K-Complex Detection Based on Synchrosqueezing Transform

K-complex is an underlying pattern in the sleep EEG. Due to the role of sleep studies in neurophysiologic and cognitive disorders diagnosis, reliable methods for analysis and detection of this pattern are of great importance. In our previous work, Synchrosqueezing Transform (SST) was proposed for analysis of this pattern. SST is an EMD-like tool, which benefits from wavelet transform and reallocation approaches. This method is able to decompose signals into their time-varying oscillatory ingredients. In addition, it provides a time-frequency representation with less blurring compared to wavelet transform. In this paper, firstly, the ability of SST is investigated by applying the ANOVA test, which is approved by proper p-values. This paper proposes SST for K-complex detection. The proposed method is based on a so-called “detection of K-complexes and sleep spindles” (DETOKS) framework. DETOKS is based on spares optimization and decomposes signals into four components, namely transient, low frequency, oscillatory, and a residual. Applying the Teager-Kaiser energy operator and setting a threshold on the low-frequency component result in K-complex detection. We modify DETOKS using SST. The proposed method is applied to DREAMS dataset. The dataset provides two visual scorings accompanied by an automatic one. As the visual labels were extremely different, the automatic detection is considered as the third expert’s scoring and data is re-labeled by a voting approach among three experts. For DETOKS, DETOKS modified by CWT, and the proposed method, MCC measure is 0.62, 0.71, and 0.76, respectively. It shows superiority of the proposed method.