Relation Between Singular Values And GraphDimensions Of Deterministic EpileptiformEEG Signals

Computerised detection and prediction of epileptic discharges from EEG data is a problem whose solution may lead to the prediction of epileptic seizures and planning of treatment. The recently confirmed fact that the EEG has a fractal nature enables a new approach to analysis of epilepsy. A conventional signal processing approach is not appropriate for highly complex signals, such as chaotic deterministic signals including epileptiform EEG. In our previous work we found that the graph dimension is the most appropriate measure for real-time fractal dimension estimation of EEG signals, and that it could be used for differentiation between parts of EEG signals with and without epileptic discharges. However, we also found that this measure is especially sensitive to signal artefacts and signal noise. In this paper we present the results of our search for a more robust measure for differentiation. The mean singular value is our new quantitative measure which in combination with the graph dimension can be used for differentiation of different EEG states (with and without epileptic discharges).