Snoring Detector

Document Sample
Snoring Detector Powered By Docstoc
					                                             Snoring Detector

Introduction
Snoring is (usually) a nocturnal sound which, unbeknownst to the snorer, may disturb and cause misery
to the snorer’s family. It is an extremely common condition and it has been estimated that up to 50% of
the adult population snores (1, 2). A snore is a respiratory noise generated by turbulent air flowing
through an occluded airway during sleep. During waking breathing the airway is sufficiently open so
that flow is laminar, and therefore relatively quiet. Typically a snore is generated on the inspiratory
phase of the breathing cycle, although not exclusively. Turbulent air causes soft tissues of the
oropharynx to vibrate and cause the snoring sound. Dalmasso et al. describe the upper airway (the
section from the lips and nostrils to the vocal cords) as comprising of several cylindrical segments of
differing length and cross sectional area (3). Flow through these tubes will be turbulent or laminar
dependent on its Reynolds number. This approximation is only partial however as some of the upper
airway is collapsible, and thus does not meet the conditions of the flow formulae.

There are many reasons for the constriction of the airway and these include position of the person,
relaxation of the muscles in the throat, alcohol consumption, age, hormone levels, obesity and general
oropharangeal dysfunctions.

Snoring is not just a domestic problem disrupting the sleep of those around the snorer. It can also be
symptomatic of a more serious condition called Obstructive Sleep Apnoea (OSA).

Obstructive Sleep Apnoea (OSA)
In sufferers of OSA the upper airway becomes totally blocked during sleep causing blood oxygenat ion
levels to drop. The exact cause of OSA is unknown, but in most patients it is the soft palette and area at
the base of the tongue which collapse and block the airway. This area has no rigid structures, such as
bone or cartilage to prevent collapse. During waking breathing the muscles in this area are active and
keep the airway open, but during sleep they relax and, in the case of OSA, collapse completely (figure
1).




                             (a)                                             (b)

Figure 1. Diagrammatic view of (a) an open airway during sleep and (b) a blocked airway. The blockage shown is
caused by the soft tissue at the base of the tongue collapsing and preventing the flow of air.

During sleep, the OSA sufferer cycles through a series of events:

      The airway becomes blocked and the sufferer does not breathe,
      Blood oxygenation saturation (SaO 2) decreases, causing the heart to pump faster,
      The sleeper arouses to open their airway and breathe,
      The sufferer falls asleep again.

The sufferer needs only to wake for a few seconds to breathe and then fall asleep again, however
these apnoea events can occur, in severe sufferers, hundreds of times every night. OSA is a very
serious condition which untreated can lead to cardiovascular disease, hypertension and daytime
somnolence, which in turn can lead to other serious problems e.g. falling asleep whilst driving.
Diagnosing OSA
The gold standard in diagnosis of OSA is polysomnography. Originally developed by Dr. Nathaniel
Kleitman at the University of Chicago in the 1950s, polysomnography takes multiple physiological
measurements of the patient being studied. This combined data is used in the diagnosis of apnoea
events. The physiological measurements recorded can include:

       Electroencephalogram EEG (brain electrical activity)
       Electroculogram EOG (eye movement)
       Electromyogram EMG (jaw muscle movement)
       Leg muscle movement
       Airflow
       Respiratory effort (chest and abdominal excursion)
       Electrocardiogram ECG
       Oxygen saturation SaO2
       Audio and visual recording of nocturnal sounds and movements

EEG, EOG and EMG measurements can provide information about the stage of sleep of the patient,
and combined data can be used to diagnose particular sleep disorders. OSA is typified by periodic
breathing with reduced chest and abdominal excursion, decreases in SaO2, broken EEG patterns and
snoring sounds.

Lugaresi et al. proposed a four stage scale of snoring noises based on several measurements (noise of
snoring, endothoracic pressure, SaO 2) and is reproduced in table 1 (4).

                         Stage 0 (or preclinical)      Sporadic obstructive
                                                       apnoeas
                         Stage I (or initial)          Obstructive apnoeas
                                                       persisting during light (stage
                                                       1-2) and REM sleep
                         Stage II (or overt)           Obstructive apnoeas
                                                       persisting for the whole
                                                       length of sleep
                         Stage III (or complicated)    Alveolar hypoventilation
                                                       persisting during
                                                       wakefulness.
Table 1. Four stage scale of snoring noises proposed by Lugaresi et el.


Owing to the complexity of the measurements taken and the number of specialised sensors and
transducers used, polysomnography requires overnight observation in a sleep disorders unit. The
financial implications of this preclude it from being a widely available screening technique for the
general population. On a practical and cost effective level, screening the general population for
potential OSA must be a simple test that the patient can self administer in their own home. The results
from such a test can then be given to the clinician for analysis. If the results meet some predefined
criteria the patient can then be admitted for polysomnography to diagnose OSA. One such test that can
be used is frequency and power amplitude analysis of the snoring noise generated by the patient.

Treatment of OSA
It was between 1960 – 1980 that OSA was clearly defined as a clinical disease and the only clinical
treatment was surgical intervention (3). In 1981 a device was described that kept the airway open by
passing a continuous stream of air through the airway, which is known as Continuous Positive Airway
Pressure (CPAP) (5). This device is the method of choice for prevention of airway blockage for
sufferers of OSA.

Snoring Detector
One simple method to screen a person for potential OSA in their own home is the analysis of their
snore sounds. Several studies in the mid-1990s analysed the acoustic properties of snoring sounds (3,
6, 7). The basic premise behind these studies was to determine whether the snoring sound generated
was related to OSA status.

Figure 2 shows a functional block diagram for snoring                             Microphone
sound analysis. To acquire the signal a microphone is
used. This signal is then amplified and filtered to                                Amplifier
remove any unwanted interference. The choice of
                                                                                     Filter
filter is important so that none of the important signal
information is removed.                                                  Analogue to Digital Conversion
Next the filtered signal is digitised and imported into
the computer for processing and display. Once input                                Computer
to the computer, the time domain signal (time vs.
amplitude) can be displayed and measurements of the                          Time Domain Display
snoring cycle period, frequency and intensity can be
made. By applying a Fast Fourier Transform, the                             Fast Fourier Transform
signal can then also be displayed in the frequency
                                                                           Frequency Domain Display
domain, allowing yet further measurements and
analysis of the signal.
                                                                Figure 2. Block diagram for snoring detector

Microphone
The primary signal is acquired using a microphone. The microphone is a transducer which converts
audio signals to electrical signals. There are several different types of microphone available and all use
the same principal: a moveable diaphragm is displaced by the sound wave and the size of the
deflection is relative to the power of the sound. The movement of the diaphragm creates a voltage
change that is proportional to the power of the sound wave.
In a condenser or capacitor microphone, the diaphragm is one plate of a capacitor. As the sound wave
impinges on the diaphragm, the capacitor plate is deflected. The capacitance of the plates is inversely
                                                 A
proportional to their separation ( C   0  r     ) and they have a fixed amount of charge. Charge,
                                                 d
capacitance and voltage are related by the formula Q = C x V The charge and capacitance are fixed, so
as the capacitor plate moves, the voltage varies. A capacitor microphone requires a power supply.
In an electret microphone, the need for a power supply is removed as the material provides the
necessary charge. An electret is a dielectric material that is permanently electrically charged or
polarised. The electret forms the diaphragm and its distance from the plate causes a voltage to be
induced. Cheaper than capacitor microphones, the electret microphone is a simple and easy way to
measure a signal.
Both of these types of microphone can measure sounds in the range of 0 – 20 kHz and have a linear
response (voltage is proportional to the power of the sound). The typical range of snoring sounds is 0 –
5 kHz, with the vast majority of the power in the lower end of this range, so either of these types of
microphone would be suitable.

Amplification
The voltage created by the microphone is very small and needs to be amplified. The intensity of the
signal is increased equally at all frequencies. Once amplified it becomes easier to then process the
signal.
                                                                 Author                 Frequency Range
Filtering                                                                               of Snoring Sounds
                                                                 Beck et al.(6)         0 – 1500 Hz
Once amplified, the signal can then be filtered to remove        Fiz et al.(7)          0 – 1000 Hz
any unwanted parts. We are interested only in the snoring        Dalmasso et al.(3) 0 – 5500 Hz
sound of the patient, and so need to remove any other
sources of noise that have been recorded by the                  Table 2. Maximum snoring sound
microphone, such as sounds created by the movement of            frequencies used in snore analysis studies
the patient. The snoring sounds are generally low
frequency, so an appropriate filter to use would be a low pass filter. The value of the cut-off frequency
needs to be chosen carefully as it is important to keep the entire snoring sound signal, whilst trying to
remove the unwanted frequencies. Table 2 shows snoring frequency values used in three studies. The
accuracy of a filter determines how well it performs and is generally related to its order. The higher the
order of filter, the better defined the pass and stop frequencies are. This is known as the roll-off.

Analogue to Digital Conversion
To analyse the signal with a computer, the voltage must now be converted to a digital signal. This is
done using an Analogue to Digital Converter (ADC). The voltage is converted into a binary number and
the resolution of the conversion is a measure of how accurate the conversion is. Resolution is
measured in bits: an 8-bit converter has 28 = 256 available quantisation levels. The voltage range
divided by the number of quantisation levels give the resolution of the converter in volts, that is the
smallest voltage change that the converter can quantify. The higher the number of conversion bits, the
more sensitive the converter.
In every ADC there exists a quantisation error that arises from the nature of the conversion. By
definition the ADC takes a continuous data set and converts it into a discrete data set. There are a finite
number of quantisation levels available, so voltages that are in a certain range in the continuous data
set will be converted to a single value in the discrete set. The higher the number of bits, the smaller the
range and the more accurate the conversion. The size of this quantisation error is between zero and ½
of the value of the least significant bit (LSB). The LSB is the smallest unit in the digital data – for an 8-
bit converter the LSB is 1/256.
The signal needs to be sampled adequately frequently so that no information is lost. The minimum
frequency at which sampling needs to be performed at is determined by the Nyquist limit. The Nyquist
theorem states that in order for no information to be lost, a signal needs to be sampled at least twice as
frequently as the highest frequency component of that signal. If the sampling frequency is below the
Nyquist limit then aliasing will occur, where the signal at the extremes is corrupted. High frequency
signals are not sampled frequently enough and so they appear as low frequency signals in the sampled
signal.

Computer
Once converted the digital signal is imported into the computer. This gives the ability to perform many
operations on the signal. For portability this could be a laptop computer.



Time Domain Display

Figure 3 is an example of a time domain display
and shows the signal obtained from a single snore.
Viewing the signal in the time domain gives
important information like the average time taken
for a snore, the frequency of snores and the
structure of the snore in the time domain. The y
axis is the intensity of the sound. From this display
average snores can be determined and used for
analysis.

                                               Figure 3. Time domain display of a snore sound. Taken from (7)

Fast Fourier Transform
A Fourier Transform is a mathematical operation that changes a signal from the time domain to the
frequency domain.

Every signal can be thought of as being composed of a series of sine waves with different frequencies
and amplitudes. The summation of these individual sine functions is called a Fourier series. For a
frequency limited signal, such as the low pass filtered snore signal, there is a maximum frequency and
so the series is finite. For an unlimited series the summation becomes an integral.

A Fast Fourier Transform (FFT) is an algorithm that is used to quickly compute a Fourier Transform on
a signal – to change it from the time domain to the frequency domain.
Frequency Domain Display
Once the signal is in the frequency domain, certain characteristics can be seen, such as if there are any
harmonics. In the study of snore sounds, it has been determined that OSA sufferers have very low
frequency snores that do not contain any noticeable harmonics. In contrast to this, people who snore
but do not suffer from OSA have snores that have a fundamental frequency and several harmonics.
This method is has been used and verified in several studies. Figure 4 is an extract from Fiz et al. In
4(a) clear harmonics can be seen in the simple snorer’s frequency domain snoring signal, whilst 4 (b)
shows the signal from an OSA sufferer. The y axis is the intensity of the sound, in no particular units.




                         (a)                                                       (b)

Figure 4. Frequency domain spectra of (a) a normal snorer and (b) an OSA sufferer. The normal snorer has a
clear fundamental frequency and harmonics. The OSA sufferer has lots of energy in the low frequencies, but no
harmonics. Taken from (7).

Conclusion
The method of OSA screening described above works well as an initial test, but without full
polysomnography it would be difficult to definitively diagnose OSA. The advantage that a snoring
detector has over polysomnography is its ease of use and availability. The method described uses a
computer to directly monitor the snoring sounds produced. If a computer was not available to
immediately analyse the signals, they could be recorded onto a digital tape and then analysed at a l ater
time. Further investigation could be made into analysis of averaged snores, and whether information
contained in the mean of a number of snores gave better information than that of a single snore.

As useful as this described method is at screening for OSA, it is not however in common use. We have
contacted a sleep expert from The Manchester Royal Infirmary to ask the reasons why it is not used
(Appendix A). Whilst not giving a definitive answer to this question, he gave an alternative method that
is in common use for OSA screening. The patient monitors blood oxygenation levels throughout the
night by means of a simple probe connected to the finger or earlobe. The probe shines light at two
frequencies and records the amount reflected. Oxygenated haemoglobin reflects different amounts of
light to unoxygenated. This information is then used to diagnose OSA.

References
1.     Lugaresi, E., Cirignotta, F., Coccagna, G., and Piana, C. Some epidemiological data on snoring and
       cardiocirculatory disturbances. Sleep, 3: 221-224, 1980.
2.     Nort on, P. G. and Dunn, E. V. Snoring as a risk factor for disease: an epidemiological survey. Br Med J
       (Clin Res Ed), 291: 630-632, 1985.
3.     Dalmasso, F. and Prota, R. Snoring: analysis, measurement, clinical im plications and applications. Eur
       Respir J, 9: 146-159, 1996.
4.     Lugaresi, E., Mondini, S., Zucconi, M., Montagna, P., and Cirignotta, F. Staging of heavy snorers'
       disease. A proposal. Bull Eur Physiopathol Respir, 19: 590-594, 1983.
5.     Sullivan, C. E., Issa, F. G., Berthon-Jones, M., and E ves, L. Reversal of obstructive sleep apnoea by
       continuous positive airway pressure applied through the nares. Lancet, 1: 862-865, 1981.
6.     Beck, R., Odeh, M., Oliven, A., and Gavriely, N. The acoustic properties of snores . Eur Respir J, 8: 2120-
       2128, 1995.
7.     Fiz, J. A., Abad, J., Jane, R., Riera, M., Mananas, M. A., Caminal, P., Rodenstein, D., and Morera, J.
       Acoustic analysis of snoring sound in patients with simple snoring and obstructive sleep apnoea. Eur
       Respir J, 9: 2365-2370, 1996.
Appendix A


Dear James and Chris,

I would not profess to be an expert on this topic, although we do screen people here for OSA. The paper that you
enclosed has included a very small number of subjects which would not be enough to make a definitive
conclusion about the value of this approach.

We use overnight oximetry as a screening tool. This is almost as easy as the above since the patient just takes
the oximeter home and sleeps with the probe on their finger or earlobe – I have done it myself so I can vouch for
its simplicity. I guess that the advantage is that this approach has been validated against full polysomnography in
a much bigger subject population.

I hope that this is helpful.

Yours sincerely

Mark Woodhead

         -----Original Message-----
         From: James Cullen [mailto:James.Cullen-2@postgrad.manchester.ac.uk]
         Sent: 23 November 2005 15:38
         To: Woodhead Mark (RW3) CM&MC Manchester
         Cc: Christopher Boylan
         Subject: Snore Sound Analysis for help in OSA Diagnosis

         Dear Dr Woodhead,

         We are postgraduate students studying for an MSc in Medical Physics at the University
         of Manchester. As part of one of our modules we have been tasked with writing a short
         report on "snoring detection". From our study of the literature, we have found that a
         frequency analysis of the snoring sounds can reveal whether the snorer is
         suffering from Obstructive Sleep Apnoea (OSA). Seemingly, the snorer could record
         their nocturnal sounds at home using a commercially available microphone and digital
         recorder, and send the recording to a professional for analysis. If this analysis shows
         the snorer to be positive for OSA, they could then attend the Respiratory
         Department/Sleep Disorder Unit for polysomnography and further investigation.
         Importantly, this sound analysis would not form the basis of a diagnosis, but would be
         used as an indicator of possible OSA.

         Through our literature searches however, we cannot find why this simple and cheap
         'screening' for OSA is not currently used. We are able to guess possible reasons, but
         would like to have your expert opinion on the subject. Approaching the subject from a
         physics point of view, we are more interested in the measurement of the signal, its
         analysis and interpretations of the results, rather than clinical effects.

         For your information I have attached a paper that illustrates the snoring sound analysis.

         Thanking you in advance,

         James Cullen and Chris Boylan