Abstracts

Rationale: Successful surgery in epilepsy can be conceptually understood as safe removal of the causal part of a seizure network, preventing future seizures. If we could reveal the causal network without seeing actual ictal events, advantages would accrue for reducing cost and risk involved in surgery. Recently we have reported that Granger causality (GC) analysis can reveal seizure networks from interictal baseline obtained from grid electrodes. We have applied the same method to interictal data recorded from stereoelectroencephalography (sEEG) depth electrodes and produced Granger causality maps well corresponding to topography of ictal networks. Technically, this process is fast enough that it would be possible to analyze short segments of intraoperative electrocorticography (ECoG) data, returning GC maps to the intraoperative clinical team in time to impact decision-making. However, the question remains: can we still see GC pattern corresponding to desired surgical targets under anesthesia? In this study, we generated GC maps from intraoperatively recorded ECoG data (pre-resection stage) and tested the hypothesis that the regions of high GC would statistically resemble the topography of the actual resection. Methods: Segments of intraoperatively recorded ECoG data (ranging from two to five minutes: 3.5 ± 1.1 min (mean ± SD)) from patients who underwent one-stage epilepsy surgery were analyzed (n=8). Causality maps were quantitatively compared to actual resections and areas of active spiking used to help direct resection boundaries, by using non-parametric rank-order statistics. Results: In 7 of 8 cases, the causality rankings of the electrodes mapped to the resection had higher causality than predicted by chance (range: p < 10^-5 to 0.04) but surprisingly, only one case showed a statistically significant higher causality within areas of active spiking (p = 0.01), suggesting that high spike frequency is neither necessary nor sufficient for detection of Granger causality. Conclusions: The results suggest that GC maps produced using intraoperatively obtained ECoG can potentially be used to predict the anatomy of surgical resection in the operating room. The GC maps could provide qualitatively different and complementary information compared with traditional spike topography, which is currently used in single-stage resective surgeries. Potentially useful aspects of GC networks appear to persist under general anesthesia and craniotomy, suggesting that real-time GC calculations may be useful intraoperatively during epilepsy surgery.  Funding: This work is funded in part by NIH (1R01NS069696 and U01EB023820-01) and in part by Wyss Institute for network data storage and management.