A new approach towards predictability of epileptic seizures: KLT dimension
Venugopal, R.; Narayanan, K.; Prasad, A.; Spanias, A.; Sackellares, J.C.; Iasemidis, L.D.
Biomedical Sciences Instrumentation 39: 123-128
2003
ISSN/ISBN: 0067-8856 PMID: 12724880 Document Number: 563778
This paper proposes a measure of complexity of the epileptic electroencephalogram (EEG) based on the dimensionality of the Karhunen-Loeve Transform (KLT) in the time domain. We estimate the KLT dimensionality by assuming the same observation noise level in the EEG during the interictal period (between the seizures) as the one during an epileptic seizure (ictal period). Utilizing an optimality criterion based on the T-index [1] and the predictability time, derived from the created KLT dimensionality profiles, we show that 10 out of 15 seizures in one patient with temporal lobe epilepsy were predictable with an average predictability time of about 36 minutes.