Automatic detection of K-complexes: validation in normals and dysthymic patients

A model-based automatic K-complex (Kc) detector was applied to all-night single-channel sleep electroencephalographic (EEG) recordings from normal and dysthymic patients. The performance of the detector was analyzed in the two groups, and the differences obtained were discussed. The results showed that the detection rate of Kc in the normal group was around 92% through all stages of non-rapid eye movement (NREM) sleep, but with high numbers of "false" positives in stage 4 NREM, which reached 57%. In the dysthymic patients "true" detection included 85% of the Kc, but the percentage of "false" positives dropped to 25% in stage 4 NREM. Most of the "false" detections in the normal group were due to sharp delta activity during slow wave sleep (SWS). The results in the patient group were expected, because sleep in dysthymics showed a reduction in SWS when compared to normals. The behavior and automatic artifact rejection mechanisms of the detector are briefly presented. The model-based Kc detector performed significantly better than other automatic detectors described in the literature; it was found to be a useful tool for routine sleep EEG studies.