English abstract
Abstract
Obstructive Sleep Apnea (OSA) is the cessation of breathing during sleep due to the collapse of upper airway. Apnea patients have increased risk of hypertension and heart failures. Polysomnographic recording is a conventional method for detection of OSA. Although it provides reliable results, it is expensive and cumbersome. Time domain and frequency domain analysis of heart rate variability (HRV) are new promising and non-invasive methods for detection of OSA. Many physiological activities are reflected as cyclic variations in RR interval (RRI) measures of HRV. Many studies have shown that OSA patients tend to have spectral peaks between 0.01 and 0.05 Hz of RRI spectrum. ECG records for 55 severe OSA patients and 35 normal subjects are acquired from Physionet website and from Department of Physiology in Sultan Qaboos University (SQU) Hospital. RRI data are extracted from these ECG records using QRS detection software provided by Physionet itself. The generated RRI data are then smoothed and filtered to remove false intervals and to substitute for missed intervals, as well as to screen the oscillations of OSA. Three types of RRI data groups are generated: non filtered RRI (RRI-NF), band pass filtered (RRI-F1) and high pass filtered (RRI-F2). Each RRI data group is analyzed independently. In this thesis, two new analysis methods are used to screen OSA and normal subjects. These methods are: Hilbert Transform followed by Statistical Signal Characterization (HTSSC) method and Sub-band decomposition (SB) method. HTSSC is a time domain analysis that manipulates RRI data in minute-by-minute basis and extracts 8 statistical parameters for each minute. The ability of these parameters to screen OSA is assessed by Receiver Operating Characteristics (ROC). Three of those parameters: mean of amplitude (MA), deviation of amplitude (DA) and maximum amplitude (MAXA), are found to produce the best range of accuracy (93-96%), specificity (93-100%), and sensitivity (87-100%) results. SB is a frequency domain analysis that is used to estimate the power spectral density (PSD) of RRI data without performing actual transformation to frequency domain, and to define the three bands of RRI spectrum: very low frequency (VLF), low frequency (LF) and high frequency (HF). The power ratios LF/HF and LF/VLF are then calculated for each RRI data group. The performance is also evaluate produced the best results when used with non-filtered RRI data. The maximum accuracy, specificity and sensitivity obtained by SB are: 93%, 94% and 98% respectively.
