Abstracts

Rationale: The visual interpretation of intracranial EEG (iEEG) is the clinical standard for mapping focal seizure onsets targeted for resection during epilepsy surgery. Still, visual iEEG interpretation is labor-intensive and on occasion, prone to localization errors due to interpreter dependency. Multivariate, parametric, frequency-based connectivity measures related to the concept of Granger-causality have been evaluated for their ability to map electrographic seizure onsets with promising results. However, these methods have been generally applied to a limited number of iEEG channels/time series (<50-60), often hand-picked based on visible involvement in the course of a seizure. Here, we tested whether the inclusion of unselected iEEG time series in connectivity estimations from high-density iEEG studies is feasible for seizure-onset mapping, given an intensive computational cost and concern for low signal-to-noise in the presence of a large number of channels.Methods: Focal ictal oscillatory activity was simulated in up to 160 iEEG channels. In addition, a 113-channel, artifact-free iEEG epoch was analyzed, containing the first 35 sec of a seizure from a patient who underwent iEEG monitoring at the University of Texas Health Science Center at San Antonio, and was rendered free of seizure following resective surgery. A Kalman filter-based multivariate adaptive autoregressive model was fitted to the iEEG time series. Time-varying formulations of the directed transfer function (DTF) and partial directed coherence (PDC) connectivity measures were derived. The outdegree, a graph theoretical connectivity measure encoding the number of outgoing connections, was then calculated for each channel. Channel-specific outdegrees were summed over time to obtain the total outdegree. The highest total outdegrees indicated the channels involved at the seizure onset. Results were compared for colocalization with the visually identified seizure onset, and where applicable, with the resection volume. All analysis was performed using GPU and CPU-optimized code written in Matlab.Results: In both simulations and the real-data seizure analyzed, the connectivity-based seizure onset overlapped with the onset determined by visual iEEG interpretation. Measure-specific results are contrasted. In most cases, the run time necessary for connectivity calculations was less than 12 hours on a regular PC.Conclusions: In this study, we demonstrated the feasibility of mapping accurately the seizure onset based on a large number of unselected iEEG channels. If confirmed by future reports, this approach may obviate the time-consuming and subjective step of time-series selection for connectivity estimation.