Electrocardiogram Signal Preprocessing Based on the SG-HP Filter for Sleep Apnea Detection Using Convolutional Neural Network (CNN)

Abstract

Sleep is one of the first behaviors that is disturbed due to the change of environmental conditions. Sometimes, those environmental discomforts that cause anxiety and disturbance or cause the failure of desires and failure to satisfy the basic needs of a person disturb his sleep. Sleep apnea is known as one of such disorders in scientific societies. So far, many researches have been conducted in this field, which are associated with advantages and disadvantages. In this article, an efficient method based on the use of Savitzky-Gulai (SG) and Hodrick–Prescott (HP) filters is presented in order to improve the electrocardiogram signal pre-processing for the accurate detection of sleep apnea. In this paper we use PhysioNet apnea-ECG database. The extracted features are subsequently used to train, test and validate a deep artificial neural network. The training and testing sets are obtained by randomly dividing the data until good performance is achieved using k-fold cross-validation (k=10). According to the results, the CNN classification has sufficient accuracy to detect and diagnose sleep apnea (99.1%), which proves the good performance of the proposed method.