Abstracts

A number of laboratories have recently reported success in the application of non-linear dynamic systems analysis to the detection of clinical seizures in humans. To date, most of these studies are a) performed in adults, b) with temporal lobe seizures and c) small series of patients. While the results of these studies are generally positive, there has been limited discussion of the circumstances in which these metrics fail, virtually no studies limited to the pediatric population, or specifically examine pediatric neocortical epilepsies. We reports our preliminary findings in a group of more than 100 children with seizures.
Samples were obtained from long term video-EEG recordings of patients admitted to the pediatric epilepsy service at the University of Chicago, under an IRB approved protocol. Three, 30 second consecutively selected seizure epochs and three awake and sleep non-seizure epochs were analyzed for each patient. Segments were selected on the basis of the first thirty-second artifact free segment in the record. Epochs were transferred to a signal processing package [ndash] RRChaos[ndash] where Eigenvalues, Kolmogorov Entropy (KE), two forms of Correlation Dimension (CD)and a global measure of non-linearity (Z) were performed. These values were then compared across patients by: age, seizure types, seizure location, and seizure versus non-seizure conditions.
Over 70 percent of the seizures showed a clear difference in nonlinear measures between the seizure and non-seizure conditions. Clinical seizure types showed similar changes in these measures within patients. Unlike, most published reports of KE which have demonstrated KE declines with seizure onset, we found, that KE may decline or increase with seizure onset. The likelihood of an increase is higher for patients with a low baseline KE. In addition we identified a group of patients in which there were no changes in KE, virtually all of whom had very low baseline KE ([lt]20bits/s). Low KE alone did not exclude the detection of seizures. Seizures in a number of premature infants reduced the KE from [lt]20 to [lt]10 bits/s. CD was a less sensitive metric, but the same trends of increases or decreases in the CD with seizure onset were observed. The most stable metric, the Eigenvalue, at times demonstrated changes when the KE failed, no single metric detected all seizures.
Dynamical systems measures are effective at separating seizure from non-seizure states in children with neocortical epilepsies. The use of absolute values for significant changes do not appear as sensitive as measures simply based on dynamic change. Low baseline KE was the most common reason for failure to detect seizure onset. Our data suggest that it is possible to identify, in advance, most of the patients for whom these measures will fail to detect seizure onset, thus proving a method to decrease the false negative rate of seizure detection.
[Supported by: Falk Foundation]