Classification of sleep apnea using cross wavelet transform

In this paper, a novel approach for classifying sleep apneas using cross wavelet transform has been proposed. This is the first time that cross wavelet transform has ever been applied to sleep apnea type classification. The developed method takes the airflow and thoracic effort signals, as an in-put, which are then transformed to time-frequency and phase plane in order to extract the information of correlation between the two signals during different apnea condition. As the cross-wavelet returns large number of coefficients, which may be difficult to handle in some automated detection system, therefore dimension reduction was necessary. In the work, kernel principal component analysis (KPCA) based dimension reduction technique has been applied, and four Eigen values from each of the cross-wavelet amplitude and phase coefficients found to be effective for detection of apnea into three categories i.e., obstructive, central and mixed. The proposed system has been tested on the recordings obtained from 23 subjects. The average classification rate obtained using simple threshold technique was 85% ± 0.78%, and the values for each class were 85.2% (obstructive), 86.4% (central) and 83.6% (mixed). The results show that cross-wavelet is useful in order to distinguish the apneas, as it looks into the phase and amplitude coherence between the two signals.