Author: Yeh-Liang Hsu, Ming-Chou Chen, Chih-Ming Cheng, Chang-Huei Wu (2005-11-03);
recommended: Yeh-Liang Hsu (2005-11-07).
Note: This paper is presented at International Conference on Systems, Man
and Cybernetics, Hawaii, USA
- October 10-12, 2005. This paper is published in Journal of Biomedical
Engineering - Applications, Basis & Communications, Vol. 17, No. 4, August,
2005, p.176-180.
Development of a portable device
for home monitoring of snoring
Abstract
Snoring analysis
is important for the diagnosis and treatment of sleep-related breathing
disorders (SRBD). Snoring has traditionally been assessed in clinical practice
from subjective accounts by the snorer and his/her partner. The use of
polysomnographic recording is a standard evaluation procedure for SRBD
patients. However, it is expensive and is not suitable for long term
monitoring. This paper describes the development of a portable microcontroller
based device for long-term, home monitoring of snoring. By analyzing the
temporal feature of the snoring sound, this device can output the total snoring
count, average number of snores per hour, and the number of intermittent
snoring. In our tests, the average success rate in identifying snores is over
85% in a lab environment and around 70% in a home.
Keywords: Snore, SRBD, home
monitoring, portable device.
1.
Introduction
Snoring is a
very common symptom and is an important sign of sleep-related breathing
disorder (SRBD), such as obstructive sleep apnea syndrome (OSAS). Snoring
analysis is important for the diagnosis and treatment of sleep disturbance.
Snoring has traditionally been assessed in clinical practice from subjective
accounts by the snorer and his/her partner. Various techniques have been
developed for the quantitative measurement of snoring.
Cohen presented
methods and algorithms for the quantitative and objective analysis of
acoustical pulmonary signals such as breathing and snoring sounds [1]. Lopez et
al. proposed a method to detect snores using an artificial neural network [2].
The frequency components of the pressure signals in the nasal mask of patients
were fed to the neural network. In their test on 5 patients, the overall
accuracy of correctly detected snores is 75.2%. Jane et al. also designed an
automatic detection algorithm for acoustic snoring signals based on a neural
network [3]. The input pattern of the neural network consists of 22 temporal
and spectral features of the sound segment, which distincts between snoring
sound and the remaining respiratory sounds. In their validation test, more than
500 snores taken randomly from the database of 30 subjects were analyzed. The
sensitivity of the algorithm was 82%, and the positive prediction value was
90%.
Hoffstein et al.
have pointed out that, the difficulty in measuring and quantifying snoring
using objective criteria is that snoring is the first and foremost of a
subjective perception by a listener [4]. They studied 25 patients, all had full
nocturnal polysomnography including the measurements of snoring, to compare the
objective snore count from the polysomnogram and the subjective snoring count
by two listeners during a 20-minute segment. In their study, the objective
count differed more than 25% in 7 out of 25 patents. In 7 out of 25 patients,
the difference in subjective snore counts perceived by both listeners was
larger than 25%.
The use of
polysomnographic recording is a standard evaluation procedure for SRBD
patients. However, it is expensive and is not suitable for long term
monitoring. Penzel et al. developed a digital recording device called MESAM IV,
to monitor oxygen saturation, heart rate, snoring, and body position in order
to screen subjects for OSAS [5, 6]. They also indicated that this inexpensive
implementation would allow the development of homecare systems for the analysis
and long term monitoring of snoring.
The automatic
scoring system of MESAM IV calculates oxygen desaturation index (ODI), heart
rate variation index (HVI) and intermittent snoring index (ISI) to obtain an
apnoea-hypopnoea index (AHI). The snoring sounds are recorded by means of a
laryngeal microphone. If the proportion of sounds between 50Hz and 800Hz
exceeds 50%, it is assumed that there is snoring. Intermittent snoring is
defined as intervals between two detected snores longer than 5 seconds and shorter
than 60 seconds. Evaluating snoring interval analysis on 68 patients with all
degrees of obstructive sleep apnea, Penzel et al. reported that correlation
between snoring interval analysis and polysomnographically scored apneas was
moderate (r=0.51) [6]. Following this development, a number of validation
studies on MESAM IV were presented [7, 8, 9, 10, 11]. In those studies, the
intermittent snoring index was found to have high sensitivity (92%-96%) but low
specificity (16%-27%).
This paper
describes the development of a portable device for home monitoring of snoring,
which performs detection and selection of the snores, while discarding any
other events that are present in the sound recording, as cough, voice, and
other artifacts. Besides counting snores, this device also outputs the count
for intermittent snoring. The measurement using this device with the subjective
assessment is also compared.
2.
Design of the snoring detector
Figure 1 shows
the structure of the snoring detector developed in this research. Snores are
low-frequency, regularly repeated sounds. In this device, a microphone records
the snoring sound. The microphone used is an omnidirectional electret condenser
type with a specified flat frequency response of 50-30,000Hz. High frequency
sounds are then removed by a first order low-pass filter (cut-off frequency at
200Hz).
%20Development%20of%20a%20portable%20device%20for%20home%20monitoring%20of%20snoring.files/image003.gif)
Figure 1. Structure of the snoring detector
The analog
signals are then sent to a microprocessor PIC16f877 to convert into digital signals. The sampling
rate for the A/D conversion is 2kHz, and the resulting signals are shown in
Figure 2(a). A series of smoothing procedures further process the signals:
(1) The reference voltage is set to zero, and all signals are set to
have positive voltage (Figure 2(b)).
(2) The profile of the signals is extracted by finding the maximum
voltage value in every 10 points (Figure 2(c)).
(3) The peaks are amplified by taking the sum of every 10 points (Figure
2(d)).
(4) The final profile is further smoothed by taking 10-point moving
average (Figure 2(e)).
Note that at the
end of this process in Figure 2(e), the sampling rate is reduced to 20Hz, and
the vertical axis of Figure 2(e) becomes Vsum (sum of voltage in Figure 2(d),
unit in mV).
%20Development%20of%20a%20portable%20device%20for%20home%20monitoring%20of%20snoring.files/image005.jpg)
Figure 2(a). Original signal taken by a microphone
and A/D converter at 2kHz
%20Development%20of%20a%20portable%20device%20for%20home%20monitoring%20of%20snoring.files/image007.jpg)
Figure 2(b). The reference voltage is set to zero,
and all signals are set to have positive voltage
%20Development%20of%20a%20portable%20device%20for%20home%20monitoring%20of%20snoring.files/image009.jpg)
Figure 2(c). Extraction of the profile by finding
the maximum voltage value in every 10 points (200Hz)
%20Development%20of%20a%20portable%20device%20for%20home%20monitoring%20of%20snoring.files/image011.jpg)
Figure 2(d) 10-points sum to amplify the peaks
(20Hz)
%20Development%20of%20a%20portable%20device%20for%20home%20monitoring%20of%20snoring.files/image013.jpg)
Figure 2(e). 10-point moving average to smooth the
curve
The snoring
signals in Figure 2(e) are further analyzed by PIC16f877 microprocessor to determine whether a snore
occurs. While most researches in snore analysis focus on the processing of the
frequency content of snores, we focus on the temporal features of snore. To
analyze the temporal features of the signals in Figure 2(e), the first problem
is to determine when a snore starts and when a snore ends.
In this study,
if the snoring signal in Figure 2(e) (Vsum) increases consecutively for 5 times
(about 0.25 second), the 5th signal is defined as the start point of a snore.
Similarly, if Vsum decreases consecutively for 5 times, the 5th signal is
defined as the end point of a snore. The time between a start point and the
adjacent end point is defined as Tduration, the duration of the snore; the time
between an end point and the next start point is defined as Tinterval, the
interval between snores.
We collected
5024 snoring samples from 5 patients who were diagnosed to have sleep-related
breathing disorders. Tduration of 99.0% of these snoring samples
ranged from 0.6 seconds to 1.8 seconds. Tinterval of 98.5% of these
snoring samples ranged from 1.4 seconds to 4.0 seconds. Moreover, in our snoring
samples, the maximum number of snores in one minute is 20, and the minimum
number of snores in one minute is 12. So we define that the total length of a
snore (Tduration + Tinterval) has to be from 2.8 (60/21)
seconds to 5.5 seconds (60/11). Therefore a piece of sound signal has to
satisfy the following 3 conditions repeatedly (at least 2 times) to be
qualified as a snore:
(1)
0.6
Tduration
1.8
(2)
1.4
Tinterval
4.0
(3)
2.8
(Tduration + Tinterval)
5.5
As mentioned
earlier, in MESAM IV developed by Penzel et al. [1990, 1991], intermittent
snoring was defined as intervals between two detected snores longer than 5
seconds and shorter than 60 seconds. This definition is also used in our
device, that is, a piece of sound signal is detected as intermittent snoring if
5.0
Tinterval
60.0. Figure 3 shows a piece of sample data in which some of
the sound signals were identified as “snores”, and some of the sound signals
were identified as “intermittent snoring”.
%20Development%20of%20a%20portable%20device%20for%20home%20monitoring%20of%20snoring.files/image017.gif)
Figure 3. Sample data
Figure 4 shows a
prototype of the snoring detector developed in this research. The total number
of snores and intermittent snores, as well as the average number of snores per
hour can be shown on the LCD display in the front panel, or stored in the multi
media card (MMC) for using in long term monitoring. The real-time clock chip
DS1302 is also integrated to provide the date and time of the system. Therefore
this device can also be used as a clock for home use. The user can switch from
“normal mode (as a clock)”, “sleep mode (snore detection)” and “report mode
(reading data)” from the knob and buttons on the front panel.
%20Development%20of%20a%20portable%20device%20for%20home%20monitoring%20of%20snoring.files/image019.gif)
Figure 4. The prototype of the snoring detector
3.
Evaluation of the snoring
detector
We did two tests
to compare the subjective perception of snoring by a listener and the
measurement by the snoring detector. In our first test, 4 10-minutes snoring
sounds were recorded and played in a lab environment. The speaker and the
snoring detector were placed in fixed positions. Each snoring sample was played
3 times. Table 1 shows the results of this test. The success rate in
identifying the snores ranges from 74% to 96%. The success rate for snore
sample C is lower because the volume of the snore sound is very low.
In our second
test, we actually placed the snoring detector at the home of 3 patients for
overnight monitoring. In the home environment, the snoring detector was placed
along the bedside the patient. The relative positions between the snoring
detector and the patient cannot be fixed, and there can be other noise
interferences during the detection. We also recorded the snoring sounds
overnight so that we could listen to the recording and compare the subjective
snoring count with the measurement by the snoring detector. Table 2 shows the
result. The success rate ranges from 66% to 79%, lower than that obtained in a
lab environment. The successful rate for identifying intermittent snores ranges
from 64% to 80%.
Table 1. Evaluation of the snore detector in a lab
environment
|
Snores
|
Intermittent
snores
|
Snores
counted by human
|
Number
of snores detected
|
Success
rate
|
Average
success rate
|
Intermittent
snores counted by human
|
Number
of intermittent snores detected
|
Snore
sample A
|
151
|
148
|
96%
|
96%
|
1
|
1
|
145
|
96%
|
145
|
96%
|
Snore
sample B
|
137
|
123
|
88%
|
87%
|
1
|
1
|
121
|
86%
|
124
|
88%
|
Snore
sample C
|
83
|
63
|
74%
|
74%
|
1
|
2
|
62
|
73%
|
64
|
76%
|
Snore
sample D
|
104
|
96
|
92%
|
91%
|
1
|
0
|
94
|
90%
|
84
|
90%
|
Table 2. Evaluation of the snoring detector in a
home environment
|
Snores
|
Intermittent
snores
|
Snores
counted by human
|
Number
of snores detected
|
Success
rate
|
Number
of snores per hour
|
Intermittent
snores counted by human
|
Number
of intermittent snores detected
|
Success
rate
|
Patient
1
|
1952
|
1292
|
66%
|
258.4/Hr
|
236
|
188
|
80%
|
Patient
2
|
2675
|
1833
|
69%
|
366.6/Hr
|
113
|
74
|
65%
|
Patient 3
|
1877
|
1485
|
79%
|
297.0/Hr
|
28
|
18
|
64%
|
4.
Conclusions
This paper
describes the development of a portable microcontroller-based device for
long-term, home monitoring of snoring. This device is intended to be a home
appliance and can be put along the bedside as a clock. By analyzing the
temporal feature of the snoring sounds, this device can output total snoring
count, average number of snores per hour, and number of intermittent snores.
This information indicates the severity of snoring and the likelihood of having
OSAS. In our tests, the average success rate of this device in identifying
snores is over 85% in a lab environment and around 70% in a home.
“What sound is a
snore” is a very subjective judgment, and snoring sound varies from person to
person. In the device presented here we focused on analyzing the interval of
snoring sound. In the early stage of this research, we tried to use information
extracted from frequency of the sound to help judging what are snoring sounds
but failed. However, we still believe that the average success rate of this
device can be improved if frequency information is properly utilized.
References
[1]
Cohen, A., “Signal processing
methods for upper airway and pulmonary dysfunction diagnosis,” IEEE Engineering
in Medicine and Biology Magazine, March 1990, p.72-75.
[2]
Lopez, F.J., Behbehani, K.,
Kamangar, F., “An artificial neural network based snore detector,” Annual
International Conference of the IEEE Engineering in Medicine and Biology -
Proceedings, v.16, n. pt 2, p.1107-1108, 1994.
[3]
Jane, R., Sola-Soler, J., Fiz,
J.A., Morera, J., “Automatic detection of snoring signals: validation with
simple snorers and OSAS patients,” Annual International Conference of the IEEE
Engineering in Medicine and Biology - Proceedings, v 4, p 3129-3131, 2000.
[4]
Hoffstein, V., Mateika, S.,
Nash, S., “Comparing perceptios and measurements of snoring,” Sleep, v.19,
n.10, p.783-789, 1996.
[5]
Penzel, T., Amend, G., Meinzer,
K., Peter, J. H., von Wichert, P., “MESAM – A heart-rate and snoring recorder
for detection of obstructive sleep-apnea,” Sleep, v.13, n.2, p.175-182, 1990.
[6]
Penzel, T., Althaus, W.,
Meinzer, K., Peter, J. H., von Wichert, P., “A device for ambulatory heart
rate, oxygen saturation and snoring recording,” Annual International Conference
of the IEEE Engineering in Medicine and Biology Society, v.13, n.4,
p.1616-1617, 1991.
[7]
Stoohs, R., Guilleminault, C.,
Chest, v.101, n.5, p.1221-1227, 1992.
[8]
Koziej, M., Cieslicki, J. K.,
Gorzelak, K., Sliwinski, P., Zielinski, J., “Hand-scoring of MESAM-4 recordings
is more accurate than automatic-analysis in screening for obstructive
sleep-apnea,” European Respiratory Journal, v.7, n.10, p.1771-1775, 1994.
[9]
Esnaola, S., Duran, J.,
InfanteRivard, C., Rubio, R., Fernandez, A., “Diagnostic accuracy of a portable
recording device (MESAM IV) in suspected obstructive sleep apnoea,” European
Respiratory Journal, v.9, n.12, p.2597-2605, 1996.
[10]
Verse, T., Pirsig, W.,
Junge-Hulsing, B., Kroker, B., Chest, v.117, n.6, p.1613-1618, 2000.
[11]
Cirignotta, F., Mondini, S.,
Gerardi, R., Mostacci, B., Sancisi, E., “Unreliability of automatic scoring of
MESAM 4 in assessing patients
with complicated obstructive sleep apnea syndrome,” Chest, v.199, n.5,
p.1387-1392, 2001.