New Method For Detection Of Sleep Apnea Biology Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

In recent years, increase of patients with a sleep apnea syndrome has become a serious problem. Obstructive sleep apnea syndrome is the most common respiratory disturbance in humans. Many new diagnosis and treatment methods are constantly being proposed. EEG signal parameters, extracted and analysed are highly useful in diagnostics, and has become an important tool for detection of sleep apnea. This paper proposes a new method for detecting sleep apnoea by separating EEG signals into its different frequency components by performing power spectral analysis. A comparative study of these components are done and fed to the artificial neural network as a training input. Furthermore, data from various other subjects are also fed to the neural network for detection of SAS.


Sleep is a circadian rhythm essential for human life. Good sleep is an essential precondition to the maintenance of mental and physical health. Sleep apnoea is a common sleep disorder characterized by brief interruptions of breathing during sleep. It is defined as sleep-disordered breathing distinguished by recurrent episodes of upper airway collapse during sleep[9]. These breathing pauses typically last between 10 to 20 seconds and can occur up to hundreds of times a night causing changes in cardiac and neuronal activity and discontinuities in sleep pattern[10]. When people enter middle age and beyond, the upper respiratory tracts of some parts shrink, which may lead to obstruction of nasal passage and snoring during sleep. This apnea may affect the quality of sleep and health when it occurs frequently, and may even cause death in severe cases [8]. The different types of sleep apnea syndromes are: Obstructive sleep apnea, Central sleep apnea and Complex sleep apnea. Obstructive sleep apnea occurs when the soft tissue in the back of your throat relaxes during sleep, causing a blockage of the airway (as well as loud snoring)[9], Central sleep apnea involves the central nervous system, rather than an airway obstruction. It occurs when the brain fails to signal the muscles that control breathing. Complex sleep apnea is a combination of obstructive sleep apnea and central sleep apnea [6]. The most common type of sleep apnoea is Obstructive Sleep apnea syndrome (OSA). The onset of sleep is typically characterized by gradual changes in cortical electroencephalographic (EEG) activity. EEG is the recording of electrical activity along the scalp and is typically described in terms of (1) rhythmic activity and (2) transients. This rhythmic activity is divided into frequency and amplitude bands- Delta waves (frequency up to 4Hz and amplitude 20-100uv), Theta waves (frequency 4->8Hz and amplitude 10uv), Alpha waves (frequency 8-13Hz and amplitude 2-100uv), Beta waves (frequency >13-30Hz and amplitude 5-10uv), Gamma waves (frequency 30-100Hz), Mu (also 4-8Hz).

The main objective of this work is to explore various possible relationships among sleep stages and apneic events and improve on the clinical accuracy of algorithms for sleep classification and apnea detection. Studies have shown that there are changes in cortical activity that occur during sleep disordered breathing (SDB) events like sleep apnoea. EEG signals will be assessed using advanced signal processing approaches in which EEG signal is segregated into different frequency bands-delta ,alpha, theta and gamma. These parameters are used for the detection of sleep apnea with the help of artificial neural network. Therefore, to devise a more economical method, we focus mainly on the sole ability of cortical EEG for the detection of sleep apnoea.


The normal human EEG is observed to show activity over the range of 1-30 Hz with amplitudes in the range of 20-100 µV. The surface EEG shows typical patterns of activity that can be correlated with various stages of sleep and wakefulness.

Sleep is not a uniform state, but is characterized by a cyclic alternating pattern of non-rapid eye movement (REM) and REM sleep. Non-REM is divided into four stages of sleep:

A person falling asleep is observed to first enter stage 1 characterized by low-amplitude, high frequency(6-8 Hz) EEG activity. Alpha rhythms are more predominant in this state of drowsiness. Sleep 2 stage is the light sleep state, where the eye movements stop and our brain waves become slower. Special waves 'K-complexes and sleep spindles begin to appear. In this state, EEG amplitude is medium ,of about 10-50 µV and EEG frequency is 4-7Hz.In stage 3 (deep sleep), extremely slow brain waves called delta waves begin to appear, interspersed with smaller, faster waves. By stage 4 (deep sleep, slow wave sleep), the brain produces delta waves. In stage 4, the amplitude of EEG will be high, but the frequency will be less than 2 Hz.

REM sleep is observed as rapid, low voltage, irregular EEG activity associated with muscle twitches and rapid eye movements. Theta wave is more predominant in this sleep stage.


Some physiological observations of sleep apnea patients during apneic events have been recorded as follows :

During each episode of obstruction there is a decrease in oxygen saturation which is sometimes accompanied by a slowing heart rate. At the end of the episode the EEG is said to show a brief (3-10 seconds) burst of alpha activity, the electromyogram (EMG) is elevated and the heart rate is accelerated. After this, breathing resumes and oxygen saturation returns to the level of wakefulness. This pattern occurs recurrently throughout the night which results in sleep fragmentation and hence the disorder is associated with daytime symptoms, most often excessive sleepiness. This also results in cognitive decline, memory loss etc[10].

Full polysomnography (PSG), which measures sleep, respiratory variables and oxygenation, is still regarded as the standard method to diagnose Sleep Apnoea Syndrome. Each patient undergoes a single night of PSG test using standard techniques in which Central and frontal EEG (C3-C4, CZ-PZ, F3-FP1, F4-FP2), two electro-oculograms (EOG), submental and left and right tibial electromyograms (EMG) and the electrocardiogram(ECG) are monitored. Respiratory variables monitored include nasal airflow, oral airflow measured by thermistor, thoraxo-abdominal movement by inductance plethysmography and oxygen saturation (SaO2). At the end of the overnight sleep study a sleep specialist scores the polygraph recording, identifying sleep stages and apnoeic and hypoapnoeic events causing oxygen desaturation.

From this analysis an Apnoea Hypopnea Index (AHI) is calculated for the patient:

Apnoea-Hypoapnea Index: This index is used to measure the severity of the sleep apnoea. The average number of apneas and hypopneas during one hour of sleep is called the apnea/hypopnea index (AHI) or respiratory disturbance index (RDI). It measures the frequency of reductions in airflow associated with upper-airway collapse or narrowing that occurs with the state change from wakefulness to sleep. Apnoea is complete cessation of airflow for 10s or more. Hypoapnoea is associated with a decrease of respiratory volume by 50% for more than 10 seconds. A patient having an AHI between 5 and 15 is said to have mild Obstructive Sleep Apnoea (OSA), whereas 15 to 30 is moderate and more than 30 events per hour is diagnosed as having severe sleep apnea. These classifications also depend on factors such as sleepiness etc.


Many researchers have various techniques including unconventional approaches such as engineering diagnostic techniques, for determining patient's condition. A review of literature includes, rapid electroencephalographic changes in response to cerebral anoxia were observed as early as 1925. Since then many researches has been conducted on detection of sleep apnoea by EEG signal processing involving various methods and algorithms. Further on [1] Guilleminault et al. (1996) reported a delta band amplitude increase starting on average 13 seconds after the onset of apnea. During NREM, the average differences between initial and maximal values were found to be 268% and 202% between initial and final values during the event duration. [2] Berry et al. (1998) further studied the variation in delta power and reported a cyclic increase in delta power which was in synchronization with increased respiratory effort in NREM sleep. [3] Baumgaurt-Schmitt et al. (1998) have used neural network to classify the various sleep stages by extracting the features from the genetic algorithms. Here, outputs of nine different networks were created by using data of 9 different subjects which were used simultaneously for classification. [4] Dingli et al. (2002) have shown the spectral analysis technique for the detection of cortical activity changes in sleep apnoea subjects. The most consistent significant change is the decrease in theta power, during Non-Rapid Eye Movement (NREM) sleep which is either associated with an increase in high frequencies (alpha and sigma) or with delta increase. [5] Lin et al. (2006) have implemented a new technique for classification and analysis of EEG signals, by using wavelet transforms and then feeding the spectral components to the inputs of an artificial neural network for recognition of EEG signal characteristics of Sleep Apnoea Syndrome which was configured to give three outputs to signify the SAS situation of the patient. The recognition threshold for all test signals turned out to have sensitivity level of approximately 69.64 and a specificity value of approximately. [6] T.Sugi et al. (2009) proposed the method for automatic detection of EEG arousals in SAS patients . To effectively detect respiratory-related arousals, threshold values were determined according to pathological events. The proposed method was applied to Polysomnographic (PSG) records of eight patients with SAS and accuracy of EEG arousal detection was verified by comparative visual inspection. [7] Chein-Chang Hsu et al. (2010) used the Hilbert-Huang transformation for extracting the frequency element from Hilbert spectrum. The main contribution of the system is to preserve time information in the EEG by Hilbert-Huang transformation mechanism as well as find frequency variation information. [8] Boyu Wang et al. (2009) has performed the different off-line methods for two different EEG signal classification task-motor imaginary and finger movement.


Sleep apnoea is currently diagnosed by the analysis of clinical polysomnography test which measures EOG, EMG, to identify various leg movements , ECG, to document cardiac arrhythmias, EEG, to document various sleep states, nasal/oral airflow, SaO2(oxygen desaturation ) and thoracic movement. EEG signals being non-stationary, non- invasive are quite useful since they give off no radiation and can be recorded over a long interval of time. Here, we mainly focus on analysis of EEG signals for detection of sleep apnoea.


EEG sleep data (in .txt format) has been obtained from sleep database (MIT-BIH Polysomnography database). Data from five subjects are taken, among which all are male, age ranging from 32-56 years and their weights ranging from 89-152 kgs . These records were analysed over 30 seconds window for a period of 1 hour. Following the 10-20 System of electrode configuration, the electrode placement for the EEG signal was either C4- A1 or C3-O2. The signal from the data has been sampled at a frequency of 250 Hz. The algorithm starts with acquisition of EEG data of multiple subjects which has been processed to generate an artifact and interference free signal.


The data having sleep apnoea events, acquired from database ( is processed in MATLAB. We load EEG data taken from one subject (slpdb59), in MATLAB and plot the signal. Next, we take 30 seconds windows for a period of 1 hour and analyse them using Power Spectral Density, which is done by using Welch method of averaged periodograms, using Hamming window with 64 points overlap(50% overlap). This was performed by creating a Welch object in MATLAB by using the function spectrum.welch and applying it in the function psd written in MATLAB. This results in an average power spectral density curve (µV2 /Hz) that lies over the frequencies ± π. Thus, the data is converted to its frequency domain which is sampled at N equidistant samples, which is given by the power of two, having value greater than the length of the signal. PSD is estimated separately for all the windows, simultaneously segregating the signal into its delta, theta, alpha, and beta frequency bands. The power spectral density corresponding to respective frequency bands is calculated and their power ratios are calculated by dividing average power of individual sleep wave frequency band by the total average power of all the bands. The total average power of each frequency band is calculated by estimating Power spectral analysis over 30s window for the entire signal, and thus taking mean of the individual frequency bands of all the windows. Now we compare the window containing sleep apnoeic event with the non-event sleep window and note the differences between the power ratios of the respective frequency bands such as Delta, Theta, Alpha and Beta . This procedure is repeated for the entire signal duration by taking 30 s windows, thus distinguishing between apnoeic and non-apnoeic regions

Now, that we have an established a threshold, we feed the data to an artificial neural network, which adapts and learns by itself, and then use this information for the detection of sleep apnoea for more subjects or data(in this case five).


The power spectral density is performed by creating a welch object in MATLAB, and taking N point FFT for power spectrum estimation. For calculating N points, we suppose that the signal representation in the frequency domain is X(ω) and is periodic with a period 2π. Considering the frequency domain is sampled at N equidistant samples, the interval between two successive samples is δω. This is written mathematically as:


Therefore, the estimated power spectrum using Welch method is given by:

At frequencies fn = n/N where n= 0,1,...,N-1


By the end of this endeavour we expect to accomplish newer methods for the detection of sleep apnea by detailed study and analysis of the parameters- Delta, Theta, Beta and Gamma, which offers a clinical reference value for identifying Sleep Apnoea Syndrome, thus reducing diagnosis time and improving medical service efficiency.


In this paper we have tried to find a new solution to the problem of identifying Sleep Apnoea Syndrome (SAS) episodes. In order to achieve this goal, the EEG signal characteristics of SAS episodes were extracted using power spectral density and then detected using an Artificial Neural Network.