Abstracts

Rationale: Detection of seizure activity, or pre-ictal processes that may lead to seizure, is of great interest to both clinicians and researchers. As such, many different measures characterizing EEG have been tested for their ability to detect or anticipate seizures. Studies have generally examined relatively short (up to several hours) epochs from adults with mesial temporal lobe epilepsy. The aim of this study is to evaluate the ability of wavelet power (WP) to discriminate seizures in ECoG and EEG from a continuously monitored pediatric population with both temporal and extra-temporal epilepsy. Methods: 36-72 hour recordings, consisting of 96-128 channel ECoG and EEG from five surgical patients (ages 13 mo, 12 yr, 13 yr, 13 yr, and 21 yr) at the University of Chicago Comer Children's Hospital, were analyzed; there was no pre-selection other than that the data be free of recording problems. Each channel was filtered with a 1-35 Hz bandpass and a 60 Hz notch; 2.5s data windows were decomposed using the Daubechies wavelet, similar to [1]. WP from the resulting level 3, 4, and 5 coefficients (WP3, WP4, WP5 respectively) were used to register the detections (≥10 s above threshold, with 120 s max. for a single detection) compared against events marked by a board certified neurologist or registered EEG technician; the researchers and human experts were blind to each other's findings until the results were formally compared. ROC-like thresholding was used to assess effectiveness. Results: WP5 and WP4 performed better than WP3. In the ECoG of 4/5 patients, WP detected all seizures in at least one channel while keeping the false alarm rate below the seizure rate (ie, a given detection had >50% chance of being true) - our minimum standard for detection efficiency (DE); this was true for 2/5 patient with EEG. In 1/5 patients, the averaged WP over EEG meets the min. DE, while the averaged WP from ECoG meets min. DE in 2/5 patients. Since these are continuous data, the ratio of interictal to ictal epochs is realistic for each patient. Conclusions: (1) WP5 is the best seizure discriminator, WP3 the least powerful; (2) In each patient, WP5 matches the most successful metric found in a previous study using a subset of these data [2], so using only WP5 would give as good a result as possible without having to choose a different metric for each patient; (3) The patient matters: WP does a consistently good job except on the 13 mo. old, whose active record proved problematic for the human expert as well as all metrics in [2]; (4) In 2/5 patients, the most efficient surface channels allow detections that are correct more than 50% of the time. WP holds promise as a practical seizure discriminator in our patients. Funding support by: Falk Foundation and a Grant from the Linn Family [1] Khan, YU, and Gotman, J. Wavelet based automatic seizure detection in intracerebral electroencephalogram. Clin Neurophysiol. 2003 May;114(5):898-908. [2] Lee HC, van Drongelen W, McGee AB, Frim DM, Kohrman MH. Comparison of seizure detection algorithms in continuously monitored pediatric patients. J Clin Neurophysiol. 2007 Apr;24(2):137-46.