Detection of Epileptic Seizures Based on Gabor Atom Density
Abstract number :
1.114
Submission category :
Year :
2001
Submission ID :
1992
Source :
www.aesnet.org
Presentation date :
12/1/2001 12:00:00 AM
Published date :
Dec 1, 2001, 06:00 AM
Authors :
C.C. Jouny, PhD, Department of Neurology, Johns Hopkins University School of Medicine, Baltimore, MD; P.J. Franaszczuk, PhD, Department of Neurology, Johns Hopkins University School of Medicine, Baltimore, MD; G.K. Bergey, MD, Department of Neurology, Joh
RATIONALE: The Matching Pursuit (MP) algorithm produces a time-frequency decomposition based on Gabor functions called atoms. We use a derived measure from this decomposition, which we have designated the atom density, to detect the occurrence of epileptic seizures in long-term recordings.
METHODS: The MP algorithm was applied to continuous EEG data recorded from intracranial electrodes in patients with intractable complex partial seizures. Decompositions were performed on signals recorded from single channels close to the seizure foci. As the decompositions based on the coherence criterion were not sensitive enough, we have introduced a new criterion based on the energy of residuals. Parameters for the analysis were set to obtain sensitive responses with the smallest computation time. As the dependence of the density with the energy threshold chosen is not linear, the maximum sensitivity with the best computational efficiency was found for the energy threshold slightly below the coherence limit of the interictal period. The number of atoms was normalized to the signal window length to obtain the atom density.
RESULTS: Preliminary results are based on seven seizures from three patients. The density of atoms allow us to mark the onset and the termination of the seizure. Complex partial seizures are invariably accompanied by a sharp increase in atom density compared to baseline levels obtained during interictal periods. The mean percentage of increase between the average baseline atom density level and maximum during the seizure is 580% ranging from 470 to 720%. This high atom density level is maintained throughout the seizure, with a pattern related to the different periods across the seizure. Analysis speed depends on the signal contents. We obtained factors ranging from 6 (1 hour to analyze 6h of data) to 10 on a standard computer (Athlon 800MHz) allowing scanning of long-term recordings. Analyses can be performed on-line. At the conclusion of a seizure, during the postictal period, the atom density drops below the preictal levels, and then returns to interictal values.
CONCLUSIONS: Increased Gabor atom density of EEG signals during seizures is a new criterion for detection methodology. It can also be used to distinguish between periods of artifacts and ictal periods. By using this new criterion based on residual energy, the matching pursuit decomposition could achieve better sensitivity. The atom density pattern during the seizure may also be used as a tool for further classification based on the atom density dynamic. How the Gabor atom density is related to process complexity remains to be elucidated.
Support: NIH grant NS 33732.