Abstracts

Rationale: Characterization of an epileptogenic focus from interictal intracranial EEG (iEEG) data could be very helpful in intraoperative decision making about resection location and extent, potentially obviating need for extra-operative invasive monitoring required to record an actual typical seizure.  Recently, Granger Causality (GC) measures (Park and Madsen, 2017) and high frequency oscillations (HFO) latency measures (Tamilia et al. 2017) have been successfully applied to the problem of statistically predicting seizure onset and resection zones using interictal iEEG data. It is therefore important to know if the two techniques invariably produce similar results, or alternatively yield complementary topographic information in surgical cases. We compared the results of analysis of a 5-7min interictal ECoG interval from each of 5 pediatric patients undergoing invasive monitoring, to determine the following: 1) do both methods statistically predict the actual resection determined based on conventional iEEG reading, 2) are the rankings of electrodes correlated between the two techniques, and 3) are there uncorrelated cases where the synergistic use of both analyses could improve identification of the seizure focus over either technique alone? Methods: 5 to 7 min segments of interictal iEEG data obtained from five patients were analyzed using GC and HFO techniques. We calculated GC out-degree and HFOs latency for each electrode and ranked them. The rankings from two techniques were quantitatively compared using Spearman’s rank correlation and the rank orders were used to statistically compare them with ictally identified seizure onset zone electrodes (SOZ) and electrodes in the resection zone using non-parametric rank-order methods (Park and Madsen 2017). Results: Using Spearman’s rank correlation, 4 of the 5 cases with rich HFOs showed  extremely high correlations between the rank ordering of GC and HFO latency (log10p values -9.7 to -4.5). One case with few HFOs did not show significant correlation with GC. In 2 out of 5 cases, the HFOs latency rankings of the electrodes in the SOZ set were lower than predicted by chance (p < .0.05) as was the case in 3 of 5 cases by GC ranking.  When looking at the rank ordering within the resection set, all 5 cases reached significance for GC rankings (log10p values ranging from -5 to -4.5) as did 3 out of 5 cases for HFOs latency based rankings. Conclusions: The results suggest that both methods can statistically point out actual resection and seizure onset zone and also the rankings of electrodes form the two methods are generally correlated. The results also suggest that GC can provide useful information in terms of prediction of resection area in cases with few HFOs. In these cases, use of the two methods together might improve identification of the seizure focus over either technique alone. Funding: This work is funded in part by NIH (1R01NS069696) and in part by Wyss Institute for network data storage and management.