Content uploaded by Hisham Alshaer
Author content
All content in this area was uploaded by Hisham Alshaer on Oct 09, 2015
Content may be subject to copyright.
Journal of Clinical Monitoring and Computing manuscript No.
(will be inserted by the editor)
A System for Portable Sleep Apnea Diagnosis Using
an Embedded Data Capturing Module
Hisham Alshaer ·Alexander Levchenko ·
T. Douglas Bradley ·Steven Pong ·
Wen-Hou Tseng ·and Geoff R. Fernie
Received: date / Accepted: date
This stage of the project was supported by MaRS Innovation. The project has been also
supported by the Ministry of Research and Innovation of Ontario, Ontario Centre of Excellence,
and Johnson and Johnson Inc. Toronto Rehabilitation Institute receives funding from the
Ontario Ministry of Health and Long-Term Care. Dr Alshaer is a recipient NSERC scholarship.
Hisham Alshaer
Institute of Biomaterial and Biomedical Engineering
University of Toronto, and
Sleep Research Laboratory & Technology Development Team
University Av, Toronto, ON, Canada, M5G 2A2
Tel.: +1 416 597-3422 ext 7959, Fax: +1 416 597-3027
E-mail: Hisham.Alshaer at uhn.ca
Alexander Levchenko
Technology Development Team
Toronto Rehabilitation Institute, University Health Network
550 University Av, Toronto, ON, Canada, M5G 2A2
E-mail: Alex.Levchenko at uhn.ca
T. Douglas Bradley
Toronto General Hospital, University Health Network
9N-943, 200 Elizabeth St., Toronto, ON, M5G 2C4, Canada
E-mail: douglas.bradley at utoronto.ca
Steven Pong
Technology Development Team
Toronto Rehabilitation Institute, University Health Network
550 University Av, Toronto, ON, Canada, M5G 2A2
E-mail: Steve.Pong at uhn.ca
Wen-Hou Tseng
Sleep Research Laboratory
Toronto Rehabilitation Institute, University Health Network
550 University Av, Toronto, ON, Canada, M5G 2A2
E-mail: wenhou.tseng at utoronto.ca
Geoff R. Fernie
Toronto Rehabilitation Institute, University Health Network
550 University Av, Toronto, ON, Canada, M5G 2A2
E-mail: Geoff.Fernie at uhn.ca
(Postprint)
2 Alshaer, Levchenko, et al
Abstract
Introduction: Sleep apnea (SA) is a very common disease with serious
health consequences, yet is very under-diagnosed, partially because of
the high cost and limited accessibility of in-laboratory polysomnogra-
phy (PSG). The purpose of this work is to introduce a newly developed
portable system for the diagnosis of SA at home that is both reliable and
easy to use.
Methods: The system includes personal devices for recording breath
sounds and airflow during sleep and diagnostic algorithms to process the
recorded data. The data capturing device consists of a wearable face frame
with an embedded electronic module featuring a unidirectional micro-
phone, a differential microphone preamplifier, a microcontroller with an
onboard differential analogue to digital converter, and a microSD memory
card. The device provides continuous data capturing for 8 hours. Upon
completion of the recording session, the memory card is returned to a
location for acoustic analysis. We recruited 49 subjects who used the de-
vice independently at home, after which each subject answered a usability
questionnaire. Random data samples were selected to measure the signal-
to-noise ratio (SNR) as a gauge of hardware functionality. A subset of
11 subjects used the device on 2 different nights and their results were
compared to examine diagnostic reproducibility. Independent of those,
system’s performance was evaluated against PSG in the lab environment
in 32 subject.
Results: The overall success rate of applying the device in un-attended
settings was 94% and the overall rating for ease-of-use was excellent. Sig-
nal examination showed excellent capturing of breath sounds with an av-
erage SNR of 31.7 dB. Nine of the 11 (82%) subjects had equivalent results
on both nights, which is consistent with reported inter-night variability.
An Embedded Module for Sleep Apnea Diagnosis 3
The system showed 96% correlation with simultaneously performed in-lab
PSG.
Conclusion: Our results suggest excellent usability and performance of
this system and provide a strong rationale to further improve it and test
its robustness in a larger study.
Keywords home diagnosis ·usability ·portable system ·embedded
module
1 Introduction
Sleep apnea (SA) is a breathing disorder characterized by repetitive ces-
sations of breathing during sleep for intervals of 10-90 seconds in length.
The frequency of these events ranges from 5 to 100 times/ hour depending
on the severity of the case. Obstructive sleep apnea, the most common
type SA, results from collapse of the upper airway either partially (hy-
popnea) or totally (apnea) during sleep. Central sleep apnea, on the other
hand, results from intermittent loss of central nervous system’s respira-
tory drive, which can also be partial of complete. These events alternate
with episodes of hyperventilation during which loud snoring occurs. These
events result in episodes of oxygen deprivation, and arousals from sleep
that cause sleep fragmentation. Subsequently, patients suffer from poor
sleep quality, daytime sleepiness, and poor cognitive performance. Repet-
itive apneas and intermittent hypoxia also elicit sympathetic nervous sys-
tem activation that cause repetitive surges in blood pressure at night and
increase the risk of developing daytime hypertension and atherosclerosis
independently of other risk factors [1–3]. Patients with SA are at three
to four fold greater risk of developing heart failure and stroke than sub-
jects without SA [4,5]. In the US, it has been estimated that SA-related
motor-vehicle collisions due to sleepiness in the year 2000 caused 1,400
4 Alshaer, Levchenko, et al
deaths and cost $15.9 billion [6]. SA also imposes a significant financial
burden on the health system; Canadian data show that patients with un-
treated SA consume twice as many health care resources for treatment
of cardio-vascular diseases than subjects without SA [7]. On the other
hand, we have demonstrated that treating SA lowers blood pressure in
hypertensive patients, improves cardiovascular function in patients with
heart failure and hastens recovery from stroke [8–11]. Therefore, diagnos-
ing and treating such patients could have a very substantial beneficial
medical and public health impact [12].
SA poses a significant challenge to anesthesiologists in surgical popu-
lations. It has been shown that SA is more prevalent in surgical patients
yet, most of which has not been diagnosed, which put them at higher
incidence of complications and deaths. This includes difficult intubation,
postoperative complications, increased intensive care unit admissions, and
greater duration of hospital stay [13–15]. Therefore, identification of SA
patients in the first place is of great value for anesthesiologists towards
the prevention of peri-operative complications.
Despite the high prevalence of SA, which affects approximately 7%
of adults [16], the majority of patients (87%) remain undiagnosed [17],
corresponding to approximately 15,000,000 patients in Canada and the
US alone. This is partly attributable to the lack of accessibility to ex-
pensive overnight monitoring in a sleep laboratory that is currently re-
quired for diagnosis [12]. Presently, the “gold standard” for diagnosing
SA is polysomnography (PSG). PSG requires patients to be monitored
overnight in a sleep laboratory using multi-channel recordings while they
are connected to sophisticated equipment with a technician in atten-
dance [18]. Because of the expense and specialized training required to
perform such testing, access to PSG is very limited in most jurisdictions.
An Embedded Module for Sleep Apnea Diagnosis 5
As a result, there is a need for alternative methods to diagnose SA
at home, in a more accessible and cost-effective manner. The challenge
in developing a portable monitoring device is finding the right balance
between simplicity on the one hand and accuracy on the other. Portable
SA diagnostic devices have been divided into 4 categories on the basis
of the number of transducers attached–Type 4 devices are those with
1 or 2 channels increasing up to Type 1, which is full in-lab PSG [19]
as illustrated in Figure 1. The currently available portable single chan-
nel devices suffer from low accuracy while multi-channel devices suffer
from high failure rates due to data loss in un-attended home settings. As
more channels are added to improve accuracy, failure rates increase to
as high as 33% [20]. Failure of the study in home settings is due to the
patient’s difficulty connecting the electrodes to the body and the inad-
vertent detachment of these electrodes during sleep. On the other hand,
devices that employ fewer channels have been found to be less accurate
than devices with more monitoring channels [21]. Extensive reviews have
concluded that there is no convincing evidence that any of the available
portable monitors could be used in unattended settings and still provide
reliable signals, which reduces their validity for use within the general
population [22, 23].
Fig. 1 Illustration of the reciprocal relation between the number of channels, success rate,
and accuracy
6 Alshaer, Levchenko, et al
In recent years, analysis of breath sounds recorded by a microphone,
has been an emerging tool for the diagnosis of SA. Breath sounds are rich
in information about breathing patterns, differentiating habitual snor-
ing from snoring in SA [24], the site from which snoring arises [25], and
patency of the upper airway [26]. We have shown that respiratory air-
flow has a characteristic acoustic signal signature [27]. Our recent results
show that acoustic analysis of breath sounds can accurately identify res-
piratory events, apneas and hypopneas, as compared with PSG overnight
subjects who underwent an overnight simultaneous PSG. Our next step
towards a fully portable system for the diagnosis of SA was to develop a
portable personal unit that patients can use at home for data collection
of breath sounds during sleep in unattended settings. Achieving this ob-
jective required the design of a unit that is easy to use and apply by a lay
person, yet sophisticated to obtain un-interrupted capturing of quality
breath sounds signal required for the bioacoustic analysis and identifica-
tion of apneas and hypopneas. The purpose of this paper, therefore, is
to describe the technical aspects of an innovative single channel portable
device and system for inexpensive and reliable diagnosis of sleep apnea in
the home. The initial evaluation of the system’s functionality, usability,
and diagnostic capability in the home environment will also be presented.
2 Materials and Methods
2.1 Development of the Diagnostic System
At Toronto Rehabilitation Institute, we have developed a system for res-
piratory monitoring and diagnosis of sleep disorders where data cap-
turing is done at home and data processing is centralized. The system
consists of 2 components. First, a self-contained, breath sounds captur-
ing module, with removable data storage media, embedded in a custom
An Embedded Module for Sleep Apnea Diagnosis 7
made face frame with no external wires to an external unit or power
supply—hereafter referred to as the ‘acoustic device’. This makes it self-
administrable and easy to use at home. Upon completion of a recording
session the memory card is returned to a diagnostic office for data extrac-
tion and analysis. Second, acoustic signal processing algorithms for data
analysis and diagnosis hosted on a server. The 2 components are further
explained below.
2.1.1 Self-Contained Acoustic Device With an Embedded Data-Capturing Module
The acoustic device consists of a data-capturing module embedded in an
open lightweight face frame to hold the microphone in optimum location.
The block diagram of the data-capturing module is shown in Figure 2. The
module includes an ATXmega128A3 microcontroller (MCU), a TS472
differential microphone preamplifier, a precision voltage reference source,
a MCP1640D step-up DC-DC converter with bypass option, a micorSD
memory card. The ATXmega MCU was selected for its high performance
12-bit ADC, featuring differential mode to utilize the advantages of the
TS472, 4-channel direct memory access (DMA) controller, and the ability
to operate down to 1.6 V that makes it particularly suitable for battery
powered applications.
Fig. 2 Block diagram of the embedded module for SA diagnosis. ADC: analog to digital
converter; DC: direct current; DMA: direct access memory; RAM: random access memory;
GPIO: general purpose input output.
8 Alshaer, Levchenko, et al
On power-on/reset the MCU initializes peripherals, puts the TS472
in a standby mode, activates the bypass feature of the MCP1640D and
enters power saving mode. While the converter is in bypass mode, the
MCU is essentially powered directly from batteries as the voltage range
is sufficient for operation in sleep mode thus minimizing overall power
consumption of the device. As the device is switched on, the MCU enables
the DC-DC converter, initializes the microSD memory card, configures
ADC and DMA channels and starts continuous sampling, at sampling
frequency of 16000 Hz, with the DMA controller configured in a double
buffer mode. In this mode one of the memory buffers is directly loaded
with the samples from ADC, while the previously stored buffer is being
recorded on a memory card. The device is powered by 2 alkaline AAA
batteries that provide 8 hours of continuous data sampling sufficient for
a typical overnight recording.
A unidirectional microphone with noise cancelling properties was se-
lected in order to minimize capture of external noise. Noise cancelling
properties of the microphone are achieved by highly directional dual-port
design. Since the microphone in the mask is specifically oriented towards
the mouth and nostrils, this directivity pattern helps to capture desired
sound and airflow signals and to filter out ambient noise. This feature
helped to achieve very low baseline noise levels and thus a better contrast
between breath sounds and no activity status. This in turn enhanced cap-
turing of breathing interruptions. The microphone is placed in the centre
of a specially designed funnel-shaped directional element, as illustrated
in Figure 3, that helps to capture not only audible breath sounds but also
airflow from both the mouth and nostrils.
Light emitting diodes (LEDs) provide a visual feedback to indicate the
status of the device. The device does not have any external wires, since
An Embedded Module for Sleep Apnea Diagnosis 9
the transducer and electronics are attached to the same frame, which
simplifies the application of the device by un-trained person. Users are
required to power the device by pressing the power button, wait for the
LED feedback signal, and wear the device by attaching a pair of head
straps. Patients have the option of terminating data collection upon com-
pletion of the session or otherwise leave it until it turns off automatically.
Fig. 3 Hardware prototype of the acoustic device showing the data capturing module embed-
ded in an open face frame. The microphone is held at position that allows maximum capture
of sound and nasal and oral airflow
2.1.2 Central Data Analysis
Upon completion of each session, data on the microSD card was uploaded
to a server. Apneas and hypopneas were detected from breath sounds us-
ing algorithms developed and validated in our laboratory earlier, which
10 Alshaer, Levchenko, et al
have shown excellent agreement (R=95%) with PSG [28,29]. Briefly, the
amplitude modulation envelopes of breath sound waveforms were adap-
tively normalized to a uniform baseline, which was then scanned for val-
leys below a certain threshold. Each valley is then examined against a set
of rules including its depth, width, and patterns of the falling and rising
edges, to be classified as an apnea or hypopnea, or to be discarded. The
frequency of apneas and hypopneas per hour of recording time, or the
apnea-hypopnea index (AHI) was found for each overnight session.
2.2 System Evaluation in the Home Environment
The acoustic device was evaluated in 49 subjects. Each subject was given
brief instructions on operating the device and was asked to wear it during
his/her usual sleep time in their home.
2.2.1 Evaluation of Functionality and Signal Quality in the Home Environment
Signal quality is a reflection of the functionality of the embedded module.
To assess the quality of breath sound signals captured in the home en-
vironment, respiratory signal to background noise ratio (SN Rbr eath) was
calculated. For this purpose, data were classified manually into breath-
ing and non-breathing components, similar to the method implemented
by Duckitt et al [30]. Here, 5-minute segments (L) were extracted from
the first, middle, and last part of randomly selected 10 home overnight
recordings. A total of 150 minutes was extracted. An experienced oper-
ator listened to each segment and manually annotated each sound unit
into one of 5 classes: inspiration, expiration, snoring, not-audible, other-
noise. The first and last 25% of each sound unit were discarded in order
to avoid boundary contamination from the adjacent sounds. The first 3
classes (inspiration, expiration, and snoring) originate from the physiolog-
ical breathing events. These were combined in each L segment and treated
An Embedded Module for Sleep Apnea Diagnosis 11
as the true breath sound signals. The ‘non-audible’ class included low am-
plitude non-intelligible portions such as brief pauses between breaths and
during apneas (Figure 4), which consist mainly from baseline noise in-
jected into the module in the absence of a true acoustic source, and thus
was treated as the background noise. The ‘other-noise’ class included ac-
cidental non-respiratory sounds, such as noise from bedsheets, somnolent
speech, and other external sounds, and was discarded because it does not
belong to the classes required in the analysis. To calculate the S N Rbreath,
first the normalized energy of breath sounds in L segments was calculated
as:
Ebreath =1
n
n
!
i=1
S2
breath(i) (1)
where breath ∈classes: {inspiration, expiration, snoring},S(i) are
data samples, and n is the number of all data samples of breath sounds
in L. The normalized energy of background noise (Enoise) was calculated
similarly. The SN Rbreath was calculated in decibels in each L, in which
Enoise was not zero, as in equation 2:
SN Rbr eath = 10 log10 "Ebreath
Enoise #(2)
The system’s overall S N Rbreath was then calculated as the average
SNR of all included L segments.
2.2.2 Evaluation of Usability in the Home Environment
Participants were given brief instructions on using the device and were
asked to take the device home to use it independently. Subsequent to the
overnight sessions, each participant answered 5 questions (Table 1) about
his/her experience, including the ease-of-use and comfort of the device.
Answers were given qualitatively as one of four options: very poor, poor,
12 Alshaer, Levchenko, et al
good, or excellent, each of which was quantified to a numerical value 1 to
4 respectively. In the case of inability to operate the device or sleep with
it on, a rating of 0 was given.
2.2.3 Evaluation of Diagnostic Capability and Reproducibility in the Home
Environment
One important limitation of PSG that needs to be addressed is the diffi-
culty to perform multiple sessions because of the high cost, limited acces-
sibility, and patients discomfort. On the other hand, it has been shown
that approximately one third of patients have remarkably variable AHI
between 2 nights, which might result in improper diagnosis if based on the
results of one night only. Therefore, a potential application of a portable
SA diagnostic device is performing multiple nights data collections and
inter-night comparison. To examine the feasibility and reproducibility of
multiple-night based diagnosis, a subset of eleven volunteers, who did not
undergo a sleep study for at least 1 year, agreed to use the acoustic de-
vice for 2 nights. The AHI for each night was calculated using acoustic
analysis of breath sounds (as described previously in 2.1.2.)
2.3 Agreement with PSG
To evaluate the accuracy of AHI derived from data captured by the em-
bedded module, the portable device was worn by 32 consecutive patients
referred to the Toronto Rehabilitation Institute Sleep Laboratory for sus-
picion of sleep apnea. None of this set of 32 subjects data were used in
developing the algorithm described in section 2.1.2. Subjects underwent
overnight polysomnography (PSG) using standard techniques and scor-
ing criteria [36,37]. Thoracoabdominal movements and tidal volume were
measured by respiratory inductance plethysmography (RIP) [38]. Airflow
was measured by nasal pressure cannulae [38] and arterial oxyhemoglobin
An Embedded Module for Sleep Apnea Diagnosis 13
saturation (SaO2) by oximetry. A sleep laboratory technician scored PSG
manually to find apneas and hypopneas. Apneas were scored as a drop in
sum of thoracoabdominal movement by ≥90% lasting ≥10 seconds [39].
Hypopneas were defined as a 50% to 90% reduction in thoracoabdominal
sum lasting ≥10 seconds as described previously [40].
BS were recorded simultaneously with PSG. At the end of the study, data
were transferred from the microSD memory card to a computer for acous-
tic analysis using the aforemention procedure described in 2.1.2. The AHI
derived from PSG (AHPP) and that from acoustic analysis (AHIA) were
quantified as the number of apneas and hypopneas per hour of recording.
AHIAwas compared against AHPPusing Pearson correlation coefficient
and diagnostic parameters were calculated.
The study protocol was approved by the Research Ethics Board of
Toronto Rehabilitation Institute and subjects provided written consent
prior to enrollment.
3 Results
3.1 Functionality and Signal Quality in the Home Environment
The overall breath sounds S N Rbreath in home sessions was 31.7 dB. Fig-
ure 4 displays representative samples of breath sounds captured by the
embedded module and the low amplitude background noise in the inter-
breath segment.
3.2 Usability in the Home Environment
Feedback from the 49 subjects was used to evaluated ease-of-use and
usability in the home environment. All the subjects (100%) indicated
being able to fully administer the acoustic device at home in the absence
of a trained person. Upon examination of the data, 3 subject (6%) were
14 Alshaer, Levchenko, et al
Fig. 4 A: single breathing cycle showing inspiration (time ∼
=0.5 to 2 seconds) and expiration
(time ∼
=2-3.5 seconds) separated by the vertical dashed line. The inter-breath interval is marked
to show a sample of ambient background noise and its relative amplitude. B: magnified wave-
form of part of the inspiration revealing a snoring episode characterized by quasi-periodicity,
resulting from tissue vibration. C: magnified waveform of part of expiration showing a turbulent
non-periodic pattern
found to have faulty data resulting from improper powering of the device,
which resulted in failure of data collection. Therefore, the overall success
rate in using the device was 94%. Answers to the usability questionnaire
are displayed in Table 1. The overall subjective rating of the ease-of-use
related questions was ‘excellent’ and for the comfort related questions was
‘good’.
Table 1 Usability questionnaire, the mean rating for the 49 subjects, and the qualitative
equivalent based on the mean rating (0-1: very poor; 1-2: poor; 2-3: good; 3-4: excellent)
Question Mean Rating Qualitative Equivalent
How easy could you operate the device? 3.2 Excellent
How easy you wore the device? 3.2 Excellent
How comfortable the mask was on your face
after wearing it?
2.9 Good
How quickly you fell asleep comparing with
other nights?
2.8 Good
How comfortable was your sleep comparing
with usual nights?
2.6 Good
An Embedded Module for Sleep Apnea Diagnosis 15
3.3 Diagnostic Capability and Reproducibility in the Home Environment
AHI resulting from acoustic analysis of data of the 11 subjects who used
the device for 2 nights are displayed in Figure 5. Their AHI ranged from
!0 to 80 apneas and hypopneas/hours. Nine out of 11 patients (82%)
showed equivalent results on both nights i.e. difference in AHI of less than
10.
Fig. 5 Comparison between the AHI obtained from the first and second nights in the (sorted
by severity.) Panel (A) shows subjects whose AHI varied by <10 on the 2 nights, while panel
(B) shows subjects whose AHI varied ≥10 on the 2 nights.
3.4 Agreement with PSG
We recruited 32 consecutive subjects who underwent full PSG while wear-
ing the acoustic device simultaneously. AHIAderived from data captured
by the acoustic device showed 96% correlation (R2=0.93%) with AHIP
obtained from manual scoring of PSG data as illustrated in Figure 6. Us-
ing a diagnostic cut-off of AHI ≥10, sensitivity was 100%, specificity was
85%, negative predictive value was 100%, positive predictive values was
16 Alshaer, Levchenko, et al
80%, and the overall accuracy was 91%. Using a diagnostic cut-off of AHI
≥15, sensitivity was 89%, specificity was 96%, negative predictive value
was 96%, positive predictive values was 89%, and the overall accuracy
was 94%.
Fig. 6 Agreement between AHIAand AHIP
4 Discussion
This work presents the development and initial evaluation of an inte-
grated system for accessible diagnosis of SA at home. The core of this
system is a self-contained device with a small embedded module to cap-
ture breath sounds. The acoustic device was well accepted by users as
shown by the very high success rates and complete subject compliance.
It also demonstrated excellent performance as revealed by the high SNR
and reproducibility of AHI in the home environment supported with re-
markable agreement with in-lab PSG
An Embedded Module for Sleep Apnea Diagnosis 17
The device captured high quality breath sound signals including snor-
ing sounds (characterized by a semi-periodic waveform) and other inspi-
ratory and expiratory sounds (characterized by turbulent non-periodic
waveforms.) The use of a unidirectional microphone, the low-noise elec-
tronic design, and the funneling structure in front of subjects’ mouths
were all deployed to enhance capture of breath sounds and improve signal
quality, thus achieving good SNR. SNR of breath sounds to ambient back-
ground noise was 31.7 dB. Other researchers in the field recommended a
minimum SNR of 10.5 [31] and 8 dB [32] to effectively distinguish breath
sounds and snoring from ambient noise. Thus, the SNR in uncontrolled
home environments with this design was clearly sufficient to perform the
desired acoustic analysis. Beside the low noise properties of the embedded
module, the low power design is also an essential factor for obtaining a
continuous 8-hour overnight recording required in this monitoring appli-
cation.
An essential, but quite overlooked aspect of a home SA monitor is
usability. The Portable Monitoring Task Force of the American Academy
of Sleep Medicine, for example, recommended developing easy to use
portable SA monitors that would reduce time required to attach sen-
sors, and simplify data collection and transfer [33]. The acoustic device
was designed to be single-channel and self-contained without external
wires in order to make it easy to use by an untrained person in unat-
tended settings, which is an important feature of a portable SA monitor.
In the current study, 100% of the subjects could administer the device
independently without the support of a trained assistant at home, with
94% success rate as revealed by objective examination of the data, which
demonstrates excellent usability of the device. It is noteworthy, however,
that comfort ratings were lower i.e. ‘good’ for comfort versus ‘excellent’
18 Alshaer, Levchenko, et al
for ease-of use (Table 1), which was attributed mostly to the experience
of wearing the acoustic device while sleeping for the first time. These
findings, in addition to other qualitative feedback can be used to improve
the ergonomics of future iterations of the device such as by reducing the
size of the embedded module.
AHI obtained from home sessions in 11 subjects were used to asses
performance and reproducibility in the home environment, all of which
did not have a recent PSG study. PSG studies have previously shown
that the majority of patients have reproducible AHI as measured by 2-
night difference in AHI <10 [34, 35]. Nevertheless, there may very well
be true inter-night variability, denoted by difference in AHI exceeding 10,
in approximately 18% [35] to 32% [34] of patients between the 2 nights.
In the current study, 9 out of 11 subjects demonstrated the equivalence
of AHI on both nights (2-night difference in AHI under 10). In 2 out of
11 (18.2%) subjects, greater variability in the AHI of ≥10 was observed,
which lies well within, indeed at the lower end of the previously reported
proportions. This is a sound indicator of the reproducibility of the results
in the home settings and that captured signals reflected the underlying
respiratory condition in the subjects who used the device while sleeping
at home.
To ensure the validity of AHI results obtained in the home environ-
ment, the system’s performance was assessed against in-lab PSG, the
current gold standard for diagnosing sleep apnea. The system showed ex-
cellent agreement (96%) with PSG in terms of AHI concordance in 32
subjects who wore the acoustic device simultaneously with PSG. Diag-
nostics accuracy was 91% and 94% for an AHI cut-off greater of equal 10
and 15 respectively . This is inline with our previously reported agreement
with PSG (95%) in our work describing the development the apneas and
An Embedded Module for Sleep Apnea Diagnosis 19
hypopneas detection algorithm [28]. In the current study, however, the
AHI was obtained from data collected using the newly designed acoustic
device and embedded module that has no external electrodes. This step of
validation of the acoustic device in a controlled environment demonstrates
its accuracy supports the AHI results obtained in the home studies.
Accordingly, we believe that this system has the potential to make
diagnosis of SA more accessible, particularly in remote areas where sleep
laboratories are not located. There is also the possibility to make multi-
ple recordings in subjects’ homes without the need for multiple visits to a
sleep laboratory, so that, for example, severity of SA could be monitored
over time in response to interventions such as weight loss. This device
makes it feasible to perform multiple studies and subsequently to explore
the causes of variability in AHI and lead to new treatment options that
recreate the circumstances that are associated with the lowest AHI scores.
In the proposed model, data collection is performed independently from
the data analysis. This setup gives the advantage of developing and im-
proving the diagnostic algorithms centrally without the need to re-design
the data acquisition module and disrupt the peripheral data collection.
This setup help in maintaining a simple structure of the portable unit
which will facilitate the use by untrained individuals and potentially re-
duce cost per test.
Future improvements of the embedded module might include migra-
tion to one of the rapidly growing ARM Cortex-M MCU families offering
more functional and cost effective platform for embedded applications
with their 16 bit onboard ADC, SD host controller, and high RAM-to-
flash ratio. Digital MEMS microphones with their small form factor, high
noise immunity, and low power consumption also offer a viable improve-
ment choice.
20 Alshaer, Levchenko, et al
5 Conclusion
We have described a novel SA diagnostic system using a portable bat-
tery operated acoustic data acquisition device as well as its functionality
and applicability to diagnose SA in the home environment. Its excellent
usability by lay people makes it suitable for use at home in un-attended
settings. The evaluation results point to the feasibility of making mul-
tiple home recordings in the same subject with capture of high quality
data. These data provide a strong rationale to perform home studies in a
larger population and compare it with PSG. This new portable home SA
monitor has the potential to improve accessibility and reduce the costs of
diagnosing SA.
Acknowledgements The authors are thankful to the contributions of Inas Kadri in devel-
oping the illustrations and to Azadeh Yadollahi for her editorial input. This work might result
in a commercializable device and patent applications have been filed.
References
1. Leung R S, Bradley T D, Sleep apnea and cardiovascular disease. Am J Respir Crit Care
Med 164, 2001: 2147–65.
2. Nieto F J, Young T B, Lind B K, Shahar E, Samet J M, Redline S, D’Agostino R B,
Newman A B, Lebowitz M D, Pickering T G, Association of sleep-disordered breathing,
sleep apnea, and hypertension in a large community-based study. sleep heart health study.
Jama 283, 2000: 1829–36.
3. Shahar E, Whitney C W, Redline S, Lee E T, Newman A B, Javier Nieto F, O’Connor
G T, Boland L L, Schwartz J E, Samet J M, Sleep-disordered breathing and cardiovascular
disease: cross-sectional results of the sleep heart health study. Am J Respir Crit Care Med
163, 2001: 19–25.
4. Bradley T D, Floras J S, Sleep apnea and heart failure: Part i: obstructive sleep apnea.
Circulation 107, 2003: 1671–8.
5. Arzt M, Young T, Finn L, Skatrud J B, Bradley T D, Association of sleep-disordered
breathing and the occurrence of stroke. Am J Respir Crit Care Med 172, 2005: 1447–51.
6. Sassani A, Findley L J, Kryger M, Goldlust E, George C, Davidson T M, Reducing motor-
vehicle collisions, costs, and fatalities by treating obstructive sleep apnea syndrome. Sleep
27, 2004: 453–8.
7. Otake K, Delaive K, Walld R, Manfreda J, Kryger M H, Cardiovascular medication use
in patients with undiagnosed obstructive sleep apnoea. Thorax 57, 2002: 417–22.
8. Kaneko Y, Floras J S, Usui K, Plante J, Tkacova R, Kubo T, Ando S, Bradley T D,
Cardiovascular effects of continuous positive airway pressure in patients with heart failure
and obstructive sleep apnea. N Engl J Med 348, 2003: 1233–41.
9. Usui K, Bradley T D, Spaak J, Ryan C M, Kubo T, Kaneko Y, Floras J S, Inhibition of
awake sympathetic nerve activity of heart failure patients with obstructive sleep apnea by
nocturnal continuous positive airway pressure. J Am Coll Cardiol 45, 2005: 2008–11.
10. Ryan C M, Usui K, Floras J S, Bradley T D, Effect of continuous positive airway pressure
on ventricular ectopy in heart failure patients with obstructive sleep apnoea. Thorax 60,
2005: 781–5.
An Embedded Module for Sleep Apnea Diagnosis 21
11. Ryan C M, Bayley M, Green R, Murray B J, Bradley T D, Influence of continuous positive
airway pressure on outcomes of rehabilitation in stroke patients with obstructive sleep
apnea. Stroke 42, 2011: 1062–7.
12. Leung R S, Bradley T D, Sleep apnea and cardiovascular disease. Am J Respir Crit Care
Med 164, 2001: 2147–65.
13. Chung F, Yegneswaran B, Liao P, Chung S A, Vairavanathan S, Islam S, Khajehdehi A,
Shapiro C M, STOP questionnaire: a tool to screen patients for obstructive sleep apnea.
Anesthesiology 108, 2008: 812–821.
14. Gupta R M, Parvizi J, Hanssen A D, Gay P C, Postoperative complications in patients
with obstructive sleep apnea syndrome undergoing hip or knee replacement: a case-control
study. Mayo Clin Proc 76, 2001: 897–905.
15. Gentil B, de Larminat J M, Boucherez C, Lienhart A, Difficult intubation and obstructive
sleep apnoea syndrome. Br J Anaesth 72, 1994: 368.
16. Young T, Palta M, Dempsey J, Skatrud J, Weber S, Badr S, The occurrence of sleep-
disordered breathing among middle-aged adults. N Engl J Med 328, 1993: 1230–5.
17. Young T, Evans L, Finn L, Palta M, Estimation of the clinically diagnosed proportion of
sleep apnea syndrome in middle-aged men and women. Sleep 20, 1997: 705–6.
18. Hult P, Fjallbrant T, Wranne B, Engdahl O, Ask P, An improved bioacoustic method for
monitoring of respiration. Technol Health Care 12, 2004: 323–32.
19. Ferber R, Millman R, Coppola M, Fleetham J, Murray C F, Iber C, McCall V, Nino-Murcia
G, Pressman M, Sanders M, et al, Portable recording in the assessment of obstructive sleep
apnea. asda standards of practice. Sleep 17, 1994: 378–92.
20. Golpe R, Jimenez A, Carpizo R, Home sleep studies in the assessment of sleep ap-
nea/hypopnea syndrome. Chest 122, 2002: 1156–61.
21. Liesching T N, Carlisle C, Marte A, Bonitati A, Millman R P, Evaluation of the accuracy of
snap technology sleep sonography in detecting obstructive sleep apnea in adults compared
to standard polysomnography. Chest 125, 2004: 886–91.
22. Flemons W W, Littner M R, Rowley J A, Gay P, Anderson W M, Hudgel D W, McEvoy
R D, Loube D I, Home diagnosis of sleep apnea: a systematic review of the literature. an
evidence review cosponsored by the american academy of sleep medicine, the american
college of chest physicians, and the american thoracic society. Chest 124, 2003: 1543–79.
23. Ahmed M, Patel N P, Rosen I, Portable monitors in the diagnosis of obstructive sleep
apnea. Chest 132, 2007: 1672–7.
24. Michael H, Andreas S, Thomas B, Beatrice H, Werner H, Holger K, Analysed snoring
sounds correlate to obstructive sleep disordered breathing. European Archives of Oto-
Rhino-Laryngology 265, 2008: 105–113.
25. Saunders N C, Tassone P, Wood G, Norris A, Harries M, Kotecha B, Is acoustic analysis of
snoring an alternative to sleep nasendoscopy? Clinical Otolaryngology and Allied Sciences
29, 2004: 242–6.
26. Alshaer H, Garcia M, Radfar M H, Fernie G R, Bradley T D, Detection of upper airway
narrowing via classification of LPC coefficients: Implications for obstructive sleep apnea
diagnosis, in IEEE International Conference on Acoustics, Speech and Signal Processing
(ICASSP), Prague, Czech Republic, 2011, pp. 681–684.
27. Alshaer H, Fernie G R, Bradley T D, Phase tracking of the breathing cycle in sleep-
ing subjects by frequency analysis of acoustic data. International Journal of Healthcare
Technology and Management 11, 2010: 163–175.
28. Alshaer H, Fernie G, Bradley T D, Development and validation of an algorithm for detec-
tion of apneas and hypopneas using overnight breath sound recordings. Am J Respir Crit
Care Med 183, 2011: A6317.
29. Alshaer H, Fernie G R, Sejdic E, Bradley T D, Adaptive segmentation and normalization of
breathing acoustic data of subjects with obstructive sleep apnea, in Science and Technology
for Humanity (TIC-STH), 2009 IEEE Toronto International Conference, pp. 279–284.
30. Duckitt W D, Tuomi S K, Niesler T R, Automatic detection, segmentation and assessment
of snoring from ambient acoustic data. Physiological Measurement 27, 2006: 1047–56.
31. Ng A K, Zhi Jie H, Tong San K, Effects of microphone position on snore signal quality
and patient comfort, in Signal Processing, 2008. ICSP 2008. 9th International Conference
on, pp. 2130–2133.
32. Yadollahi A, Moussavi Z, Automatic breath and snore sounds classification from tracheal
and ambient sounds recordings. Med Eng Phys 32, 2010: 985–90.
33. Collop N A, Anderson W M, Boehlecke B, Claman D, Goldberg R, Gottlieb D J, Hudgel
D, Sateia M, Schwab R, Clinical guidelines for the use of unattended portable monitors in
22 Alshaer, Levchenko, et al
the diagnosis of obstructive sleep apnea in adult patients. portable monitoring task force
of the american academy of sleep medicine. J Clin Sleep Med 3, 2007: 737–47.
34. Chediak A D, Acevedo-Crespo J C, Seiden D J, Kim H H, Kiel M H, Nightly variability
in the indices of sleep-disordered breathing in men being evaluated for impotence with
consecutive night polysomnograms. Sleep 19, 1996: 589–92.
35. Bliwise D L, Benkert R E, Ingham R H, Factors associated with nightly variability in
sleep-disordered breathing in the elderly. Chest 100, 1991: 973–6.
36. Eeg arousals: scoring rules and examples: a preliminary report from the sleep disorders
atlas task force of the american sleep disorders association. Sleep 15, 1992: 173–84.
37. Rechtschaffen A, Kales A, A Manual of Standardized Terminology, Techniques and Scor-
ing System for Sleep Stages of Human Subjects., Los Angeles: UCLA Brain Information
Service/Brain Research Institute, 1968.
38. Sleep-related breathing disorders in adults: recommendations for syndrome definition and
measurement techniques in clinical research. the report of an american academy of sleep
medicine task force. Sleep 22, 1999: 667–89.
39. Iber C, Ancoli-Israel S, Chesson A L, Quan S, The AASM Manual for the Scoring of Sleep
and Associated Events: Rules, Terminology, and Technical Specifications, Westchester, IL:
The American Academy of Sleep Medicine, 1st ed., 2007.
40. Redolfi S, Yumino D, Ruttanaumpawan P, Yau B, Su M C, Lam J, Bradley T D, Re-
lationship between overnight rostral fluid shift and obstructive sleep apnea in nonobese
men. Am J Respir Crit Care Med 179, 2009: 241–6.
