noise, or motion artefacts in the signal can
cause misreadings to occur. Furthermore, results showed that the signal becomes
corrupted in the presence of metal, which makes the Respiration Pad unusable on a
spring mattress. It may be possible to overcome this using shielding to remove the
effect of metal in the spring mattress; however, this does solve the problem when
metal is present above the Respiration Pad near the patient area such as in patient
clothing or on the mattress itself due to electrodes and wires from other devices.
4. Complex breathing: The Respiration Pad has only 1 measurement channel,
which means its use in sleep apnea monitoring and detecting complex breathing
disorders is limited. For example, paradoxical breathing, which occurs in obstructive
sleep apnea patients requires two independent channels measuring signals of both
the chest and abdominal walls for proper assessment. Furthermore, sleep apnea
monitoring requires a more complete picture of ventilation status, which includes
tidal volume measurement. Since the Respiration Pad may not have the ability to
reliably and accurately measure tidal volume, maintain a calibrated state, and avoid
misreadings, it will be of very little clinical use in monitoring complex breathing
disorders including sleep apnea.
7.4 Directions for future work
With these limitations in mind, the clinical usefulness of this technology beyond simple
respiratory rate monitoring is limited. However, if respiratory rate measurement alone is
Chapter 7. Conclusion 45
sufficient, then there is potential for improvement in this area. Directions for future work
include:
1. Technology improvement: Signal quality may be improved by optimizing the
electrode geometry through a computation model. Other improvements include
upgrading the wireless transmitter to Bluetooth low energy to reduce size and power
consumption, adding shielding to allow the sensor to work on a spring mattress,
and adding multiple channels to potentially allow detecting more complex breathing
disorders such as paradoxical breathing and sleep apnea.
2. Clinical pilot: Address limitations of feasibility study by repeating the study with
a diverse patient population, performing longer breathing captures, investigating
the effect of motion and breathing artefacts on signal quality, and analyzing the
effect of patient posture, position and orientation on accuracy.
3. Sleep quality screening: There is potential for sleep quality assessment in the
home by measuring proxy variables such as restlessness during sleep, bed-exit events,
and respiratory variability during sleep.
Overall, this work was a major step in pushing the Respiration Pad forward. The
conclusion is that the Respiration Pad has promise for simple respiratory rate monitoring
to improve vital signs documentation in hospitals and has the sensitivity to accurately
measure respiratory rate even during shallow breathing.
Bibliography
[1] K. Bobay, K. Fiorelli, and A. Anderson. Failure to rescue. J Nurs Care Qual, 2008.
[2] K. Hillman, P. Bristow, and T. Chey et al. Antecedents to hospital deaths. Intern
Med J, 2001.
[3] M. Buist, E. Jarmolowski, and P. Burton et al. Recognising clinical instability in
hospital patients. Med J Australia, 1999.
[4] A. Tee, P. Calzavacca, and E. Licari et al. The challenges of researching and adopting
medical emergency teams. Crit Care, 2008.
[5] E. Thomas, D. Studdert, and H. Burstin et al. Incidence and types of adverse events
and negligent care in utah and colorado. Med Care, 2000.
[6] M. Semler, D. Stover, and A. Copland et al. A single-day, multicenter, resident-
directed study of respiratory rate. American College of Chest Physicians, 2013.
[7] H. Gao, A. McDonnell, and D. Harrison et al. Systematic review and evaluation of
physiological track and trigger warning systems for identifying at-risk patients on
the ward. Intensive Care Med, 2007.
[8] M. DeVita, G. Smith, and S. Adam et al. Identifying the hospitalised patient in
crisis. Resuscitation, 2010.
[9] Goldhill and A. McNarry. Physiological abnormalities in early warning scores are
related to mortality in adult inpatients. British Journal of Anaesthesia, 2004.
[10] K. Hillman, J. Chen, and M. Cretikos. Introduction of the medical emergency team
(met) system. Lancet, 2005.
[11] J. Welton, L. Unruh, and E. Halloran. Nurse staffing, nursing intensity, staff mix,
and direct nursing care costs across massachusetts hospitals. J. Nurs Admin, 2006.
46
BIBLIOGRAPHY 47
[12] P. Lynch. Managing the costs of medical equipment maintenance. Biomedical
instrumentation and technology, 2009.
[13] M. Cretikos, R. Bellomo, and K. Hillman et al. Respiratory rate: the neglected vital
sign. Medical Journal of Australia, 2008.
[14] R. Parkes. Rate of respiration: The forgotten vital sign. Emergency Nurse, 2011.
[15] A. Hart, K. Tallevi, and J. Cafazzo et al. A contact-free respiration monitor for
smart bed and ambulatory monitoring applications. EMBC, 2010.
[16] J. Silber, S. Williams, H. Krakauer, and J. Schwartz. Hospital and patient character-
istics associated with death after surgery. Med Care, 1992.
[17] C. Leuvan and I. Mitchell I. An observational study of vital sign measurements.
Critical Care Resuscitation, 2008.
[18] S. Mukkamala, C. Gennings, and R. Wenzel. Bias in the reporting of respiratory
rates. American Journal of Emergency Medicine, 2008.
[19] T. Brennan, A. Localio, and L. Leape et al. Identification of adverse events occurring
during hospitalization. Ann Intern Med, 1990.
[20] F. Al-Khalidi, R. Saatchi, and D. Burke et al. Respiration rate monitoring methods.
Pediatric Pulmonology, 2011.
[21] K. Ashutosh, R. Gilbert, and J. Auchincloss et al. Impedance pneumograph and
magnetometer methods for monitoring tidal volume. J. Appl. Physiol., 1974.
[22] K. Cohen, W. Ladd, and D. Beams et al. Comparison of impedance and inductance
ventilation sensors on adults. IEEE Trans. Biomed. Eng., 1997.
[23] M. Folke, L. Cernerud, M. Ekstrom, and B. Hok. Critical review of non-invasive
respiratory monitoring in medical care. Med. Biol. Eng. Comput., 2003.
[24] P. Hult, T. Fjllbrant, and B. Wranne et al. An improved bioacoustic method for
monitoring of respiration. Technology and Health Care, 2004.
[25] N. Bu, N. Ueno, and O. Fukuda. Monitoring of respiration and heartbeat during
sleep using a flexible piezoelectric film sensor. IEEE EMBS, 2007.
[26] T. Kang, C. Merritt, and B. Karaguzel et al. Sensors on textile substrates for