Abstracts

Over the past three decades, many different linear and nonlinear measures characterizing EEG and ECoG signals have been tested for their ability to detect or anticipate seizures. However, few studies have reported on the performance of these metrics in pediatric patients over extended epochs ([gt] several hours); this study aims to evaluate several linear (largest eigenvalue, EI; power, POW) and nonlinear (correlation dimension, CD; Kolmogorov entropy, KE) measures of scalp EEG and ECoG in a continuously monitored pediatric population with both temporal and extra-temporal epilepsy., 36-72 hour recordings, consisting of 96-128 channel EEG and ECoG, were analyzed from four successive surgical patients (ages 13 mo, 12 yr, 13 yr, and 21 yr) at the University of Chicago. Each channel was sampled at 400 Hz and filtered between 1-35 Hz, after which the metrics were calculated from 20s windows with a 10 s overlap. The resulting time-series of metrics were compared against events marked by a board certified clinical neurophysiologist or registered EEG technician; the researchers and the human experts were blind to each others[apos] findings until the results were compared. A Receiver Operator Curve-like thresholding was used to assess how well the metrics detected seizures, pre- and post-seizure intervals, and other events noted by the human experts., Overall the metrics showed cyclic behavior as well as sudden changes during epileptiform activity. In one patient, seizures occurred only when EI had cycled below a long-term average (for 6/6 seizures recorded over 2 days). In addition, EI identified a distinct 3-10 min post-ictal period in 3 patients. KE also showed consistent post-ictal behavior in 2 of the patients.
In 2 patients, at least one metric, averaged over the scalp leads, achieved 100% sensitivity (SENS: the fraction of seizures that was detected) and a false detection rate (FDR: the number of falsely detected events/seizures per hour) [le]1/hr; this was true for 3 patients using averaged intracranial metrics.
For individual channels, only EI and KE had SENS=1.0 and FDR[le]1/hr for a significant number of scalp leads ([ge]25% in 2 patients); POW satisfied this for only 1 patient, while CD generally did poorly for almost all scalp leads. None of the metrics was effective at detecting seizures with 100% sensitivity and an acceptable FDR in the 13 mo old., 1- No single metric outperformed the others across patients and generally the metrics calculated from ECoG performed better than those from the scalp EEG.
2- For a given sensitivity, EI and KE generally had lower FDR than POW or CD
3- Consistent with previous studies, the results suggest that the usefulness of a metric is patient dependent.
4- Development may play a critical role: the results for the toddler were much worse that the other patients.
5- EI, and to a lesser extent KE, indicates a 3-10 min post-ictal period exists., (Supported by Falk Foundation and the Linn Family.)