Detecting K-Complexes in Brain Signals Using WSST2-DETOKS

Sleep studies play an underlying role in diagnosing neurophysiologic and cognitive disorders. K-complex is an important pattern in sleep EEG. Therefore, reliable methods for detection and analysis of this pattern are needed. Detection of K-complexes and sleep spindles (DETOKS) is a powerful method which benefits from modeling EEG as three components including transient, low frequency and oscillatory, in addition to a fast nonlinear optimization algorithm to estimate them. In this chapter, we propose two modifications on DETOKS using wavelet-based synchro-squeezing transform (WSST), and second order WSST (WSST2) called WSST-DETOKS and WSST2-DETOKS, respectively. The WSST is an EMD-like approach which is a mixture of wavelet transform and reallocation approaches. The WSST decomposes signals into their time-varying oscillatory ingredients. It provides a time-frequency representation with less blurring compared to wavelet transform. Our proposed methods are applied to a DREAMS dataset. Two different visual scorings are available for this dataset. To address the issue, we propose using an automatic scoring, which is available, as the third expert’s annotation. Therefore, data is re-labeled using a voting approach. Results declare that WSST2-DETOKS with MCC measure 0.86; F1 and F2 scores of 0.85 and 0.92, respectively, outperforms the standard DETOKS as well as the CWT-DETOKS and WSST-DETOKS.