Detection of stationary segments in interictal electroencephalographic epileptic recordings.

The electroencephalographic (EEG) recording is frequently used to acquire both ictal and interictal epileptiform abnormalities. The objective of this paper is to detect the stationary parts of the interictal EEG signals. This is achieved by means of a lattice filter based on an optimal orthogonal linear prediction algorithm. It calculates recursively a set of reflection coefficients and a likelihood ratio test built on them is applied by means of a threshold linked to its statistical properties. The method is applied on three types of interictal recordings: signals with single spikes, sequences of spikes and signals with spikes and slow waves with comparable amplitudes. The best results were pointed out for the first and the last types. When dealing with spike-sequences, the algorithm reached its lowest rate of success. The proposed method permits an adequate segmentation and therefore facilitates the automatic interpretation of EEG before epileptic seizures.