Detection of K-complexes in sleep EEG using CD-HMM

In the current paper a new approach for K-complex detection using a continuous density hidden Markov model (CD-HMM) is presented. The system performance was evaluated in two manners. First using three seconds long segments of K-complexes and of background EEG (classification problem). Second using a whole night record and detecting the K complexes (detection problem). The fist test achieved an equal error rate of 7%. In the second test the system performance was compared to four trained scores that scored the signal independently. The performance of the algorithm was within the variance of the human scorers.