A PROJECT REPORT ON

FOURIER AND WAVELET TRANSFORM BASED SLEEP

APNEA DETECTION IN ECG SIGNALS
Submitted in partial fulfillment of the requirements for the award of the
degree of

BACHELOR OF TECHNOLOGY
In
ELECTRONICS AND COMMUNICATION ENGINEERING

Submitted by
POTHUGANTI MANASA 23EG104C53
CHILUKAMARI KARTHIK 23EG104C08
SUTHOJU VAMSHI KRISHNA 23EG104C29
Under the esteemed guidance of
Dr. POLI LOKESHWARA REDDY, [Link]., Ph.D
Assistant Professor
Department of ECE

DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING

ANURAG UNIVERSITY
Venkatapur , Ghatkesar , Medchal-Malkajgiri District, Hyderabad,
Telangana, India-500088
2023-2024
 ANURAG UNIVERSITY
Venkatapur, Ghatkesar, Medchal-Malkajgiri District, Hy derabad,
Telangana, India-500088

DEPARTMENT OF ELECTRONICS & COMMUNICATION ENGINEERING

CERTIFICATE
This is to certify that the project report entitled “FOURIER AND WAVELET
TRANSFORM BASED SLEEP APNEA DETECTION IN ECG SIGNALS” is a
bonafide record submitted by [Link] , [Link] KRISHNA , [Link] ,
Department of Electronics and communication Engineering, Anurag University and
is submitted in partial fulfillment for the award of Degree of Bachelor of Technology
in “Electronics and Communication Engineering” for the year 2023-2024. The work
reported herein does not form part of any other thesis on which a degree has been
awarded earlier.

Project Supervisor Head of the Department

Dr. Poli Lokeshwara Reddy, [Link]., Ph.D Prof. N. Mangala Gouri
Assistant Professor, Dept. of ECE Professor, Dept. of ECE
Anurag University Anurag University

Viva-Voce held on: Examiner

2
 ACKNOWLEDGEMENT
It is indeed a great pleasure to thank all the people who had given support in
completion of the project.
We would like to express our deep sense of gratitude and respect to our project guide
Dr. Poli Lokeshwara Reddy, Assistant Professor, Department of Electronics and
Communication Engineering, for his encouragement, excellent guidance and kind
support during our project, without which it would not have been possible to complete
this work. We are thankful to him for being highly cooperative throughout our project
work.
We are extremely grateful to Prof. N. Mangala Gouri, Head of the Department of
Electronics and Communication Engineering, for her encouragement and support
throughout the project.
Special thanks are dedicated to Prof. [Link] Kumar, Dean School of Engineering,
who helped us for the successful completion of our project.
We are extremely grateful to Dr. Palla Rajeshwar Reddy, Chairman, Anurag Group
of Institutions, who encouraged and helped us for the successful completion of our
project.
It is our pleasure to express thanks to all the teaching and non-teaching staff of
Electronics and Communication Engineering Department for their excellent
monitoring and their suggestions that helped in the successful completion of
coursework for [Link], Program.
Finally, we thank our parents and the Almighty, whose divine grace provided us the
opportunity to do our project work according to our wish.

Project associates…….

POTHUGANTI MANASA (23EG104C53)

CHILUKAMARI KARTHIK (23EG104C08)
SUTHOJU VAMSHI KRISHNA (23EG104C29)

3
 DECLARATION

We hereby declare that the project report entitled, “FOURIER AND WAVELET
TRANSFORM BASED SLEEP APNEA DETECTION IN ECG SIGNALS”, is
the work done by us and submitted for the partial fulfillment of the requirements for
the award of the degree of Bachelor of Technology in Electronics and
Communication Engineering, under the guidance of Dr. Poli Lokeshwara Reddy,
[Link]., Ph.D,, Assistant Professor, Department of Electronics and Communication
Engineering, Anurag University.

We further declare that this project report has not been previously submitted before
either in part or full for the award of any degree or any diploma by any organization
or any universities.

Project associates…….
POTHUGANTI MANASA (23EG104C53)
CHILUKAMARI KARTHIK (23EG104C08)
SUTHOJU VAMSHI KRISHNA(23EG104C29)

4
 ABSTRACT

Sleep apnea, a common sleep disorder characterized by interruptions in breathing during sleep,
poses significant health risks and can lead to various cardiovascular and neurological
complications. This project focuses on the development of a robust method for the detection of
sleep apnea using Electrocardiogram (ECG) signals. The proposed approach employs both
Fourier and wavelet transforms, integrating signal processing techniques to enhance detection
accuracy.

The initial phase of the project involves the application of a Butterworth bandpass filter to
preprocess the ECG signals, removing noise and isolating relevant frequency components. The
Fourier transform is then applied to compute the Power Spectral Density (PSD), capturing
frequency-domain features indicative of sleep apnea. Additionally, wavelet transform analysis
is performed to extract wavelet coefficients, offering time-frequency information that aids in
detecting apnea-related patterns.

Subsequent analysis involves the comparison of PSD and wavelet coefficients with dynamic
thresholds. These thresholds are adaptive and designed to identify significant deviations in
signal features, facilitating the detection of apnea and non-apnea events. By leveraging both
Fourier and wavelet-based analysis, this research provides a reliable and efficient
computational approach for sleep apnea detection, offering a non-invasive diagnostic tool for
early identification of individuals at risk.

Keywords: ECG, Fourier Transform, Wavelet Transform, Sleep Apnea Detection, Feature
Extraction, Power Spectral Density, Dynamic Thresholding.

5
 CONTENTS

CERTIFICATE

ACKNOWLEDGEMENT

DECLARATION

ABSTRACT

CONTENTS

LIST OF FIGURES

LIST OF TABLES

LIST OF ABBREVIATIONS

CHAPTER-I

INTRODUCTION 1-9

1.1 What is a signal 1

1.2 Types Of Signals 1-3

1.2.1 Electrocardiogram (ECG) 3

1.3 What is Signal Processing ? 3

1.4 Application of Signal Processing 4-5

6
 1.5 Preproccessing approaches 6-7

1.6 Problem Statement 7

1.7 Objectives of the proposed work 7-8

1.8 Organization of Report 8-9

CHAPTER-II

LITERATURE REVIEW 10-20

CHAPTER-III

EXISTING METHODS 20-23

3.1 Autoregressive (AR) Models 20-21

3.2 Periodogram 21-22

CHAPTER-IV

PROPOSED METHODOLOGY 23-26

4.1 Introduction 23

4.2 Methodology 25

4.2.1 Collection of data 23

4.2.2 Preprocessing 24

4.2.3 Feature Extraction 24-

25 [Link] Power Spectral density 24

[Link] Wavelet Coefficients 25

7
 4.2.4 Threshold-Based Detection Of Apnea Events 25

4.2.5 Classification: Apnea Vs Non-Apnea 25

4.3 Performance Metrics 25-26

4.4 Block diagram of proposed methodology 26

CHAPTER-V

SOFTWARE DESCRIPTION 27-36

5.1. Introduction 27
5.2 Basics of Software 27-33
5.2.1. Basic Building Blocks of MATLAB 27-29
5.2.2. MATLAB Files 30
5.2.3. MATLAB Commands Required 30-32
5.2.4. Software Requirements 33
5.3 Design and Implementation 33
5.3.1. Introduction 33
5.3.2. System Architecture 33
5.3.3. Starting the MATLAB 33
5.3.4. Creating a New Project 33-34
5.4. Method of Implementation 34-36
CHAPTER-VI

RESULTS AND ANALYSIS 37-45

6.1 Input Signals 37-38

8
6.2 Preprocessing 38-39

6.3 Feature Extraction 39-43

6.3.1 Fourier Transform Analysis 39-41

6.3.2 Wavelet Transform Analysis 41-43

6.4 Detecting Sleep Apnea Events 42-45

CHAPTER-VII

ADVANTAGES AND APPLICATIONS 46-47

7.1 Advantages 46

7.2 Applications 46-47

CHAPTER-VIII

CONCLUSION AND FUTURE SCOPE 48-50

REFERENCES 50-51

APPENDIX 51-56

9
 LIST OF FIGURES
Figure No. Name of the Figure Page
No.

Fig 3.2.1 23

Fig 4.4.1 Block diagram of proposed methodology 26

Fig 4.2 Flowchart of the proposed methodology 29

Fig 5.1(a) Command Window 31

Fig 5.1(b) Command Window Showing Values 32

Fig 5.2 Workspace Window 32

Fig 5.3 MATLAB Icon 36

Fig 5.4(a) MATLAB Layout 38

Fig 5.4(b) Browse the Source File 38

Fig 5.4(c) Select the Source File 39

Fig 5.4(d) Main Source File 39

Fig 6.1 Average Temperature plot in New York 40

Fig 6.2 Average Temperature plot in Begumpet 41

10
 LIST OF TABLES
Table No. Name of the Table Page No.

Table 2.1 Literature Review Summary 12-20

Table 3.3 Advantages and limitations of 22-23

Existing methods

Table 6.2 Comparing predicted value to the 42

actual value in Begumpet

ABBREVATIONS

ECG Electrocardiogram

PSD Power Spectral Density

MATLAB Matrix Laboratory

11
 CHAPTER - I

INTRODUCTION

1.1 What is Signal?

A signal is a measurable quantity that varies with time, space, or another independent
variable. In simpler terms, it's a representation of information or data that changes
over a particular parameter, such as time. Signals can be found in various forms,
including audio, video, temperature, pressure, or even biological activities like the
electrical impulses in our bodies. Signals are mainly classified into two types. They
are Continuous time signals and discrete time signals.

1.2 Types of Signals

Signals can be broadly categorized into various types based on their characteristics,
properties, and applications. Here are some common types of signals:

[Link]-Time Signals:

Continuous-time signals are representations of data that vary smoothly and

continuously over an infinite range of time. These signals are often described by
mathematical functions, allowing for precise analysis and modeling. Examples
include sinusoidal waves, exponential growth, and various analog electrical signals.

[Link]-Time Signals:

12
In contrast to continuous-time signals, discrete-time signals are defined only at
specific time points. These signals are commonly obtained through sampling
continuous-time signals. Sequences of numbers and digital audio samples are typical
examples of discrete-time signals, frequently encountered in digital signal processing.

[Link] Signals:

Digital signals are discrete in both time and amplitude, often taking on quantized
values. These signals are essential in digital communication and computing systems,
where information is represented in a binary form. Digital signals include binary code,
digital images, and any data stored or transmitted in a digital format.

[Link] Signals:

Periodic signals exhibit a repetitive pattern at regular intervals. These signals are
characterized by their periodicity, allowing for the analysis of cyclical phenomena.
Examples encompass sine waves, square waves, and other waveforms commonly
encountered in signal processing and communications.

[Link] Signals:

Aperiodic signals lack a repeating pattern over time. These signals are crucial in
representing transient events and phenomena without a regular structure. Examples
include impulse signals, step functions, and other non-repeating waveforms.
[Link] Signals:

Deterministic signals have predictable and precisely defined behaviors. They are often
described using mathematical functions or equations, making them amenable to
analytical techniques. Deterministic signals are fundamental in areas such as control
systems and mathematical modeling.

[Link] Signals (Stochastic Signals):

Random signals exhibit an element of randomness or unpredictability. These signals

are characterized by statistical properties and are commonly encountered in fields
such as communications and signal processing. Examples include noise signals and
random vibrations.

13
[Link] Signals: Analog signals are continuous signals that can take on a range
of values. They are commonly found in natural phenomena and analog electronic
systems. Analog signals include voltage signals in electrical circuits and analog audio
signals.

[Link] Signals:

Binary signals are a specific type of digital signal that takes on only two discrete
levels or states. They form the basis of digital communication and computing, where
information is represented using binary code and conveyed through on/off states.

[Link] Signals:

Multichannel signals consist of multiple components or channels. These signals are

prevalent in multimedia applications, where different channels may represent various
aspects of the signal, such as color channels in images or audio tracks in a recording.

1.2.1 Electrocardiogram(ECG)
The electrocardiogram (ECG) is a graphical representation of the heart's electrical
activity, recorded through electrodes placed on the skin. These electrodes detect the
small electrical changes caused by the polarization and depolarization of the heart's
cells during each heartbeat. Depolarization refers to the process where the heart's
muscle cells lose their electrical charge, leading to contraction, while repolarization is
the process of regaining the electrical charge, allowing the heart to relax. The ECG
captures these changes as distinct waveforms, such as the P wave (atrial
depolarization), QRS complex (ventricular depolarization), and T wave (ventricular
repolarization). By analyzing these waveforms, healthcare providers can diagnose
cardiac conditions like arrhythmias and myocardial infarction, and researchers can use
ECG signals to study physiological processes like heart rate variability and sleep
apnea detection.

1.3 What is Signal Processing?

Signal processing is the analysis and manipulation of signals, such as sound,
images, or physiological data like ECG, to extract useful information or enhance

14
their quality. It involves techniques like filtering to reduce noise, transforming
signals using methods like the Fourier transform for frequency analysis, and
feature extraction. In digital signal processing (DSP), signals are converted into
discrete data and processed using algorithms. Signal processing is crucial in
various fields, including communications, audio and image processing, and
biomedical engineering, where it plays a key role in applications like speech
recognition, image compression, and medical diagnostics.

1.4 Applications of Signal Processing

Signal processing plays a critical role in various fields, offering significant
advancements and improvements:

[Link]:
Signal processing enhances clarity, reduces noise, and facilitates data
compression in telecommunication systems, ensuring efficient and reliable
communication.
2. Audio Processing:
It improves sound quality in music and speech, supports noise reduction, and
enables advanced voice recognition systems, contributing to clearer and more
accurate audio experiences.
3. Image and Video Processing:
Techniques in signal processing enhance image quality, enable effective image
compression, and support applications like facial recognition and computer
vision, which are essential for security and automated systems.

4. Biomedical Engineering:
In this field, signal processing is crucial for analyzing physiological signals such
as ECG and EEG. It aids in medical diagnostics, patient monitoring, and the
development of diagnostic tools.
5. Radar and Sonar:

15
Signal processing improves the detection and tracking of objects by filtering and
analyzing reflected signals, which is vital for navigation, surveillance, and
environmental monitoring.
6. Speech Processing:
It supports speech recognition and synthesis, enhancing applications such as
virtual assistants and transcription services by making interactions with
technology more intuitive.
7. Seismology:
Signal processing techniques are used to analyze seismic signals, helping to
monitor and understand earthquakes and other geological activities, which is
essential for disaster preparedness and research.
8. Control Systems:
Signal processing is integral to control systems for tasks such as feedback
control, filtering, and system identification, contributing to automation in
industries and robotics.
[Link] Networks:
Signal processing is used in sensor networks for extracting meaningful
information from sensor data, enabling applications in environmental
monitoring, smart cities, and industrial automation.
[Link] Signal Processing:
In finance, signal processing techniques are applied for analyzing financial time
series data, predicting market trends, and risk management.
[Link] Systems:
Signal processing techniques are applied in power systems for monitoring,
control, and fault detection, contributing to the stability and efficiency of
electrical grids.
[Link] Signal Processing:
In acoustics, signal processing is applied for tasks like noise cancellation, audio
recognition, and spatial audio processing.
[Link] Signal Processing:

16
Signal processing is used in environmental monitoring for analyzing signals
from sensors measuring air quality, water quality, and other environmental
parameters.
[Link] and Defense:
Signal processing is vital in radar systems, satellite communication, and
navigation systems for applications in aerospace and defense.

1.5 Preprocessing Approaches

1.5.1 Introduction to Preprocessing
Preprocessing is a critical step in ECG signal analysis as it helps in improving
signal quality by removing noise and artifacts. In the context of sleep apnea
detection, preprocessing ensures that the subsequent analysis using Fourier and
wavelet transforms is accurate and reliable.
1.5.2 Butterworth Bandpass Filtering
Overview
Butterworth filters are commonly used in signal processing for their smooth
frequency response and minimal phase distortion. A Butterworth Bandpass
Filter (BPF) allows signals within a certain frequency range to pass through
while attenuating frequencies outside this range. This characteristic is essential
for isolating relevant features in ECG signals that are indicative of sleep apnea.

Design and Implementation

Filter Design: The Butterworth BPF is designed to have a flat frequency
response in the passband and a smooth roll-off in the stopband. The filter is
characterized by its order (determining the sharpness of the cutoff) and cutoff
frequencies.
Order Selection: The order of the filter affects its performance. A higher order
provides a steeper roll-off but may introduce more delay and potential phase

17
distortion. For ECG signals, a typical order is chosen to balance between
adequate attenuation of unwanted frequencies and minimal signal distortion.
Cutoff Frequencies: The cutoff frequencies are selected based on the specific
frequency range of interest in the ECG signal. For instance, to focus on the
frequency components relevant to sleep apnea, typical cutoff frequencies might
be set between 0.1 Hz to 30 Hz, depending on the noise characteristics and the
specific features of interest.

Implementation in MATLAB:
In MATLAB, the Butterworth filter can be designed using the butter function.
For example:
[b, a] = butter(order, [low_cutoff, high_cutoff] / (Fs / 2), 'bandpass');
filtered_ecg = filtfilt(b, a, raw_ecg);
Here, order is the filter order, low_cutoff and high_cutoff are the cutoff
frequencies, Fs is the sampling frequency, and raw_ecg is the original ECG
signal.
Hence, effective preprocessing using Butterworth Bandpass Filtering is essential
for enhancing the quality of ECG signals in sleep apnea detection projects. It
ensures that the subsequent analysis steps—Fourier and wavelet transforms—
are applied to clean and relevant data, thereby improving detection accuracy and
robustness.

1.6 Problem Statement

Sleep apnea is a severe sleep disorder characterized by frequent interruptions in
breathing during sleep. These interruptions can lead to serious health complications,
including cardiovascular diseases, hypertension, and cognitive impairments. Early and
accurate detection of sleep apnea is crucial for effective management and treatment.
Electrocardiogram (ECG) signals provide a rich source of physiological data that can
be used for detecting sleep apnea. However, the analysis of ECG signals for this
purpose is challenging due to the presence of noise, variability in apnea events, and
the need for precise analysis of transient features.

18
Traditional methods often struggle to capture the subtle changes associated with sleep
apnea and may not be sufficient for reliable detection. This project addresses these
challenges by employing advanced signal processing techniques, specifically Fourier
and wavelet transforms, combined with robust preprocessing methods to improve the
accuracy of sleep apnea detection from ECG signals.

1.7 Objectives of Proposed Work

● To Implement Preprocessing: Design and apply a Butterworth Bandpass Filter
to preprocess ECG signals, aiming to enhance signal quality by removing noise
and artifacts.

● To Analyze Frequency Components: Utilize the Fourier Transform to examine

the power spectral density (PSD) of the ECG signal, identifying key frequency
components associated with sleep apnea.

● To Extract Time-Frequency Features: Apply the Wavelet Transform to capture

time-frequency characteristics of the ECG signal, enabling the detection of
transient features indicative of sleep apnea events.

● To Develop Detection Algorithms: Create and apply dynamic thresholding

techniques to classify ECG segments as apnea or non-apnea based on the features
extracted through Fourier and wavelet transforms.

● To Evaluate System Performance: Measure the effectiveness of the proposed

detection system using performance metrics such as accuracy, sensitivity, and
specificity to validate its reliability and effectiveness in detecting sleep apnea.

1.8 Organization of Report

● Chapter I: Introduction

19
 This chapter provides an introductory overview of sleep apnea detection,
highlighting the significance of accurate diagnosis and the role of ECG signals in
detecting sleep apnea. It also outlines the challenges associated with analyzing
ECG signals and the need for advanced signal processing techniques.

● Chapter II: Literature Review

This chapter reviews existing research and methods related to sleep apnea
detection using ECG signals. It covers previous studies, discusses various signal
processing techniques, and highlights their strengths and limitations.

● Chapter III: Existing Methodologies

This chapter details existing methodologies for sleep apnea detection that are less
effective compared to the proposed approach. It includes a critique of traditional
methods and an explanation of why they fall short in addressing the challenges
posed by ECG signal analysis.

● Chapter IV: Proposed Methodology

This chapter presents the proposed methodology for detecting sleep apnea. It
describes the preprocessing approach using Butterworth Bandpass Filtering, the
application of Fourier and wavelet transforms, and the dynamic thresholding
technique for event detection. It explains how these methods collectively enhance
the accuracy of sleep apnea detection.

● Chapter V: Software Used

This chapter discusses the software tools used in the project, specifically
MATLAB. It provides an overview of how MATLAB is utilized for signal
processing, feature extraction, and simulation of results.

● Chapter VI: Simulation Results

This chapter presents the results obtained from the simulation of the proposed
methodology. It includes visual and quantitative analyses of the detected sleep
apnea events, comparisons with existing methods, and the evaluation of
performance metrics such as accuracy and sensitivity.

20
● Chapter VII: Advantages and Applications

This chapter outlines the advantages of the proposed methodology over existing
methods. It also explores potential applications of the detection system in clinical
settings, wearable devices, and other practical scenarios.

● Chapter VIII: Conclusion and future scope

This chapter summarizes the key findings of the project, reiterates the significance
of the proposed methodology, and discusses the overall impact on sleep apnea
detection. It also includes recommendations for future work and potential
improvements.

CHAPTER – II

LITERATURE REVIEW
Quiceno-Manrique et al. [1] proposed a methodology to detect obstructive sleep
apnea (OSA) using electrocardiogram (ECG) recordings. The method applies time-
frequency analysis to heart rate variability (HRV) signals, which are extracted from
ECG data. They used Cohen’s class of quadratic time-frequency distributions to
analyze the HRV signals and extract dynamic features such as linear frequency
cepstral coefficients and spectral centroids. These features were then input into a k-
nearest neighbor (k-NN) classifier, which distinguished between normal and
pathological signals. The proposed system achieved an accuracy of up to 92.67% in
classifying one-minute intervals of ECG data. This approach allows for a simplified
diagnosis tool for OSA, using a non-invasive and cost-effective method compared to
standard polysomnography.

In the study by L. Hejjel and I. Gál [2], the authors provided an in-depth analysis of
heart rate variability (HRV) through both time-domain and frequency-domain
methods. The HRV fluctuations, controlled by sympathetic and parasympathetic

21
influences, were used to assess autonomic nervous system functioning under various
conditions such as stress, posture changes, and disease states. The study demonstrated
the importance of spectral analysis methods like Fast Fourier Transform (FFT) in
HRV evaluation and highlighted the potential clinical applications of HRV in
monitoring cardiovascular health, predicting disease outcomes, and assessing physical
and mental stress levels.

J.A. Fiz, et al. [3] proposed an acoustic analysis of snoring sounds in patients with
simple snoring and obstructive sleep apnoea (OSA). The study analyzed the snoring
sound power spectrum of 17 male patients (10 with OSA and 7 with simple snoring)
using full-night polysomnography. Two distinct acoustic patterns were observed,
distinguishing OSA patients from simple snorers based on snoring frequency. OSA
patients had significantly lower peak frequencies, with a negative correlation between
the apnoea-hypopnoea index (AHI) and snoring sound frequencies. This analysis
offers potential for developing a tool for distinguishing OSA from simple snoring.

P. Janbakhshi and M.B. Shamsollahi [4] proposed an automatic obstructive sleep

apnea (OSA) detection system based on single-lead electrocardiogram (ECG)-derived
respiration (EDR) signals. The system introduces two novel methods, EDRVar and
EDRPSR, for extracting EDR signals and evaluates their performance in detecting
OSA events. These EDR signals, along with RR interval series, are used as features
for machine learning classifiers such as Linear Discriminant, Support Vector
Machines (SVM), and Artificial Neural Networks (ANN). The proposed system
achieved high accuracy, with up to 100% in subject-based classification and
approximately 90.9% in minute-based apnea detection. This method shows promise
for developing a reliable, non-invasive OSA detection tool.

Margot Deviaene et al. [5] proposed a classification algorithm for detecting sleep
apnea using pulse photoplethysmography (PPG) signals. The algorithm is based on a
least-squares support vector machine (LS-SVM) and processes features derived from

22
PPG, including pulse rate and amplitude variability. The model was trained and tested
on polysomnographic data from 102 subjects suspected of sleep apnea. The results
showed that using PPG alone yielded a classification accuracy of 68.7%. When
combined with oxygen saturation (SpO2) features, the accuracy improved to 83.4%,
slightly better than using SpO2 alone. Although PPG features are promising, their
added value beyond SpO2 is limited.

Yuliya Zhivolupova [6] proposed an algorithm for detecting Cheyne-Stokes breathing

and identifying sleep apnea and hypopnea episodes using impedance pneumography.
The algorithm is designed to enhance the capabilities of Holter monitoring systems,
allowing for at-home screening of potential patients without the need for complex
equipment. It focuses on noise detection and removal to improve the accuracy of
apnea and hypopnea detection. The algorithm was tested on 19 night recordings, with
10 used for training and 9 for testing, yielding satisfactory results for identifying sleep
disorders but without differentiating between obstructive and central apnea.

Majdi Bsoul, et al. [7] proposed a real-time sleep apnea monitoring system, called
Apnea MedAssist, which utilizes single-lead ECG data. The system extracts features
from the nocturnal ECG of patients and employs support vector classifiers (SVC) to
detect obstructive sleep apnea (OSA) episodes. The system runs on Android
smartphones and offers two SVC models: subject-independent and subject-dependent.
The study reported an F-measure of 90% and 96% sensitivity for the subject-
independent SVC, making it suitable for home and clinical applications for detecting
OSA.

Kaicheng Feng et al. [8] proposed a sleep apnea detection method using single-lead
ECG signals, focusing on unsupervised feature learning. The method is based on a
frequential stacked sparse autoencoder (FSSAE) for automatic feature extraction and
a time-dependent cost-sensitive (TDCS) classification model, which combines the
Hidden Markov Model (HMM) and the MetaCost algorithm. The approach aims to

23
improve classifier performance by addressing data imbalance and considering
temporal dependencies. The model was trained on polysomnographic data and
achieved 85.1% accuracy, 86.2% sensitivity, and 84.4% specificity. The study
highlights that ECG-based sleep apnea detection using unsupervised learning can
provide a viable and efficient alternative to traditional methods.

LITERATURE REVIEW SUMMARY:

24
S. Journal Title Methodology Advantages Limitations
Name/Vol./
No.
Page No.
Year

1 IEEE Detection of The approach Time- Time-frequency

international
Obstructive involves frequency analysis can be
conference sleep apnea analyzing ECG analysis is computationally
on in ECG signals with sensitive to intensive ,
engineering recordings time-frequency the subtle requiring
in medicine using time- distributions , changes in the significant
frequency which allows ECG that are processing
And biology
distributions for the indicative of power, which
society
and dynamic detection of OSA , making might limit its
(EMBC),(pg
features. changes in it effective for real-time
no:5559-
heart rate and this specific application.
5562,2009 )
other dynamic application. The accuracy of
features the method may
associated with decrease with
sleep apnea poor signal
events quality , which
is common in
long-term ECG
recordings
during sleep.

2 Acta Heart rate time-domain Provides Requires clean,

Physiologica variability and frequency- detailed artifact-free
Hungarica, analysis domain insight into ECG data and
Volume 88 methods for autonomic long recordings
(3-4), pp. heart rate regulation and for accuracy.
219–230, variability is a non- Non-linear
2001 (HRV) invasive methods are not
measurement. method that yet widely used
Also discusses can predict in clinical
non-linear cardiovascular practice.
methods like and stress
Poincaré plots conditions.
and Fast
Fourier
Transform
(FFT).

25
 Table 2.1 Summary of Literature Review

CHAPTER – III
EXISTING METHODS

Sleep apnea is detected by different methods and processes, different therapies are
used to treat the apnea. Following are the existing techniques for detection and
treatment of sleep apnea:

3.1 Autoregressive (AR) Models

Sleep apnea is a condition where a person’s breathing repeatedly stops and starts
during sleep. This can lead to poor sleep quality and other health problems like heart
disease and high blood pressure.

Sleep apnea is a condition where a person’s breathing repeatedly stops and starts
during sleep. This can lead to poor sleep quality and other health problems like heart
disease and high blood pressure.

Autoregressive (AR) models are a way of predicting a future value of a signal based
on its past values. In simple terms, they look at the recent history of the signal to
make a prediction about its next step. For example, in sleep apnea detection, an AR
model can analyze a signal like heart rate or breathing pattern to see if there’s a
repeated abnormal pattern, which may signal an apnea event. The model assumes that
past values are a good indicator of what will happen next, making it useful for
analyzing time-dependent data. AR models work by expressing the current value of
the signal as a linear combination of its previous values plus a noise term. The order
of the model, which determines how many past values are used, is crucial for
accuracy and can be selected using criteria like the Akaike Information Criterion
(AIC) or Bayesian Information Criterion (BIC).

26
 In sleep apnea detection, AR models can be particularly useful for identifying
deviations from normal breathing patterns or heart rate variability. During an apnea
event, the model may detect significant changes, indicating a potential apnea episode.
One of the advantages of AR models is their simplicity and computational efficiency,
making them suitable for real-time monitoring systems. However, they assume that
the signal is stationary, which can be a limitation for non-stationary signals. To
improve accuracy, AR models can be combined with other techniques such as
machine learning algorithms, which can learn complex patterns from large datasets.
Additionally, preprocessing steps like detrending and differencing can help in
handling non-stationary signals.

3.2 Periodogram

A periodogram is a tool used to find out which frequencies are present in a signal and
how strong they are. It transforms the signal from the time domain into the frequency
domain, helping us see if there are any dominant frequencies that repeat over time. In
sleep apnea detection, a periodogram can analyze a person's breathing or heart rate to
find abnormal frequency patterns, like those caused by interrupted breathing.
However, it’s not always the best tool for complex signals because it can struggle to
clearly separate close frequencies or deal with rapidly changing signals. To overcome
these limitations, techniques like the Short-Time Fourier Transform (STFT) or
wavelet transforms can be used, which provide better time-frequency resolution and
are more effective for non-stationary signals. Moreover, combining periodograms
with machine learning algorithms can enhance the detection accuracy by learning
patterns from large datasets.

27
FIG 3.2.1

METHOD ADVANTAGES LIMITATIONS

Assumes Stationarity:
Autoregressive (AR) Good for Predictive Modeling: Performs poorly on non-
Models Useful for forecasting future stationary data where
values based on past data - statistical properties
Captures Temporal change over time –
Limited

Flexibility:
Dependencies: Effective in May not capture complex
modeling and identifying or non-linear
patterns in time-series data relationships

Requires Parameter
Simple and computationally Tuning: Choice of model
efficient: relatively order (number of lags)
straightforward to implement can affect performance
and interpret and needs careful
selection.
Periodogram Simple to Compute: Provides a Poor Frequency
straight-fo rward estimation Resolution: Limited
of the power spectral density ability to distinguish
of a signal closely spaced
frequencies.-

Useful for Identifying High Variance: noisy and

Dominant Frequency : Helps provide unreliable
in detecting periodic Information Can be
components in the signal estimates especially with
short signal lengths.-

28
 Widely applicable : can be No Time information on :
used in various domains for lacks the capability to
frequency analysis. analysis how frequency
content changes over time
.

Table 3.3 : Advantages and limitations of Existing methods.

CHAPTER-IV

PROPOSED METHODOLOGY

4.1 Introduction This project proposes a methodology for detecting sleep apnea
using ECG signals. The approach uses advanced signal processing techniques,
including bandpass filtering, feature extraction from frequency components, and
dynamic thresholding for apnea classification. The methodology aims to enhance the
accuracy of apnea detection while maintaining robustness across different subjects.
The following sections detail the various steps involved, supported by mathematical
analysis, a block diagram, and the performance metrics used for evaluation.

4.2 Methodology

4.2.1 Data Collection

The first step is collecting ECG signals from a reliable source, such as a publicly
available sleep apnea database (e.g., PhysioNet’s Apnea-ECG Database). This dataset
contains long-duration ECG recordings during sleep from both normal subjects and

29
individuals diagnosed with sleep apnea. These ECG signals serve as the input to the
system for analysis.

4.2.2 Preprocessing: Butterworth Bandpass Filtering

ECG signals often contain noise and unwanted components, such as baseline wander
and power-line interference. A Butterworth bandpass filter is used to remove these
elements while retaining the frequency components relevant to sleep apnea. The filter
is set with a low cutoff frequency to remove slow variations and a high cutoff
frequency to eliminate high-frequency noise.

1
H ( s )=

( )
2n
The Butterworth filter transfer function is given as: s
√ 1+
ωc

Where: H(s) is the filter transfer function,

ωc is the cutoff frequency,

n is the order of the filter.

For this project, the bandpass filter is designed with a range of 0.1 Hz to 30 Hz, which
effectively retains the key ECG signal components while removing irrelevant noise.

4.2.3 Feature Extraction

[Link] Power Spectral Density (PSD) via Fourier Transform

Once the signal is filtered, frequency-domain features are extracted using the Fourier
Transform. The Fourier Transform decomposes the signal into its frequency
components, allowing for the calculation of the Power Spectral Density (PSD), which
represents the power distribution of the signal over various frequencies.

The Fourier Transform is expressed as: X ( ω )= ∫ x ( t ) e

− jωt
dt
−∞

2
From this, the PSD can be calculated as: | X ( ω )|
The PSD provides essential information about the dominant frequency components of
the signal, which are used to detect abnormalities associated with sleep apnea.

30
[Link] Wavelet Coefficients via Wavelet Transform

In addition to the Fourier Transform, the wavelet transform is employed to capture

both time and frequency information from the ECG signal. The wavelet transform
decomposes the signal into various levels of frequency bands, capturing both high-
frequency and low-frequency components. This makes it particularly useful for
detecting transient events like apnea episodes.

( t−ba ) dt
The continuous wavelet transform is given by: W ( a , b )=∫ x ( t ) ψ∗
−∞

Where ψ is the mother wavelet, ( a) is the scale, and ( b ) is the translation parameter.
The extracted wavelet coefficients provide information about the signal's energy at
different frequency levels, which is then used for feature extraction.

4.2.4 Dynamic Thresholding

Once the relevant features (PSD from the Fourier Transform and wavelet coefficients
from the wavelet transform) are extracted, a dynamic threshold is applied to classify
the data. The threshold is adaptive and changes based on the characteristics of the
signal, making it sensitive to individual variations in the ECG signal.

By comparing the extracted features with the threshold, the system identifies whether
the signal segment indicates an apnea event or normal breathing.

4.2.5 Classification: Apnea vs. Non-Apnea

The final step in the methodology is to classify each segment of the ECG signal as
either an apnea event or non-apnea event. If the extracted features exceed the dynamic
threshold, the segment is classified as an apnea event. Otherwise, it is classified as
non-apnea. This binary classification is performed based on the extracted PSD and
wavelet coefficients.

4.3 Performance Metrics

To evaluate the effectiveness of the proposed methodology, the following

performance metrics are used:

31
Accuracy: Measures the overall correctness of the classification

TP+TN
Accuracy =
TP+ TN + FP+ FN

Sensitivity: Measures the ability of the system to correctly identify apnea events.
TP
Sensitivity =
TP+ FN

Specificity: Measures the ability of the system to correctly identify non-apnea events.
TN
Specificity =
TN + FP

Where:( TP ) is the number of true positive apnea classifications, ( TN ) is the number

of true negative non-apnea classifications, ( FP ) is the number of false positive apnea
classifications, ( FN ) is the number of false negative non-apnea classifications.

4.4 BLOCK DIAGRAM:

Figure 4.8.1 : Block diagram of proposed methodology

32
 CHAPTER-V

SOFTWARE DESCRIPTION
5.1 Introduction
In this project, use the software MATLAB to execute the required results. MATLAB
(matrix laboratory) is a fourth-generation high-level programming language and
interactive for numerical, visual and programming.

Matrix laboratory is developed by MathWorks. This MATLAB allows

● Matrix manipulations
● Plotting of functions and data
● Implementing of algorithms
● Creation of user interface

It has built-in commands and math functions which help in mathematical

calculations, numerical methods and generating plots. MATLAB has many
advantages compared to conventional computer languages (e.g., C, FORTRAN) for
solving technical problems. MATLAB is an interactive system whose basic data
element is an array that does not require dimensioning.

5.2 Basics of Software

5.2.1 Basic Building Blocks of MATLAB

The basic building block of MATLAB is MATRIX. The fundamental data type is the
array. Vectors, scalars, real matrices and complex matrices are handled as specific
classes of this basic data type. The built in functions are optimized for vector
operations. No dimension statements are required for vectors or arrays.

MATLAB Window
The MATLAB works based on five windows: Command window, Workspace
window, Current directory window, Command history window, Editor Window,
Graphics window and Online-help window.

33
● Command Window: In this command window it displays a command
prompt
“>>” and the cursor starts blinking where the commands can be entered and
executed. For example we try some arithmetic expressions.

Example 1 >> 20+(3*6)

ans = 38
Example 2 >>50/50*10
ans = 10

Fig 5.1: (a) Command Window

Workspace Window
Let us consider one example by initializing the value to two variables as shown in
Figure 5.1. (b).

34
 Fig 5.1: (b) Command Window Showing Values

As shown in the Figure5.1 (b). a is initialized one value and b also with some value. z
is having the value of a +b. Workspace is a collection of all variables.
a = 10 and b =20 z = a+b =30

Fig 5.2: Workspace Window

● Current Directory Window: The current Directory tab shows the contents
of the current directory, whose path is shown in the current directory window.
For example, in the windows operating system the path might be as follows:
C:\MATLAB\Work, indicating that directory “work” is a subdirectory of the
main directory “MATLAB”; which is installed in drive C.

35
● Command History Window: The Command History Window contains a
record of the commands a user has entered in the command window,
including both current and previous MATLAB sessions. Previously entered
MATLAB commands can be selected and re-executed from the command
history window by right clicking on a command or sequence of commands.
This is useful to select various options in addition to executing the commands
and is a useful feature when experimenting with various commands in a work
session.
● Editor Window: Editor Window is to create a new file and to edit the saved
page. We can create a new file and can also open an existing file here; the
files saved with the .m extension are called M-files.
● Graphics or Figure Window: The output of all graphic commands typed in
the command window is seen in this window.
● Online Help Window: MATLAB provides online help for all its built in
functions and programming language constructs. To know about any function
or the command click on the help icon so that we can easily find out the
command description. To check the proper command or function at a
particular line click on the help command and look for command.
5.2.2 MATLAB Files
● M-Files
These are standard text files with ‘m’ extension to the file name and creating their
own matrices using M-files, which are text files containing MATLAB code.
MATLAB editor or another text editor is used to create a file containing the same
statements which are typed at the MATLAB command line and save the file under a
name that ends in .m. There are two types of M-files:

● Script Files
It is an M-file with a set of MATLAB commands in it and is executed by typing the
name of the file on the command line. These files work on global variables currently
present in that environment.

● Function Files

36
A function file is also an M-file except that the variables in a function file are all
local.

This type of file begins with a function definition line.

5.2.3 MATLAB Commands

Required clc:
(Clear Command Window) This syntax clears all the text displaying in the Command
Window which clears the entire screen.

clear:

Here it removes all the variables and functions from the workspace. For example, if
we consider three variables of x, y, z initializing some values as p = 1; q = 2; r = 3;

clear p;

Only one variable p is cleared from the workspace and the remaining two variables q
and r displays in the workspace.

uigetfile: There is different syntax to get a single or multi file. This uigetfile displays
a dialog box for retrieving files. In this project after running the main file it displays a
dialog box to select the file. Here, this project is on underwater image enhancement so
we select an input image. After selecting an input image click on the open button.
[filename, pathname] = uigetfile (...);

imread: imread is initialized to read images from the file. There are
different syntaxes to read an image. Here we used a single image in
this project so to read that image we use syntax

inp = imread(fname);

[...] = imread (filename);

Some of the formats to read an image gif, jpg or jpeg, tif or tiff etc.

To read an image from an internet URL is to be mentioned in the syntax as

mentioned below

[...] = imread (URL,...);

37
 A = imread (filename, fmt);

Reads an image from the file specified and the format of the file is also mentioned in
the above syntax as imread (filename, fmt).

size:

To know the size of the image in array dimensions the syntax is written in M and N
as rows and columns.

[M N] = size (inp)

imshow:

To display the image imshow is used and the syntax is as

follows
imshow (inp) imshow(I)
mean:
The average of the array or mean value of the array. To get
the red channel image according to the formulae the mean of
green and red of the image is to be calculated.
mn_Ig = mean (mean (Ig)); mn_Ir = mean (mean (Ir));
mat2gray:

mat means matrix. Matrix to grayscale image where the

values of the image are stored in 0 (black) to white (1).

I = mat2gray (inp) inputdlg:

Create an open input dialog box.

title:

This is to add a title to the image.

title (‘sharpen image’);

title (‘input image’);
title (‘Red channel image’);
imwrite:

38
imwrite write image to graphics file. imwrite (a, filename, fmt) writes the image a to
the file specified by filename in the format specified by fmt. a can be an m-by-n
(grayscale image) or m-by-n-by-3 (color image) array, a cannot be an empty array. if
the format specified is tiff, imwrite can also accept an m-by-n-by-4 array containing
color data that uses the cmyk color space. imshow display image in handle
graphics figure. imshow (i) displays the gray scale image i.

sqrt:

Which gives the square root value for example if we initialize f = 25 then to get the
square root value of f then we have to use the syntax as

>>f = 25 f= 25

>>g = sqrt (f) g=5

end:
To terminate any block of the code then the end is used. If we use any of these for,
while, switch, if, and try then we end the block by using end syntax. Generally the
loops are used to check any conditions according to the program. To terminate the
block and go to the next line we close the block with the end.

5.2.4 Software Requirements

Coding Language : MATLAB

Version : MATLAB R2018a

Tool : Image processing toolbox

5.3 Design and Implementation

5.3.1 Introduction In this project our white-balancing aim at compensating for the
color cast caused by selective absorption of color with depth. Primarily by removing
the undesired color casting due to various illumination or medium attenuation
properties. Image fusion is to improve underwater images without restoring. Here the
results are executed in MATLAB software .Image processing toolbox is used to
perform analysis and algorithm development which perform image segmentation,
image enhancement and noise reduction.
5.3.2 System Architecture

39
In this project a single image is given as input image and our white-balancing
approach derived into two images one is the input 1 and input 2 as shown in the
Figure using gamma correction and edge sharpening and the two input images are
used as inputs of the fusion process. Multi-scale fusion approach is here to examine
with three levels by weight maps calculation.

5.3.3 Starting the MATLAB

Double click on MATLAB software

Fig 5.3: MATLAB Icon

5.3.4 Creating a New Project:
● To create a new file go to the Toolbar >>File>>New
● A window appears on the screen to create a new file name.
● Here all the files are stored with .m extension
Open the MATLAB layout here at the top of the layout a toolbar is present which
have many tools to create files, run the program, help, window, desktop, edit and
debug. To the left current folder is present which stores the present running files. To
the top right corner a workspace which is to display all defined variables with
memory allocation. Right bottom command history which records all commands and
in the middle of this command window is present which is used to generate the small
programs.

Advantages
● It is a case sensitive language
● MATLAB does not require a compiler to execute.
● It is an object oriented language.
Disadvantages
● It is very costly and the user has to buy each and every module.

40
● It is very difficult during cross compiling.
● It uses a large amount of memory.

5.4 Method of Implementation

Double click on the MATLAB icon after displaying the MATLAB layout on the
screen at the top of the layout toolbar is present. Go for new or if the file has to be
selected from the drive go to the file drive. Open the file in the current folder
displayed at the left of the MATLAB layout. The MATLAB layout is displayed in
Figure 5.3. (a).as shown.

● Click on open and select the source file from the drive. Select the source file and
click on the open button then the source file opens in the current folder. Click on
+ sign to see the files present in the folder. If a source file is to be selected or to
be added to the current folder click on file in the toolbar present at the top.
● Select the source file and click on the open button. As shown in the Figure 5.4.
(c).code file is selected and click on open.
● In the current folder it displays our code source file. Double click on the file it
displays the main file and the image which is going to be our input image.

● Click on the main file as shown in Figure 5.3. (d).

● Run the main file and it displays a dialog box to select the input file.

41
 Figure 5.4: (a) MATLAB Layout

Fig 5.4:(b) Browse the Source File

 Click on open and select the source file from the drive. Select the
source file and click on open button then the source files opens in the
current folder. Click on + sign to see the files present in the folder. If
a source file is to be selected or to be added to the current folder click
on file in the tool bar present at the top.

 Select the source file and click on open button. As shown in the
Figure5.4.(c).code file is selected and click on open.

Figure 5.4: (c) Select the Source File

 After clicking on code file the file is going to display in the current

42
 folder list.

Figure 5.4: (d) Main Source File

In current folder it displays our code source file. Double click on the file
it displays the main file and the image which is going to be our input
image.
 Click on main file as shown in the Figure5.3. (d). Run the main file it
displays a dialog box to select the input file

CHAPTER-VI

RESULTS AND ANALYSIS

The proposed method for sleep apnea detection was simulated using ECG signals,
yielding clear results at each stage of the process. After applying the Butterworth
Bandpass Filter, the ECG signals were significantly cleaner, with reduced noise and
artifacts. Fourier Transform analysis revealed distinct frequency changes during apnea
events, especially in lower frequency bands, highlighting relevant patterns in the
power spectral density (PSD). The Wavelet Transform provided further insights by
capturing time-frequency variations, particularly during apnea episodes, with
pronounced changes in wavelet coefficients. The use of dynamic thresholding allowed
for accurate classification of apnea and non-apnea events, effectively aligning with
real apnea occurrences in the signals. The overall simulation demonstrated the

43
capability of the system to detect sleep apnea through a combination of Fourier and
wavelet transforms, supported by dynamic thresholding.

6.1 Input signals

The ecg signals of two individuals, person 1 suffering with moderate apnea and
person 2 suffering with severe apnea are taken and plotted in MATLAB.

ECG Signal of person 1 with no apnea:

FIG: ECG signal of person 1

ECG Signal of person 2 with apnea:

FIG: ECG signal of person 2 with apnea

6.2 Preprocessing with Butterworth Bandpass filter:

44
The ECG signal was filtered using Butterworth Bandpass filter with cutoff
frequencies of 0.1Hz and 30Hz to remove noise and artifacts.

PERSON 1:

FIG: Filtered ECG signal of person 1

PERSON 2:

FIG:Filter
ed ECG signal of person 2 with apnea

45
6.3 Feature Extraction

6.3.1 Fourier Transform Analysis: The Fourier Transform was applied to calculate
the power spectral density (PSD) of the preprocessed ECG signals

Person 1:

FIG:

Person 2:

46
FIG:

6.3.2 Wavelet Transform Analysis

The Wavelet Transform was used to decompose the ECG signal ,capturing both time
and frequency information.

Person 1:

47
FIG:

48
PERSON 2:

FIG:

6.4 Detecting Sleep Apnea Events

Using the features extracted and setting threshold with the outliers code in MATLAB
the potential sleep apnea events in each signal are plotted.

PERSON 1:

49
FIG:

PERSON 2:

50
FIG: Signal with Apnea episodes highlighted

51
 CHAPTER- VII

ADVANTAGES & APPLICATIONS

7.1 Advantages

1. Non-invasive: non-invasive method that can predict cardiovascular and stress

conditions.

2. Cost-effective: This approach is more affordable compared to other complex

diagnostic tools, making it accessible for broader use.

3. Improved Accuracy: By analyzing both time and frequency domains, the system
enhances detection accuracy, identifying apnea events more precisely.

4. Continuous Monitoring: The method allows for long-term, continuous monitoring

of ECG signals, helping to capture intermittent apnea events throughout the night.

5. Real-time Detection: The system enables rapid detection, analyzing patterns and
signal features to flag apnea episodes as they occur.

6. Efficient Feature Extraction: Advanced signal analysis techniques extract critical

features, simplifying classification of sleep apnea and improving diagnostic
reliability.

7. Noise Reduction: The methodology is effective at filtering noise from ECG

signals, which ensures clearer data and more accurate results.

8. Dynamic Thresholding: Adapting thresholds based on signal patterns improves

sensitivity and reduces the occurrence of false positives.

9. Home Monitoring Potential: The technology is suitable for integration into

wearable or portable devices, enabling patients to monitor sleep apnea from home.

7.2 Applications

1. Clinical Diagnostics: Used in medical settings for the analysis of ECG data to
diagnose sleep apnea, improving patient outcomes.

52
2. Home-based Monitoring: This system can be integrated into wearable devices for
at-home monitoring, providing patients with convenience and continuous oversight.

3. Real-time Detection: Enables immediate detection and response to apnea events

during sleep, alerting users or healthcare providers.

4. Data Compression: The method supports efficient data compression, facilitating

long-term storage and transmission without loss of critical information.

5. Signal Denoising: Improves the quality of ECG signals by reducing interference,

allowing for clearer and more reliable analysis.

6. Research and Development: Continually advancing the accuracy and efficiency of

sleep apnea detection, contributing to the evolution of diagnostic tools.

7. Telemedicine: Supports the remote monitoring and transmission of processed ECG

data to healthcare professionals for expert review, broadening the reach of telehealth
services.

8. Remote Patient Monitoring: Enables continuous monitoring of patients remotely

for early detection of sleep apnea events.

9. Healthcare Resource Optimization: Contributes to the optimization of healthcare

resources by providing a cost-effective and accessible diagnostic tool for sleep apnea
detection.

53
 CHAPTER – VIII

CONCLUSION AND FUTURE SCOPE

In conclusion, the methodology for detecting sleep apnea using Fourier and wavelet
transforms applied to ECG signals demonstrates significant potential. By analyzing
both the time and frequency domains of the ECG signals, this system captures critical
features associated with apnea events, providing a robust and accurate detection
mechanism. The integration of these transforms, combined with continuous
monitoring of the heart’s electrical activity, ensures that apnea events are detected
effectively. This method not only improves detection accuracy but also offers a non-
invasive, cost-effective solution suitable for home-based and clinical applications.

Looking ahead, several avenues provide opportunities for advancing and expanding
the system for sleep apnea detection using ECG signals:

Firstly, ongoing research is needed to refine and enhance the accuracy of the
detection algorithms, optimizing the performance in diverse patient populations and
conditions. The incorporation of machine learning could further improve the
adaptability and precision of the system, allowing it to identify more subtle apnea
patterns.

Additionally, integrating the method with wearable technology presents a promising

future direction, enabling real-time, continuous monitoring of sleep patterns outside of
clinical settings. This could enhance early diagnosis and intervention, reducing the
risks associated with untreated sleep apnea. Telemedicine applications could further
broaden access to this technology, ensuring that remote patients can benefit from its
capabilities.

The integration of machine learning techniques represents a promising avenue for

further development in the detection of sleep apnea using ECG signals. By
incorporating machine learning, the system can adapt to variations in ECG patterns
and improve its accuracy over time by learning from larger datasets. This evolution
allows the system to discern subtle patterns that may indicate sleep apnea, which
traditional methods might miss. The ability to continuously refine and enhance the

54
model's performance ensures that the system can offer more precise and
individualized diagnoses as data accumulates.

Validation studies, including large-scale clinical trials, are essential to strengthen the
system’s reliability and demonstrate its applicability across diverse patient groups.
These trials would not only ensure that the system meets stringent regulatory
standards but also facilitate its integration into clinical practice. By proving the
system's effectiveness in a controlled setting, it opens the door for widespread
adoption, helping to ensure that the technology can benefit a broader population.

Another exciting area of future development lies in the exploration of personalized

medicine. By tailoring the detection system to individual patient profiles and unique
sleep patterns, the system could offer more accurate and personalized diagnoses. This
level of customization would enable healthcare providers to develop targeted
treatment plans based on a patient's specific needs, ultimately leading to better health
outcomes. The potential to optimize treatment strategies based on detailed individual
data presents a promising shift in how sleep apnea is managed.

As this technology advances, ensuring data security and privacy becomes crucial.
Remote monitoring and the transmission of sensitive health data require robust
safeguards to protect patient confidentiality and comply with legal standards.
Incorporating strong data protection measures will be vital to gaining trust from both
healthcare professionals and patients. Secure data handling will not only ensure
compliance with regulatory requirements but also foster confidence in the system's
reliability and integrity.

Collaboration with healthcare providers is a key element in the successful integration

of this technology into the healthcare system. Establishing strong partnerships with
healthcare professionals will facilitate the smooth adoption of the technology and
ensure that it aligns with established medical practices. This collaboration will ensure
that the system can be effectively incorporated into real-world clinical workflows,
enhancing accessibility and usability for both patients and clinicians.

In summary, the continued development of this ECG-based sleep apnea detection

system holds immense potential to transform sleep medicine. With advancements in

55
machine learning, rigorous validation, and secure data management, this methodology
stands to become a vital tool in the early detection and treatment of sleep apnea. Its
future integration into wearable devices and personalized healthcare solutions further
underscores its potential to significantly improve patient outcomes. By collaborating
with healthcare providers, this system will continue to evolve and contribute to the
broader field of healthcare technology, offering accessible, accurate, and personalized
solutions for sleep apnea patients.

REFERENCES
[1] Quiceno-Manrique et al. Detection of Obstructive sleep apnea in ECG recordings
using time-frequency distributions and dynamic features. IEEE international
conference on engineering in medicine And biology society (EMBC),(pg no:5559-
5562,2009)

[2] L. Hejjel and I. Gál Heart rate variability analysis Acta Physiologica Hungarica,
Volume 88 (3-4), pp. 219–230, 2001

[3] J.A. Fiz, et al Acoustic analysis of snoring sound in patients with simple snoring
and obstructive sleep apnea European Respiratory Journal, Vol. 9, pp. 2365–2370,
1996

[4] P. Janbakhshi and M.B. Shamsollahi Sleep Apnea Detection from Single-Lead
ECG Using Features Based on ECG-Derived Respiration (EDR) Signals IRBM, Vol.
39, pp. 206–218, 2018

[5] Margot Deviaeneet al. Sleep Apnea Detection Using Pulse Photoplethysmography
Computing in Cardiology 2018; Vol 45 Page 1

[6] Yuliya Zhivolupova Apnea MedAssist: Real-time Sleep Apnea Monitor Using
Single-Lead ECG IEEE Transactions on Information Technology in Biomedicine,
Vol. 15, No. 3, May 2

[7] Majdi Bsoul, et al. Sleep Apnea and Hypopnea Detection Algorithm Journal
2019 Ural Symposium on Biomedical Engineering, Radioelectronics and Information
Technology (USBEREIT), pp. 92-93

56
[8] Kaicheng Feng et al. A Sleep Apnea Detection Method Based on Unsupervised
Feature Learning and Single-Lead Electrocardiogram IEEE Transactions on
Instrumentation and Measurement, Vol. 70, 2021, 4000912

APPENDIX

SOURCE CODE

% Load the .mat file

data = load ("C:\Users\manasa p\Downloads\slp37m (1).mat");

% Check the variables in the loaded .mat file

disp('Variables in .mat file:');
disp(fieldnames(data));

% Extract the variable 'val' (change if needed based on actual variable names)
if isfield(data, 'val')
signal = [Link];
else
error ('Variable "val" not found in the .mat file.');
end
% Remove NaNs and Infinities from the signal
signal = signal(~isnan(signal) & isfinite(signal));

% Bandpass filter design

fs = 1000; % Sampling frequency
low_cutoff = 0.1; % Low cutoff frequency

57
high_cutoff = 30; % High cutoff frequency

% Filter the signal

[b, a] = butter(4, [low_cutoff high_cutoff] / (fs / 2), 'bandpass');
filtered_signal = filtfilt(b, a, signal);

% Remove NaNs and Infinities from filtered signal

filtered_signal = filtered_signal(~isnan(filtered_signal) & isfinite(filtered_signal));

% Plot original and filtered signal

figure;
subplot(2, 1, 1);
plot(signal);
title('Original Signal');
xlabel('Sample Number');
ylabel('Amplitude');

subplot (2, 1, 2);

plot(filtered_signal);
title ('Filtered Signal (Bandpass)');
xlabel('Sample Number');
ylabel('Amplitude');

% Fourier Transform
N = length(filtered_signal);
if N < 2
error ('Filtered signal is too short for FFT.');
end

58
f = (0:N-1) *(fs/N); % Frequency range
Y = fft(filtered_signal);
P2 = abs(Y/N); % Two-sided spectrum
P1 = P2(1:ceil(N/2) +1); % Single-sided spectrum
P1(2: end-1) = 2*P1(2: end-1); % Adjust amplitude
psd = (1/(fs*N)) * abs(Y). ^2; % Power Spectral Density

% Plot FFT
figure;
plot (f (1: ceil(N/2) +1), P1);
title ('Single-Sided Amplitude Spectrum of Filtered Signal');
xlabel('Frequency (Hz)');
ylabel('|P1(f)|');

% Wavelet Transform
if length(filtered_signal) > 0
[wt, f_wt] = cwt(filtered_signal, 'amor', fs); % Continuous Wavelet Transform
wavelet_coeff = abs(wt); % Magnitude of wavelet coefficients

% Time vector for plotting

t = (0: length(filtered_signal)-1)/fs;

% Plot Wavelet Transform

figure;
imagesc(t, f_wt, wavelet_coeff);
axis xy;
title ('Wavelet Transform (Magnitude)');
xlabel('Time (s)');
ylabel('Frequency (Hz)');

59
 colorbar;
else
warning ('Filtered signal is too short or contains invalid values for wavelet
transform.');
end

% Feature extraction
features = [mean(psd), std(psd), mean(wavelet_coeff(:)), max(wavelet_coeff(:))];

% Thresholds for classification (these should be determined based on your data)

threshold_psd = 0.1; % Example threshold for PSD
threshold_wavelet = 0.5; % Example threshold for wavelet coefficients

% Classification based on thresholds

classification_labels = zeros(1, length(filtered_signal));

% Sliding window approach for classifying apnea episodes

window_size = 5 * fs; % 5 seconds
step_size = 2 * fs; % Step size of 2 seconds
num_windows = floor((length(filtered_signal) - window_size) / step_size) + 1;

for i = 1:num_windows
start_idx = (i-1) *step_size + 1;
end_idx = start_idx + window_size - 1;
segment = filtered_signal(start_idx:end_idx);

% Fourier Transform and Wavelet Transform for the segment

Y_segment = fft(segment);
psd_segment = (1/(fs*length(segment))) * abs(Y_segment).^2;

60
 [wt_segment, ~] = cwt(segment, 'amor', fs);
wavelet_coeff_segment = abs(wt_segment);

% Feature extraction for the segment

features_segment = [mean(psd_segment), max(wavelet_coeff_segment(:))];

% Classification for the segment

if mean(psd_segment) > threshold_psd && max (wavelet_coeff_segment(:)) >
threshold_wavelet
classification_labels(start_idx:end_idx) = 1; % Mark segment as apnea
end
end

% Plot filtered signal and highlight apnea episodes

figure;
plot(filtered_signal);
hold on;

% Highlight apnea episodes

apnea_indices = find (classification_labels == 1);
plot(apnea_indices, filtered_signal(apnea_indices), 'r.', 'MarkerSize', 10); % Red dots
for apnea

title ('Filtered Signal with Apnea Episodes Highlighted');

xlabel('Sample Number');
ylabel('Amplitude');
legend ('Filtered Signal', 'Apnea Episodes');
hold off;

61
% Display classification summary
num_apnea_episodes = sum(classification_labels) / fs;
disp(['Total Apnea Duration: ', num2str(num_apnea_episodes), ' seconds']);

62