An Empirical Mode Decomposition-Based Method for Feature Extraction and Classification of Sleep Apnea

摘要

Background: Sleep apnea is a breathing disorder found among thirty percentage of the total population. Polysomnography (PSG) analysis is the standard method used for the identification of sleep apnea. Sleep laboratories are conducting this sleep test. Unavailability of sleep laboratories in rural areas makes the detection difficult for ordinary people. There are different methods for detecting sleep apnea. Past researches show that electrocardiogram-based detection is more accurate among other signals. This paper investigates the idea of electrocardiogram (ECG) signals for the recognition of sleep apnea. Methods: In this paper, the classification of healthy and apnea subjects is performed using electrocardiogram signals. The proper feature extraction from these signal segments is executed with the help of empirical mode decomposition (EMD). EMD algorithm decomposes the incoming signals into different intrinsic mode functions (IMFs). Four morphological features are extracted from these IMF levels. These features include the morphological characteristics of QRS complex, T and P waves. The classification of healthy and apnea subjects is done using the machine learning technique called support vector machine. Result: All the experiments are carried out by using St. Vincents University Hospital/University College Dublin Sleep Apnea Database (UCD database). This database is available online in physionet. It is observed from the results that by using empirical mode decomposition; it could be possible to extract the proper morphological features from this ECG segments. This technique also enhances the accuracy of the classifier. The overall sensitivity, specificity, and accuracy achieved for this proposed work are 90, 85, and 93.33%, respectively.