BSTRUCTIVE sleep apnea syndrome is associated with

persistent snoring and upper airway obstruction, which

causes sleep disruption, changes in arterial oxygen saturation,

and daytime drowsiness. It is tough to define because it can

range from regular breathing to snoring, both of which have

serious consequences. For clinical reasons, apnea episodes of

more than 15 per hour are deemed abnormal. Each patient is

unique[1].

This article describes a reliable method for

automatically and intelligently identifying sleep apnea and

monitoring it. The procedure includes an automated

determination of OSA based on the sound signal produced

by breathing as well as a study of

the cardio-respiratory signals for better outcomes. The

recommended method sets itself apart by combining these

two factors, and due to the innovative way the respiratory

signal was processed which help us to get a very accurate

detection of apnea.

According to modern research done in the United States,

apnea incidence of 15 and higher was found in 9% of men and

4% of women in the workforce. Symptoms of sleep-disordered

breathing were reported by 5% of men and 2% of women who

were questioned [2]. These statistics show that sleep apnea is

indeed a significant public health.

The Home Sleep Apnea Test(HSAT) and other programs

rely on just a few or a single measure that is gradually being

created [3, 4]. There are a few high-quality HSAT devices on

the market right now, but none of them is based on a single

sensor. Several studies on single-channel OSA detection have

yielded encouraging results. As a result, signal analysis from

sensors that could be used to measure sleep apnea is a growing

subject of study. Nevertheless, most single- or few-sensor

OSA monitoring systems and suggested algorithms are

inadequate for real-time apnea detection. Effective OSA

treatment, on the other hand, is crucial since it recovers the

patient's state and other consequences[5, 6].

A disorder known as sleep apnea happens when an

individual stops breathing during sleep. It may be divided into

three types: central, obstructive, and complex. The most

common type is obstructive sleep apnea (OSA). According to

multiple Trusted Sources research, OSA affects anywhere

from 4% to 50% of the world's population. The prevalence of

sleep apnea within every study is determined by the

Identification of Apnea Based on Voice Activity Detection (VAD)

YOUNES EL OUAHABI1, KAOUTAR BAGGAR3, BENAYAD NSIRI2,3, MY HACHEM EL YOUSFI

ALAOUI2,3, ABDELMAJID SOULAYMANI1, ABDELRHANI MOKHTARI1, BRAHIM BENAJI3

1Laboratory Health and Biology, Faculty of Sciences, Ibn Tofail University, Kenitra, MOROCCO

2Research Center STIS, M2CS, National Graduate School of Arts and Crafts of Rabat, Mohammed V

University Rabat, MOROCCO

3Groupe of Biomedical Engineering and Pharmaceuticals Sciences - National Graduate School of Arts and

Crafts (ENSAM)-Mohammed V University Rabat, MOROCCO

Abstract: We identify obstructive sleep apnea as the most common respiratory issue associated with sleep.

Frequent breathing disruptions characterize sleep apnea during sleep due to an obstruction in the upper airway.

This illness, if left untreated, can lead to significant health problems. This article outlines a sound approach for

detecting sleep apnea and tracking it in an automated and intelligent manner. The method entails an automated

identification of OSA based on the sound signal during breathing and a cardio-respiratory signals analysis for

more efficient results. The suggested approach is put to the test under a variety of scenarios to verify its efficacy

and dependability. The benefits and drawbacks of the suggested algorithm are mentioned further down.

Keywords: Obstructive Sleep Apnea, Breathing disorder, cardio-respiratory automatic detection, OSA.

Received: May 17, 2021. Revised: July 25, 2022. Accepted: August 19, 2022. Published: September 22, 2022.

1. Introduction

1.1 Diagnosis of Obstructive Sleep Apnea

1.2 Types of Sleep Apnea

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

213

Volume 19, 2022

researchers' criteria, as well as the participants' age, gender,

and body weight, as well as any underlying health

concerns[7].

Males have a frequency of 22%, while girls had a

prevalence of 17%, according to a 2015 review of 11

studies[8]. Understanding the various types of sleep apnea can

aid people in determining what's causing their problems and

obtaining the support they require[9]. Oxidative stress has

been linked to sleep apnea, which has been linked to an

increased risk of diseases such as diabetes, hypertension, heart

problems, and strokes.

The most common type of sleep apnea is obstructive sleep

apnea (OSA). It occurs when the tongue and airway are both

functionally blocked. Breathing becomes extremely difficult

but not improbable, when the tongue rubs against the vocal

cords while sleeping, and the top of the throat and mouth then

brush against the larynx[10].

OSA can induce snoring due to the vibrating of the tongue

and soft palate. This can make you feel as if you can't breathe

when you wake up. The lungs work properly, and the system

attempts to inhale, but Apnea makes it hard to get enough air

through the upper airway[11].

OSA becomes more prevalent as people grow older, and it's

much more common with men, overweight people, pregnant

women, and those who lie flat on their backs. The following

are some of the indications and symptoms:

• Waking up at night or feeling extremely wary when

awake.

• Waking up with a worrying sense.

• Gasping or struggling to breathe in the middle of the

night.

• Frequent headaches.

• The feeling of having a dry mouth when you wake up.

• At school or work, you may be perplexed or unable to

concentrate.

Sleep apnea sufferers are treated in a variety of methods,

depending on their symptoms or where they reside. Significant

efforts are made in developed nations to identify and treat

patients who experience sleep apnea. According to current

studies, most instances of obstructive apnea are undiagnosed

and mistreated. OSA is rarely detected in resource-constrained

contexts, and diagnostic, and treatment options are either

lacking or insufficient because of the high and societal

repercussions, do to sum up sleep apnea is connected with a

high social and financial cost.

In 2015, the cost of identifying various types of sleep apnea

was predicted to reach $12.4 billion in the United States. The

exact global cost of identifying and treating OSA has still not

been determined since a further study on the prevalence rate is

required first[12][13].

Fig. 1: According on the American Academy of Sleep Medicine’s

2012 standards, the top ten nations with the largest estimated number

of patients suffering from obstructive sleep apnea

Sleep apnea is linked to poor healthcare outcomes, and

treating it improves the sleep-related life quality while

reducing adverse clinical implications[14]. As a result,

focusing on the effective treatment of OSA might be one

option for minimizing linked healthcare costs and unfavorable

symptoms of the disease, such as tiredness and cognitive

impairment.

For detecting sleep apnea, cardiorespiratory

polysomnography (PSG) is the best model[15]. While an

overnight is a must in a hospital, various respiratory factors

are gathered and examined by expert professionals.

Researchers are exploring alternate analytic

approaches such as home PSG and residential sleep apnea

diagnostics home sleep apnea testing (HSAT), which

employ monitoring systems with multiple sensors. There is

other research about detecting sleep apnea using EEG[16],

ECG[17], or EMG[18] signals or even by using video

recording devices[19]. Even though these studies have

lowered the cost of detecting sleep apnea, the

inconvenience is still present. Several other techniques use

breathing waves as the indicator to identify OSA have

always been suggested[20][21]. Our work focuses on

respiratory signals that can be simply recorded at home,

thus minimizing the cost and inconvenience without

affecting the system rate performance.

Those strategies need a whole nights and it is way too

expensive. Moreover, it has been noticed that the extensive

recording device significantly impacts sleep quality,

potentially distorting the results[22]. Several less complicated

but equally reliable techniques have been developed on this

foundation, especially for ambulatory and scanning

applications. The key distinctions are the type and quantity of

1.3 Symptoms of Obstructive Sleep Apnea

1.4 Obstructive Sleep Apnea in the World

1.5 Related Work

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

214

Volume 19, 2022

physiological signals gathered and automatically analyzed.

Polysomnography is a type of sleep study that is commonly

used. After the audio recording, complicated assessment

procedures are required to provide the information needed for

diagnosing sleep apnea. A typical way of calculating envelope

curves is to use breath sounds captured at the neck. Other

studies identified distinct patterns in tracheal audio spectrums

that distinguished apnea and non-apnea phases.

Fig. 2: Numeric breathing signals

For a vast number of people of varied ages and gender in

this study, we present a unique apnea detection approach

based on breathing signals. The goal of this study is to

employ a variety of algorithms. Unlike previous research, the

proposed method uses a threshold technique to remove

unnecessary segments while keeping the ones that would be

useful to detect apnea.

We examined the effectiveness of the recommended model

using a range of filtering techniques, allowing us to evaluate

each one's accuracy with apnea diagnosis.

This study used a database created by Dr. Thomas Penzel of

Phillips-University in Marburg, Germany, and made available

through the PhysioNet website[23].The Apnea-ECG

Database[24] has 70 records separated into 35 sets, with

recordings ranging from 7h-10h each.

Table 1: Information about the participant’s anthropoids

Description

No. of Subjects

Dataset Size

Age (year)

Male Patients

Female Patients

Body Mass (Kg)

Sleep Time (minutes)

Height

70 person

Size : 580.6MB

(27 – 58) ans

55 patients

15 patients

53 – 121

430 – 585

167 - 183

Four signals (a01 through a04, b01, and c01 through c03)

accompany recordings (a01 through a04, b01, and c01 through

c03) (Resp C and Resp A, chest and abdominal respiratory

effort signals obtained using inductance plethysmography;

Resp N, oronasal airflow measured using nasal thermistors;

and SpO2, oxygen saturation). We used breathing recordings

to support our main goal of identifying apnea using breath

analysis; an overview of the database is seen at the first table.

The process in general consists of using the numeric

form of the recorded respiratory signal and, through a

specific algorithm (explained below), we will be able to

classify the signals into apnea and non-apnea signals and

thus determine whether each patient has apnea or not.

The algorithm used in this study is divided into three major

parts: data pre-processing, reduced detection, and

classification. (for the moment, we ignored potential hypopnea

episodes).

1- Data pre-processing:

In order to create a "clear" and "clean" dataset for data

analysis, raw data from data extraction must first go

through a sequence of steps known as data pre-processing.

To enable accurate statistical evaluation, pre-processing

seeks to evaluate and enhance the quality of the data.[25]

In our case we will need data pre-processing will help us

remove any heart rate sound or background noise from

the original audio files, leaving just breathing tones in the

data.

The first part of the algorithm will follow the steps bellow:

a) Removing noises and heart sounds then producing a

signal with only breath sounds.

b) Using a FIR bandpass filter (200-2000) Hz These criteria

were made based on actual findings and relevant research that

were identified in the literature[26][27][28].

c) Eliminating background noise for the weak breath sounds

using a specific filtering method called subtraction[29].

2- Reduced detection

Finding drops and changes in the mean level of a time

series or signal is the concept of reduced detection,

1.6 Proposed Work

2. Materials

2.1 Database

2.2 Proposed Method

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

215

Volume 19, 2022

sometimes referred to as drop detection, in statistics and

signal processing. Generally, it is regarded as a specific

case of the statistical technique called change detection or

shift point detection. This technique will be necessary in

our system for identifying drops in breath amplitude

before detecting apneic segments[30].

The second part of the algorithm will follow the steps bellow:

a) Calculating the average intensity of the audio signal after

pre-processing and extracting the envelope curve E1 based on

specific method cited on those researches[31][32].

b) Removing outliers due to snoring with an adaptive

threshold (the standard deviation here is set at 30s).

c) interpolating the local maximum of a single respiratory

cycle into the first E1 envelope.

The E2 envelope is a smooth connection of the maximum

points of E1.

d) Signal segments in the E2 envelope (a connection of the

maximum points of E1) below the adaptive threshold are

identified as having decreasing respiratory amplitude.

e) The adjacent segments are extracted as probable apneic

segments.

3- Classification

The last part of the algorithm is the classification as the

name indicates. this section is used to classify the segments

detected in the previous section (reduced detection) and

then to classify them. the purpose of this section is to make

a detailed examination of the previously extracted segment

in order to distinguish apnea events from non-apnea

events.

The third part of the algorithm will follow the steps bellow:

a) The sound clip is now split to small duration events, the

E3 should is then be defined.

b) Using a low-pass filter of 2Hz.

Then all segments in E3 above a threshold that was

calculated are labeled as sound episodes.

c) all the episodes that have an activity above a defined

threshold are classified as motion noise.

d) Appling a threshold operation to the envelope curve E2.

The segments below this threshold are the only one who are

classified as apneic segments. If this segment is short, then

there is no apnea but if not the time is divided into apneic

segment and non-apneic segment.

e) Calculating the reference level for normal breathing.

f) If there is a segment with 10s and above without

breathing is to be an apnea phase.

We generate a variable threshold value and apply a low-

pass filter with a cutoff frequency of 2 Hz to the envelope E3.

All sessions in E3 that are higher than this threshold are

categorized as sound events. The noise brought on by motion

artifacts is identified using the motion signal that was

extracted from the motion measurement unit data. Events that

have activity levels over the established threshold in this case

are classified as motion artifact noise.

Then, using the decreased detection, we apply once again to

the E2 long-term envelope curve. All segments are now

classified as reference segments if they are above this

threshold and apnea segments if they are below it. Compared

to the mean value, the reference value is therefore more

resistant to unintentional outliers brought on by snoring.

The flow chart in Figure 7 provides more details on the

proposed system.

The two most common types of auditory signals are pulse

rate and apnea detection. It is worth noting that probable

hypopnea activities are currently overlooked when diagnosing

apnea occurrences. The apnea detection approach begins with

pre-processing, Reduced detection, and finally, classification.

In a basic flow chart, Figure 4 displays the framework's

essential processes.

Pre-processing removes any heart rate or background noise

from the original audio files, leaving just breathing tones in

the data. A FIR bandpass filter [200-2000] Hz is employed in

this example. These settings were selected based on published

research [27,28]. A filtration process known as spectral

reduction is used to reduce ambient noise that may interfere

with recognizing breathing patterns[29]. This approach works

by eliminating noise dents from the spectrogram's underlying

signal.

The system's second component is Reduced detection,

crucial for identifying probable apnea episodes. To aid in

further explanation, Figure 5 depicts this section's most

significant parts of the algorithms.

Fig. 4: Recognizing of probable apnea phase

3. Methodology

3.1 Apnea Detection

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

216

Volume 19, 2022

Fig. 3: The flow diagram highlights each key stage in the proposed

apnea diagnosis system

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

217

Volume 19, 2022

The two most common types of auditory signals are pulse

rate and apnea detection. It is worth noting that probable

hypopnea activities are currently overlooked when diagnosing

apnea occurrences. The apnea detection approach begins with

pre-processing, Reduced detection, and finally, classification.

In a basic flow chart, Figure 4 displays the framework's

essential processes.

Pre-processing removes any heart rate or background noise

from the original audio files, leaving just breathing tones in

the data. A FIR bandpass filter [200-2000] Hz is employed in

this example. These settings were selected based on published

research [27,28]. A filtration process known as spectral

reduction is used to reduce ambient noise that may interfere

with recognizing breathing patterns[29]. This approach works

by eliminating noise dents from the spectrogram's underlying

signal.

The system's second component is Reduced detection,

crucial for identifying probable apnea episodes. To aid in

further explanation, Figure 5 depicts this section's most

significant parts of the algorithms.

Fig. 4: Recognizing of probable apnea phase

To identify the early stages of sleep apnea. The first and

second envelope curves created during breathing amplitude

Reduced detection are blue (E1) and red (E2), respectively.

All the deleted examples concerning E1 are then induced by

noise are depicted by a dotted portions of the blue curve. The

blue zone on either side reflects the observed decrease and its

neighboring segments, resulting in a potentially retrievable

apnea segment (PAS).

First, most pre-processed audio inputs are detected over

short-term frames to construct an envelope curve E1

representing each breathing cycle. Compared to regular

breathing, snoring creates disproportionally huge outliers that

do not match the airflow volume. As a result, using a

thresholding technique for the lengthy frames' typical

fluctuation, all the anomalies in the first envelope are

removed. In Figure 5, the resultant E1 is shown in blue, while

the signal before outliers were removed is shown in the blue

line curve.

Fig. 5: Recognizing of probable apnea phase

We can observe the event categorization inside a single

possible apnea segment(PAS) in this graphic. The envelope

contour used by Eq 1 to identify sound occurrences is

represented by the red contour (E3). The lengths of the gray

surface, which identify the recorded sound occurrences,

signify the intensity of the selected features. The feature rate

of the audio signals in reference segment(RS) establishes a

cutoff to distinguish between respiratory and non-respiratory

episodes in the apnea segment.

In this illustration, all apnea segment(AS) occurrences come

up short throughout this cutoff and are classified as non-

respiratory. The region of the following identified apnea is

thus shown by the green curves.

E2, a second envelope that detects long-term respiratory

amplitude changes, is created as part of the Reduced detection

technique. The Nonlinear Cubic Hermite interpolation method

converts the local maxima of the (truncated) starting E1 into a

curve (see a red curve in Figure 5), Various research have

looked at the connection between air circulation and breathing

noises [33-35]. Within an adaptive threshold, breathing

amplitude decreases are detected in all signal parts of the

second envelope. Finally, these decreases and the segments

immediately around them are flagged as probable apnea

segments (PAS) and investigated further in the algorithm's

next phase. This algorithm phase's goal was to detect a wide

range of probable apnea segments while keeping the rejection

of false-positive occurrences to the following part.

The third and last phase of the algorithm (classification)

tries to properly assess the previously extracted PAS to

discriminate between apnea and non-apnea occurrences. The

essential phases of the algorithm in this section are depicted in

Fig.5 to facilitate comprehension. Each sound event in the

probable apnea segments should be detected separately to

discriminate between breathing and non-breathing episodes.

Then the related airflow in each breathing phase should be

computed.

4. Methodology

4.1 Apnea Detection

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

218

Volume 19, 2022

The pre-processed audio signal is separated into small

intervals, and the third envelope is obtained by calculating the

amplitude, as in the preceding portion.

E stands for the envelope size, the total number of samples

in the window stands for Ns, and the sample selection is Zi.

Because the logarithm quickly pushes to minimal negative

numbers when there is no noise, this form of the envelope is

suitable for distinguishing various sound occurrences isolated

by quiet. The final waveform is then subjected to a reduced

filtration with a stopband of (2 Hz), which is used to

determine a fixed limit. Sound episodes are E3 portions that

go over this threshold.

The active signal retrieved from the inertial measurement

unit is then employed to detect episodes produced by the

movement sound effects in the following step. Movement

artifact noise develops when activity exceeds a certain

threshold.

The present average breathing level (reference) should be

known to use the basic definition of apnea and distinguish

between breath and apnea. The importance of this stage is

underscored by the fact that, depending on sleeping posture,

the overall volume of breathing sounds can change

significantly during the night. After then, the method

determines which of the previously identified episodes in the

retrieved probable apnea segment should be employed like a

referencing, then those segments get categorized to be a

respiratory segment or not. In the next step, the long-term part

of the second envelope curve, repeat the basic threshold

approach employed in the Reduced prior detection. All

segments below that level are already referred to as apnea

segments, while those above it is referred to as reference

segments.

The next stage is to link the quantity of airflow to specific

tracheal sound properties (such as amplitude) within the good

episodes that have been recorded. Different strategies for

connecting airflow and breathing sounds have arisen from

various investigations of the two signals. Eq No (1) is

employed in the technique provided to determine an attribute

for each sound occurrence. On the other hand, individual event

outliers are deleted before feature calculation, much like

Reduced detection works. This is important because loud

clicking noises might occur during an apnea attack.

Within a particularly probable apnea, episodes are

classified. The envelope curve that recognizes sound episodes

using Eq is represented by the red color E3 (1). The heights of

the grey patches represent the feature extraction value. In

contrast, the grey sections reflect the detected sound

occurrences (negative values stem from using the ln function).

The good episodes in the reference segments have a feature

value. RS defines the referent segments. Because all apnea

segment episodes fall below that rate, they get classified as

not-breathing. As a consequence, the green colors indicate

where apnea was found.

Since relying just on respiratory signals may not always

produce better insights in some instances, we conducted a

cardio-respiratory study; more details are shown below.

The human body is affected physiologically and physically

by the functioning of the respiratory and cardiovascular

systems[36, 37]. Based on the human eye's low spatiotemporal

acuity, most of these impacts are undetectable to human

drivers, but in biological and clinical contexts, they can be

very instructive[38,39]. The cardiorespiratory signal can be

retrieved based on the following methods:

• SKIN COLOR CHANGES.

• ARTERIAL PULSE MOTION.

• CHEST MOTION.

• HEAD MOTION.

Since relying just on respiratory signals may not always

produce better insights in some instances, we conducted a

cardio-respiratory study; more details are shown below.

After the signal acquisition, and for a time period chosen.,

we studied three intervals that will lead the apnea detection

Fig. 6: Apnea detection simulation

4.2 Cardio-respiratory Signal Processing

4.3 Protocol

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

219

Volume 19, 2022

Fig. 7: A sample full set of the examined time period collected.

- [S-1]: Refers to the 75s that were recorded after the time

period chosen.

- [S-2]: Refers to the 3 minutes recorded up to 1 hour prior

to the chosen time period.

- [S-3]: Refers to the three minutes recorded one hour prior

to the [S-2] period.

we created sample complete sets of studied intervals in

linear arbitrary models with a random slope and intercept. 3-

min automatically recorded segments were accessible up to 2

hours S-2 and up to 1 hour before to the occurrence of interest

for each epoch of conventional.

These reference intervals must last at least 15 seconds after

the cycle began; they should involve at least three breaths, be

devoid of motion artifacts, and last 15 seconds after any

activity. Then we classified any breathing arrest lasting more

than 5 seconds as an apnea incident. Unfortunately, because of

the lack of data, we could not afford to include this part in our

application, more details and intimations are included in the

research’s cited [40, 41].

The results of the detection are shown in bellow. A total of

ten audios, the total of the running time is also presented

below. There are between 2 and 247 apnea episodes in each

recording.

Table 2: Detailed description of sleep health study database

Subjects

Sex

Height

Age

No.1

140

No.2

160

No.3

170

No.4

162

No.5

168

No.6

150

No.7

174

No.8

165

No.9

150

No.10

159

as for:

• TNA: the length of time without apnea that has been

appropriately categorized.

• TPA: stands for the total amount of accurately classified

apnea time.

• FNA: the length of time without apnea that was

mistakenly categorized.

• FPA: represents the amount of apnea time that has been

erroneously categorized.

• TPA: stands for the total number of apneas that have

been appropriately classified.

• FPA: the number of apneas that are wrongly categorized.

we were able to find six hundred and thirty episodes true

positive(TP) and fifty-two wrong ones false negative(FP). if

we consider true and wrong classified time segments a

sensitivity of ninety-two percent and a specificity of ninety-

nine percent.

We can see that our system was able to provide high

results compared to other systems, while minimizing costs

and disruptions without affecting system performance.

for other systems such as those based on EEG, EMG or

ECG, if they rely on a single measure to minimize cost and

time, the performance of the system decreases

significantly. and if they combine these measures (EEG,

EMG, ECG), the performance of the system will increase,

but the time and cost will increase significantly, as these

techniques require clinical assistance, leaving our system

the best performing and least expensive.

Even for systems that rely on video recording data,

there is evidence that they are only useful for infants.

The sensitivity and specificity of the devised apnea

detection algorithm were calculated as follows to assess its

performance:

Table 3: Apnea detection

Table 2: Detailed description of sleep health study database

5. Results

5.1 Results Statistics

5.2 Τhe System Performance

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

220

Volume 19, 2022

Table 3: Apnea detection statistics

The filtering part play a major role in the process of apnea

detection. We used a FIR filter bandpass (200-2000) Hz. we

tried looking for different filterers that can be used by our

system and we conducted a parallel comparison of each filter

and its effect on the result of apnea detection. the filters

information's are shown below:

a) CHEBYSHEV FILTER:

The Chebyshev category I filter optimizes the pace of cutoff

seen between passband and stopband of the sound quality at

the price of passband ripple and phase response noise [42].

b) HILBERT FILTER:

The harmonic spectrum's lower half is wiped out,

converting the real-valued input to a reinforced one [43].

c) THOMSON5 (BESSEL) FILTER:

With minimal ringing in the linear system, the analog

Bessel filter does have a maximum uniform group delay and a

fully linear phase reaction.

Bessel is an analog filter by definition. The bilinear shift is

used to build digital Bessel filtering, although the frequency

sensitivity of the analog filter is not preserved. As a result, for

harmonics below around fs/4, it is only nearly right. To get a

bandpass filter that is as flat as possible at high frequency[44].

Table 4: Filters statistics

Filters type

No of TD

No of FD

CHEBYSHEV

HILBERT

THOMSON

An innovative sleep monitoring technology has been

developed that allows people to identify apnea without

exerting too much effort.

This technique was created and supplied for recognizing

probable apnea episodes. This method was approved in further

research that deployed different methods and tools. They

demonstrate their ability to identify apnea episodes regularly

[20].

Four different filtering procedures were used to make a

more efficient experiment, and three significant stages were

needed to process the signals. All heart noises or disturbances

from the initial file during pre-processing were eliminated.

The Reduced detection detects potential apnea episodes by

analyzing the whole signal and trying to find some scraps in

the respiratory amplitude to discriminate between apnea and

non-apnea episodes. Finally, the apnea diagnostics get done in

the classification stage. This system does a very in-depth

analysis of all the signals so that any apnea episodes get

detected during all the processing stages.

Comparing our system to other systems, we can see that

it was able to deliver superior outcomes while minimizing

expenses and interruptions without degrading system

efficiency. If other systems, such those based on EEG[16],

ECG[17], or EMG[18] rely on just one technique to cut

costs and time, the system's performance suffers greatly.

and if they combine these measurements (EMG, EMG,

and ECG), the system's performance will improve, but the

time and cost will considerably increase because these

techniques demand clinical support, even if we count video

recording-based solutions, they are only helpful for young

children. making our system the best-performing and least

costly.

The algorithm's weakest spot is that it fails to detect

hypopnea episodes. Although Apnea-hypopnea index(AHI) is

usually the must study measurement that we should focus on

5.3 Filtering Options

5.4 Filtering Comparison

6. Discussion

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

221

Volume 19, 2022

to detect OSA but unfortunately we can't use this it in this

study. According to a thorough review of the PSG data that

included hypopnea episodes, most false-positive apnea

episodes in subjects five and eight are misread hypopnea

episodes. Changing classification levels to distinguish between

apnea and hypopnea episodes proved impracticable since the

study showed that we have many false-positive hypopnea

episodes, and this is so clear and repeated in files with a

number of apnea occurrences that is under than twenty.

According to American Academy of Sleep Medicine (AASM)

standards [45], hypopneas are identified as a droption in

airflow of at least 30%, incident excitation and an air

dehydration of much more over 3%.

Heart beat retrieval from the captured audio stream is

greatly interfered with during snoring bouts. This could also

be the case if the patient is talking or moving loudly while

recording an audio signal. Thus, throughout certain parts, heart

rate is adjusted. Other diagnostically significant information,

such as heart rates, could be estimated with sufficient

accuracy, making it possible to miss a rapid shift in pulse rate.

The proposed study has several pros and downsides. Even

though all of the patients were chosen because they were

suspected of having sleep apnea, the findings encompassed the

whole spectrum of sleep apnea from none to severe. The sex,

age, and BMI distributions also contain a wide variety of

people, indicating that the findings apply to the broader public.

Nonetheless, future research needs to include a larger

participant pool. Future research should be conducted in a

home situation since it would be fascinating to examine how

the suggested system functions without medical supervision.

Apneas can be detected in a clinical context using the

disclosed technique. The system's use of data acquired and

ability to estimate heart rate are distinguishing characteristics

compared to certain other technologies. Due to its fundamental

sensor arrangement, compared to current mobile sleeping

trackers, the tracking system proposed has proven to be way

more dependable and pleasant and also bought high results

throughout the development of a fully functional prototype in

other research. The device's fundamental flaw is its failure to

identify hypopnea episodes required for AHI calculation.

Future research will concentrate on incorporating these

episodes into existing algorithms and enhancing present

capabilities to provide a more comprehensive assessment of

sleep quality.

[1] M. R. Mannarino, F. Di Filippo, et M. Pirro,

« Obstructive sleep apnea syndrome », European journal

of internal medicine, vol. 23, no 7, p. 586‑593, 2012.

[2] T. Young, L. Evans, L. Finn, et M. Palta, « Estimation of

the clinically diagnosed proportion of sleep apnea

syndrome in middle-aged men and women », Sleep, vol.

20, no 9, p. 705‑706, 1997.

[3] I. M. Rosen et al., « Clinical use of a home sleep apnea

test: an American Academy of Sleep Medicine position

statement », Journal of Clinical Sleep Medicine, vol. 13,

no 10, p. 1205‑1207, 2017.

[4] M. T. Saletu, S. T. Kotzian, A. Schwarzinger, S. Haider,

J. Spatt, et B. Saletu, « Home sleep apnea testing is a

feasible and accurate method to diagnose obstructive

sleep apnea in stroke patients during in-hospital

rehabilitation », Journal of Clinical Sleep Medicine, vol.

14, no 9, p. 1495‑1501, 2018.

[5] T. Penzel, A. Blau, C. Garcia, C. Schöbel, M. Sebert, et I.

Fietze, « Portable monitoring in sleep apnea », Current

Respiratory Care Reports, vol. 1, no 2, p. 139‑145, 2012.

[6] J. Corral-Peñafiel, J.-L. Pepin, et F. Barbe, « Ambulatory

monitoring in the diagnosis and management of

obstructive sleep apnoea syndrome », European

Respiratory Review, vol. 22, no 129, p. 312‑324, 2013.

[7] G. M. Traaen et al., « Prevalence, risk factors, and type of

sleep apnea in patients with paroxysmal atrial

fibrillation », IJC Heart & Vasculature, vol. 26, p.

100447, 2020.

[8] R. Agrawal, J. A. Wang, A. G. Ko, et J. E. Getsy, « A

real-world comparison of apnea–hypopnea indices of

positive airway pressure device and polysomnography »,

Plos one, vol. 12, no 4, p. e0174458, 2017.

[9] F. Kapsimalis et M. H. Kryger, « Gender and obstructive

sleep apnea syndrome, part 1: clinical features », Sleep,

vol. 25, no 4, p. 409‑416, 2002.

[10] E. Romero, B. Krakow, P. Haynes, et V. Ulibarri,

« Nocturia and snoring: predictive symptoms for

obstructive sleep apnea », Sleep and Breathing, vol. 14,

no 4, p. 337‑343, 2010.

[11] H. Gotsopoulos, C. Chen, J. Qian, et P. A. Cistulli, « Oral

appliance therapy improves symptoms in obstructive

sleep apnea: a randomized, controlled trial », American

journal of respiratory and critical care medicine, vol.

166, no 5, p. 743‑748, 2002.

[12] A. V. Benjafield et al., « Estimation of the global

prevalence and burden of obstructive sleep apnoea: a

literature-based analysis », The Lancet Respiratory

Medicine, vol. 7, no 8, p. 687‑698, 2019.

[13] K. Sutherland et al., « A global comparison of anatomic

risk factors and their relationship to obstructive sleep

apnea severity in clinical samples », Journal of Clinical

Sleep Medicine, vol. 15, no 4, p. 629‑639, 2019.

[14] R. J. Walter, S. I. Hagedorn, et C. J. Lettieri, « Impact of

diagnosing and treating obstructive sleep apnea on

healthcare utilization », Sleep Medicine, vol. 38, p. 73‑77,

2017.

[15] S. J. Redmond, P. de Chazal, C. O’Brien, S. Ryan, W. T.

McNicholas, et C. Heneghan, « Sleep staging using

cardiorespiratory signals », Somnologie-Schlafforschung

und Schlafmedizin, vol. 11, no 4, p. 245‑256, 2007.

[16] T. Sugi, F. Kawana, et M. Nakamura, « Automatic EEG

arousal detection for sleep apnea syndrome », Biomedical

Signal Processing and Control, vol. 4, no 4, p. 329‑337,

2009.

[17] C. Varon, A. Caicedo, D. Testelmans, B. Buyse, et S.

Van Huffel, « A novel algorithm for the automatic

detection of sleep apnea from single-lead ECG », IEEE

7. Conclusion

References

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

222

Volume 19, 2022

Transactions on Biomedical Engineering, vol. 62, no 9, p.

2269‑2278, 2015.

[18] N. J. Douglas, M. A. Jan, N. Yildirim, P. M. Warren, et

G. B. Drummond, « Effect of posture and breathing route

on genioglossal electromyogram activity in normal

subjects and in patients with the sleep apnea/hypopnea

syndrome », American Review of Respiratory Disease,

vol. 148, p. 1341‑1341, 1993.

[19] S. Sharma, S. Bhattacharyya, J. Mukherjee, P. K. Purkait,

A. Biswas, et A. K. Deb, « Automated detection of

newborn sleep apnea using video monitoring system », in

2015 Eighth International Conference on Advances in

Pattern Recognition (ICAPR), 2015, p. 1‑6.

[20] L. Almazaydeh, K. Elleithy, M. Faezipour, et A.

Abushakra, « Apnea detection based on respiratory signal

classification », Procedia Computer Science, vol. 21, p.

310‑316, 2013.

[21] C. Avcı et A. Akbaş, « Sleep apnea classification based

on respiration signals by using ensemble methods », Bio-

medical materials and engineering, vol. 26, no s1, p.

S1703‑S1710, 2015.

[22] G. H. Caumo, D. Spritzer, A. Carissimi, et A. C. Tonon,

« Exposure to electronic devices and sleep quality in

adolescents: A matter of type, duration, and timing »,

Sleep health, vol. 6, no 2, p. 172‑178, 2020.

[23] T. Penzel, G. B. Moody, R. G. Mark, A. L. Goldberger, et

J. H. Peter, « The apnea-ECG database », in Computers in

Cardiology 2000. Vol. 27 (Cat. 00CH37163), 2000, p.

255‑258.

[24] A. L. Goldberger et al., « PhysioBank, PhysioToolkit,

and PhysioNet: components of a new research resource

for complex physiologic signals », circulation, vol. 101,

no 23, p. e215‑e220, 2000.

[25] B. Malley, D. Ramazzotti, et J. T. Wu, « Data Pre-

processing », in Secondary Analysis of Electronic Health

Records, MIT Critical Data, Éd. Cham: Springer

International Publishing, 2016, p. 115‑141. doi:

10.1007/978-3-319-43742-2_12.

[26] A. Yadollahi, E. Giannouli, et Z. Moussavi, « Sleep

apnea monitoring and diagnosis based on pulse oximetery

and tracheal sound signals », Medical & biological

engineering & computing, vol. 48, no 11, p. 1087‑1097,

2010.

[27] H. Pasterkamp, « KramanS. S., WodickaG. R »,

Respiratory sounds: advances beyond the stethoscope.

Am J Respir Crit Care Med, vol. 156, p. 974‑987, 1997.

[28] H. Nakano, M. Hayashi, E. Ohshima, N. Nishikata, et T.

Shinohara, « Validation of a new system of tracheal

sound analysis for the diagnosis of sleep apnea-hypopnea

syndrome », Sleep, vol. 27, no 5, p. 951‑957, 2004.

[29] C. Iber, « The AASM manual for the scoring of sleep and

associated events: Rules », Terminology and Technical

Specification, 2007.

[30] H.-J. Jang, J. W. Kim, et D.-H. Hwang, « Robust step

detection method for pedestrian navigation systems »,

Electronics Letters, vol. 43, no 14, p. 1, 2007.

[31] H. Alshaer, G. R. Fernie, E. Maki, et T. D. Bradley,

« Validation of an automated algorithm for detecting

apneas and hypopneas by acoustic analysis of breath

sounds », Sleep medicine, vol. 14, no 6, p. 562‑571, 2013.

[32] H. Alshaer, G. R. Fernie, E. Sejdić, et T. D. Bradley,

« Adaptive segmentation and normalization of breathing

acoustic data of subjects with obstructive sleep apnea »,

in 2009 IEEE Toronto International Conference Science

and Technology for Humanity (TIC-STH), 2009, p.

279‑284.

[33] K. A. Sohrabi et al., « Quantification of nasal respiratory

flow by tracheal sound analysis », Biomedical

Engineering/Biomedizinische Technik, vol. 57, no SI-1-

Track-L, p. 733‑735, 2012.

[34] A. Yadollahi, A. Azarbarzin, A. Montazeri, et Z.

Moussavi, « Acoustical flow estimation in patients with

obstructive sleep apnea during sleep », in 2012 Annual

International Conference of the IEEE Engineering in

Medicine and Biology Society, 2012, p. 3640‑3643.

[35] I. Hossain et Z. Moussavi, « Respiratory airflow

estimation by acoustical means », in Proceedings of the

Second Joint 24th Annual Conference and the Annual

Fall Meeting of the Biomedical Engineering

Society][Engineering in Medicine and Biology, 2002, vol.

2, p. 1476‑1477.

[36] M. M. Platiša, N. N. Radovanović, A. Kalauzi, G.

Milašinović, et S. U. Pavlović, « Multiscale entropy

analysis: Application to cardio-respiratory coupling »,

Entropy, vol. 22, no 9, p. 1042, 2020.

[37] R. P. Bartsch, K. K. Liu, Q. D. Ma, et P. C. Ivanov,

« Three independent forms of cardio-respiratory

coupling: transitions across sleep stages », in Computing

in Cardiology 2014, 2014, p. 781‑784.

[38] K. Niizeki, I. Nishidate, K. Uchida, et M. Kuwahara,

« Unconstrained cardiorespiratory and body movement

monitoring system for home care », Medical and

Biological Engineering and Computing, vol. 43, no 6, p.

716‑724, 2005.

[39] S. J. Redmond et C. Heneghan, « Cardiorespiratory-based

sleep staging in subjects with obstructive sleep apnea »,

IEEE Transactions on Biomedical Engineering, vol. 53,

no 3, p. 485‑496, 2006.

[40] K. Karandikar, T. Q. Le, A. Sa-ngasoongsong, W.

Wongdhamma, et S. T. Bukkapatnam, « Detection of

sleep apnea events via tracking nonlinear dynamic cardio-

respiratory coupling from electrocardiogram signals », in

2013 6th International IEEE/EMBS Conference on

Neural Engineering (NER), 2013, p. 1358‑1361.

[41] G. Guerrero, V. Le Rolle, et A. Hernández, « Parametric

analysis of an integrated model of cardio-respiratory

interactions in adults in the context of obstructive sleep

apnea », Annals of Biomedical Engineering, vol. 49, no

12, p. 3374‑3387, 2021.

[42] P. Podder, M. Hasan, M. Islam, et M. Sayeed, « Design

and implementation of Butterworth, Chebyshev-I and

elliptic filter for speech signal analysis », arXiv preprint

arXiv:2002.03130, 2020.

[43] W. M. Laghari, M. U. Baloch, M. A. Mengal, et S. J.

Shah, « Performance analysis of analog butterworth low

pass filter as compared to Chebyshev type-I filter,

Chebyshev type-II filter and elliptical filter », Circuits

and Systems, vol. 2014, 2014.

[44] J. C. T. Badajos, K. M. P. Ochoa, R. A. P. Tejadilla, et R.

T. L. Peñas, « Reduction of audio noise with lowpass

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

223

Volume 19, 2022

Chebyshev Type II filter simulated using GNU octave »,

in 2019 IEEE 11th International Conference on

Humanoid, Nanotechnology, Information Technology,

Communication and Control, Environment, and

Management (HNICEM), 2019, p. 1‑6.

[45] A. T. Thornton, P. Singh, W. R. Ruehland, et P. D.

Rochford, « AASM criteria for scoring respiratory events:

interaction between apnea sensor and hypopnea

definition », Sleep, vol. 35, no 3, p. 425‑432, 2012.

Creative Commons Attribution License 4.0

(Attribution 4.0 International, CC BY 4.0)

This article is published under the terms of the Creative

Commons Attribution License 4.0

https://creativecommons.org/licenses/by/4.0/deed.en_US

WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS

DOI: 10.37394/23209.2022.19.22

Younes El Ouahabi, Kaoutar Baggar,

Benayad Nsiri, My Hachem El Yousfi Alaoui,

Abdelmajid Soulaymani,

Abdelrhani Mokhtari, Brahim Benaji

E-ISSN: 2224-3402

224

Volume 19, 2022