Rationale: In patients presenting with a new-onset seizure, visual analysis of electroencephalogram (EEG) waveforms is widely used to evaluate the risk of future seizures. However, routine EEG interpretation has limited ability to identify patients at risk of future seizures, as many patients with normal-appearing EEGs will go on to have future seizures. Computational analysis of digital EEG recordings using the information theory measure of entropy may identify subtle underlying abnormalities that are missed by routine visual inspection, ultimately improving clinicians’ ability to identify and treat patients at risk of future seizures.
Methods: Twenty-eight 10-20 system clinical EEG recordings from patients who experienced a single unprovoked seizure were retrospectively identified. All recordings were initially interpreted as normal by a board certified electroencephalographer. However, nine of the patients went on to have future seizures within a two-year follow-up period (nineteen did not). For each recording, signals between 1-20 Hz were divided into 10 second time windows. Entropy was calculated within each time window for each of nineteen electrodes separately. The mean entropy for each electrode within each patient recording was then calculated and used for statistical analysis.
Results: Differences were observed in the magnitude of entropy between electrodes (F(18,468)=30.920, p<0.0001), with additional marginal differences between patients who went on to have future seizures versus those who did not (F(1,26)=3.963, p=0.057). The direction of means across all nineteen electrodes consistently reflected lower entropy in patients who went on to have future seizures compared to those who did not; however, these differences are not statistically significant for all electrodes individually. Further analysis evaluating only electrodes in temporal and parietal regions revealed a significant difference in entropy within these regions between patient groups (F(1,26)=4.696, p=0.040).
Conclusions: Calculated entropy is lower in the EEGs of patients who experience multiple seizures than in those who have only one. This study indicates that computational analysis of digital EEG may have a clinical role in diagnosing patients at risk of future seizures.
(Funding from Alpert Medical School Summer Research Assistantship Program and the Matthew Siravo Memorial Foundation)