Abstracts

Rationale: Epileptic seizures are debilitating because of the unpredictable loss of function, and anti-epileptic medications are titrated to achieve maximal seizure control with minimal side effects. However, optimizing seizure control can be difficult because due to subjective reporting, known to be variable and unreliable. An objective and easily quantified mechanism for detecting seizures over the long-term is needed to achieve more accurate and replicable medication titration. Although attempts are made to detect seizures through indirect methods such as autonomic responses or motion detectors, the electroencephalogram remains the gold standard for seizure detection. To meet these needs, we propose a minimally invasive approach for seizure detection and monitoring using a single differential recording from the scalp.Methods: We recorded EEG activity from two surface electrodes in patients being evaluated for seizure at Yale-New Haven Hospital (IRB Protocol# 0701002210). In addition to the leads placed for traditional 10-20 EEG mapping, two additional surface leads were placed: one each on the left and right mastoid processes (labeled X7 and X8), just adjacent to the normal A1 and A2 leads, respectively. Recordings were reviewed by a trained neurologist, seizure activity was identified, and recordings were divided up into discrete sections with or without seizure activity based on the entire 10-20 montage. For each section, the two unique signal differentials (X7-A1 and X8-A2) were low-pass filtered at either 20 Hz or 50Hz and processed with spectrographic, power spectral, and Teager energy analyses.Results: We found that seizure activity could be reliably detected utilizing this single differential recording. Seizures were detected as a broad-spectrum increase in signal power and sustained increased in Teager energy. No difference was found in detection between signals filtered at 20 Hz or 50 Hz, so all sections were low-pass filtered at 20 Hz in subsequent analyses. Abnormally elevated activity was defined as Teager energy values greater than the 95th percentile for sum recordings from each patient. Compared to sections without seizure (n=66), sections from EEG recordings with verified seizures (n=55) demonstrated significantly greater time with sustained elevations of Teager energy (p<0.001). Based on this finding, we build an algorithm that was able to differentiate seizure from non-seizure activity with a sensitivity of 76% and a specificity of 50%. Analyses of longer segments of time with other electrode combinations are on-going and will provide additional data on how better to tune the method for more accurate detection.Conclusions: These results demonstrate the feasibility of seizure detection using only a single discrete differential EEG recording.