Abstracts

Rationale: Definitions of complexity are numerous and often incorporate different concepts. The behavior of several complexity measures derived from the matching pursuit analysis are investigated during the onset of partial seizures. To assess the reliability of these measures to detect partial seizure onset, the behaviors of these measures are compared over a large set of partial seizures recorded from patients undergoing monitoring with intracranial electrodes as part of presurgical evaluations. Methods: Seizures recorded from 45 consecutive patients with intracranial recordings arrays (subdural grids, strips, depth arrays) were analyzed for this study. Fifteen patients were diagnosed with MTLE, five with neocortical temporal onset epilepsy, fifteen with extra-temporal lobe epilepsy and eight could not be classified as either. Events included sub-clinical seizures and partial seizures (simple or complex) with or without secondary generalization. All epileptic events were analyzed with the matching pursuit time-frequency decomposition method selecting the channel closest to the seizure focus. Measures computed included entropy measures based on the distribution of the atom parameters (modulus, frequency, octave, phase), GAD complexity, power of the signal and the statistical complexity C. Average levels for each measure were computed over a 10s window before and after the electrographical seizure onset for 339 seizures. Automatic segmentation was used to determine the time of GAD level change associated with each epileptic seizure onset. Average levels for each measure were also computed over a 10s window using the time of initial GAD change. Percentage of change in levels between pre-ictal and ictal periods were calculated. Results: The segmentation method determined that 330 seizures (97%) have significant GAD changes associated with seizure onset. From all the measures, GAD has the most consistent behavior with the highest proportion of increase/decrease at seizure onset (319/11). Most of the entropy-based measures and statistical complexity show slightly lower proportions. Frequency-based entropy and two of the statistical complexity based measured show inverted proportions as those measures declined at seizure onset. Signal power is a poor identifier of seizure onset. Conclusions: Complexity as measured by GAD is a highly reliable method for identification of ictal events. This method allows retrospective review to identify subclinical or unmarked events in patients undergoing intracranial presurgical evaluations. Supported by NIH NS48222