In this chapter, the authors use various signal processing techniques to analyze and gain insights on how ECG signals for patients suffering from sleep apnea (sleep apnea or obstructive sleep apnea occurs when the muscles that support the soft tissues in the throat, such as tongue and soft palate, relax temporarily) disease vary with respect to a normal person's ECG. The work has three stages: firstly, to identify waves, complexes, morphology in an ECG which reflect the presence of the disease; second, feature extraction techniques to extract features of ECG such as duration of the wave, amplitude distribution, and morphology classes; and third, detailed clustering (unsupervised) algorithm analysis of the extracted features with efficient feature reduction methodologies such as PCA and LDA. Finally, the authors use supervised machine learning algorithms (SVM, naive Bayes classifier, feed forward neural network, and decision tree) to distinguish between ECG signals with sleep apnea and normal ECG signals.
TopIntroduction
Sleep apnea is caused by collapse of the pharyngeal airway during sleep and is characterized by apneoic episodes during sleep, where the person will stop breathing periodically for up to a few minutes. The presentation of sleep apnea results in snoring, morning headache and the key feature is daytime sleepiness and concentration problem (Osman, 2018).On an average 10% of the whole world’s population is facing this problem (Dempsey, 2010).Untreated sleep apnea can increase the risk of heart attack, stroke high blood pressure and one study even showed that it may reduce 20 years off patients life. Stages of sleep apnea is measured using a index called Apnea-Hypopnea Index or AHI, when AHI is between 5-15 it is termed as mild sleep apnea, when it is between 15-30 it is called moderate sleep apnea and when it is above 30 it is severe sleep apnea state. Medical recommendations will be based on the AHI number.
Traditionally sleep apnea diagnosis is done by a technique called polysomnography (PSG). PSG is one of the most common tests used for apnea detection. It analyses physiological signals of human beings like airflow, graph (EEG), ECG and metabolic process signals during sleep. Electrodes and sensors are connected to patients' bodies and they are made to sleep in a particular posture for more than a night in the sleep laboratory under the supervision of technicians. PSG is calculated by the number of sleep apnea events per hour. It is clear that PSG is a time taking process and also expensive. In order to make it time and cost effective researchers have found some alternative techniques to attain high efficiency and flexibility compared to traditional methods.
The alternative research suggests some mechanisms in order to detect sleep apnea other than PSG which focuses on ECG, EEG signals and snoring patterns of patients. Considering that ECG signals are the most commonly used for medical diagnosis because of their accurate performance in terms of feature extraction. Figure 1 depicts a complete ECG beat consisting of P, Q, R, S and T waves wherein R-R interval and T wave area and amplitude are the primary features, which are extracted from ECG signals. These features are given as inputs to different machine learning classifier models so as to detect sleep apnea.
Most of the ECG-based sleep apnea detection algorithms in the literature have used different parameters derived from the raw ECG signal where the raw ECG signals are taken from physionet.org which is open-source software for biomedical signals. Features are extracted by measurements of P-wave, QRS complexes and T-wave. Where each QRS complex is converted into a Fourier spectrum from ECG signals, classification is achieved by Grey relational analysis of identifying the amplitude and frequency characteristics of the transmission signal accurately under a strong vibration environment (Schrader et al., 2000). In some papers different machine-learning methods were compared to find an optimal solution to the problem of OSA classification. Their results showed the top three classifiers (SVM, DT, and LDA) were compared on the basis of accuracy. Top three classifiers i.e. SVM, DT and LDA showed strong performance, they could be effective on sleep studies and OSA detections are explained (Urtnasan et al., 2017).
The work we concentrate on extraction of time-domain and frequency-domain features from the Holder’s coefficients, further the extracted features are given to six different classifiers: SVM (Support Vector Machine), KNN (k-Nearest Neighbor), SGD (Stochastic Gradient Descent), Naive Bayes Classifier. Comparison of different classification models with respect to analyzing factors such as precision, recall and f1 score for training and test data sets has been carried out.