Detection of Sleep Apnea Disorder Using an Adaptive Neuro-Fuzzy Classification Method

Abstract

The proposed work aims at developing a solution for the detection of sleep apnea disorder using ECG signal analysis, which is an established diagnostic modality. Under this work, the standard research resource, ECG-Apnea database from MIT’s Physionet.org., having ECG signal night time recordings, is used. The sequential procedure of Preprocessing, Peak or QRS complex detection, Feature extraction, Feature reduction, and Classification is used. Preprocessing of the ECG signal is performed to free it from noise resulted from baseline wander, power-line interference, and muscle artifacts. Thus, the improved signal quality is estimated in terms of its Signal to Noise Ratio (SNR) and entropy value. QRS detection is implemented using the popular Pan-Tompkins algorithm that provides the reference for the feature extraction process. The performance of the detection algorithm is measured in terms of the average values of accuracy and specificity as 98% and 96%, respectively. Feature extraction algorithm involves the collection of selected 30 feature values related to the time domain and the frequency domain gathered from each of the test recordings of the ECG database, minute-wise for 7 hours. Feature reduction technique is followed to reduce the data size to a set of 20 ECG signal features using Principal Component Analysis (PCA) and avoid redundancy. Hence the trained Adaptive Neuro-Fuzzy Classifier is used on the output feature set derived from PCA to detect the presence or absence of Sleep apnea disorder with an estimated accuracy and specificity as 95% and 96%, respectively.