Abstracts

Rationale: When clinicians analyze intracranial EEG, they rely on both interictal and ictal biomarkers of the location of seizure onset. Quantitative research has revealed several quantitative measures of the behavior of the epileptogenic network. One such measure is the Epileptogenicity Index (EI). In the current study, we applied EI to the EEG of patients in a database of clinical intracranial EEG recordings to test the feasibility of applying this method in a diverse clinical dataset.  Methods: We used clinical data from intracranial EEG recordings from pediatric and adult patients evaluated at our institution. Using the method of calculation of EI previously described, we first calculated a measure of the ratio fast to slow energy spectra at each sample in a snippet of EEG surrounding a seizure. This calculation was performed on up to 5 of each of the types of seizures captured for each patient. Next, the EI was calculated using the Page-Hinckley algorithm to detect the moment when the energy ratio changed to favor high frequency activity at the start of seizure in each channel. Finally, we counted the frequency with which this algorithm produced a robust measure of EI.  Results: This analysis was performed in a total of 61 seizures from 14 different patients taken from a consecutive cohort including all patients who underwent intracranial EEG monitoring at the University of Michigan. Patients were included if clinical data and intractanial EEG data were available at the time of analysis. In 4/14 patients this algorithm did not yield a reliable EI value; in 7/14 patients the algorithm yielded consistently reliable EI values, and in 3/14 patients the algorithm did not yield a reliable EI value in a subset of seizures or seizure types. In 46/61 seizures the algorithm yielded a reliable measurement of EI. The time of reported seizure onset detected by the EI algorithm was dependent on seizure type: among seizures with robust EI values, the detection of seizure onset was most delayed in seizures with prominent slow polarizing shift at the time of seizure onset.  Conclusions: These results show that calculation of EI, which depends on a transition from low-frequency to high-frequency spectral power at seizure onset, is a robust measure of epileptogenicity in 75% of the patients tested. This analysis also shows that a subset of seizure onset is characterized by some other pattern. For instance, rhythmic spiking less than 12 Hz at seizure onset causes a decrease in energy ratio, leading to an inability to accurately detect seizure onset with the algorithm. Furthermore, robust EI values (defined as rank order of EI stable for a broad range of parameter values) do not necessarily imply accurate identification of the time of seizure onset when there is slow depolarizing shift in onset channels. More investigation is needed to determine whether incorporating measures of epileptogenicity that are sensitive to slow polarizing shift will improve the robustness of such quantitative biomarkers.  Funding: 1. AES Research and Training Fellowship for Clinicians (LOI), American Epilepsy SocietySmith, Garnett (PI) 2. R01 NS094399-01