A wavelet and teager energy operator based method for automatic detection of K-Complex in sleep EEG

In this study, an efficient algorithm is proposed for the automatic detection of K-complex from EEG recordings. First, the morphology of the K-complex had been examined and the detection features were determined according to visual recognition criterions of human scorer. These features were based on amplitude and duration properties of K-complex waveform. The algorithm is based on wavelet and teager energy operator and includes two main stages. Both results of stages were combined to make robust decision. G recordings obtained from the Sleep Research Laboratory in Department of Psychiatry at Gulhane Military Medical Academy. All night sleep EEG data, total 1045 epochs and 690 of these are NREM 2 stage, from 25 years old healthy female subject were used. Three scorers inspected recording separately to score K-complexes. The detection algorithm was then tested on the same recording. The results show that the agreements between the scorers were fairly different. The results are evaluated with the ROC analysis which proves up to 91% success in detecting the K-complex.