Detection of the EEG Artifacts by the Means of the (Extended) Kalman Filter

This paper presents a new approach for detection of artifacts in sleep electroencephalogram (EEG) recordings. The proposed approach is based on Kalman filter. The idea of this approach consist in embedding the AR model into the Kalman Filter which makes possible to use such KF AR (Kalman Filter AR) models for linear prediction of non-stationary signals. Such model can be set up to detect and follow discrete dynamic changes of the signal. For detection of the EEG artifacts we have exploited the evolution of the state noise - increase in state noise indicate the dynamic change of the signal. The evaluation of the results was done by the Receiver-Operator Characteristics (ROC) curves in terms of the specificity and the sensitivity. For 90% of the specificity the best achieved value of the sensitivity using KF AR model was 33%. In order to achieve better results we have tried the following modification: instead of the Kalman Filter we have used extended Kalman Filter and instead of the AR model a neural network. The preliminary results look promissing: for 90% of the specificity we have achieved 65% of the sensitivity.