|
  |
|
|
Krasnow Institute > Monday Seminars > Abstracts Seizure Detection and Prediction Objective: We sought to identify the best algorithm for detecting the earliest dynamical changes in the electroencephalogram (EEG) leading to the seizures. Methods: This work compares results of seven linear and nonlinear methods in detecting the earliest dynamical changes preceding 12 intracranially recorded seizures from 4 children. Changes were identified by normalizing all measures by their standard deviation during a baseline interictal period, marking the first time the baseline maximum was exceeded. These times are compared to electrographic seizure onset times identified by an EEG board-certified neurologist. Results: All methods succeeded in identifying seizures 25-72 seconds before the neurologist-determined electrographic seizure onset for results averaged over all recording sites. Identification was improved by using data from a subset of "most active" electrodes, preceding that of the neurologist by an average of 100 seconds for phase analysis (n=12). Conclusions: The methods described here were successful in detecting changes leading to most seizures one to three minutes before the first changes noted by a neurologist. Although differences between detection times of linear and nonlinear methods were not significant, we note that phase correlation, a sensitive detector of linear and nonlinear synchronization, appeared slightly more sensitive than the other methods tested.
The Krasnow Institute for Advanced Study |
|
