Abstracts

Rationale: : Given that both Granger Causality (GC) and HFO generate prognostic data about seizure onset from interictal data, the possibility of an approach which would synthesize both data streams to a single statistical mapping has appeal. We determined the probability of resection of individual electrode loci based on whether each electrode had both sufficient causality (Park and Madsen, 2017) and short enough latency of HFO propagation (Tamilia et al, 2017). We sought to find optimal thresholds for each quantity by maximizing "informedness" (sensitivity+specificity-1), calculated for each case by varying parameters. Since the goal is a priori estimation of most ictogenic areas, we then determined the consequences of using the mean or median value from the entire patient set on the set of electrodes predicted to be ictally important. We reasoned that if the optimized thresholds prove similar to actual resection zone, a stronger computational approach for iEEG and intraoperative ECoG could result. Methods: We analyzed 5 to 7 minute interictal iEEG data intervals obtained from five patients using GC and HFOs techniques and calculated GC based out-degree and HFOs latency. All electrodes for each case were plotted in GC out-degree against 1/HFO latency. Threshold points (consisted of GC out-degree and 1/latency values) were used to calculate specificity and sensitivity with respect to resected set of electrodes (for example, we state “true positive” if a certain electrode is present in the threshold space and also in the resected set of electrodes). To estimate the probability of informed decision, we calculated informedness from the specificity and sensitivity. For each case we marked the set of electrodes determined by the threshold from within patient maximum informedness on the cortical map and compared the map with the resection. Since all the thresholds were close to each other, we also examined the effect of substituting mean and median of the threshold values from the entire group over all 5 cases. Results: By visual comparison with resection map, for all 5 cases we found that electrodes set determined by the within-subject optimized threshold at the maximum informedness were inside the resected area. Interestingly, for one of the cases with few HFOs, the electrodes set determined by the threshold were well within the resection area. Whether we used individual threshold or mean and median of the thresholds for all cases, maps looked quite similar to each other. Conclusions: Use of a threshold in both GC and HFO latency may improve discrimination of ictal onset, therefore surgically relevant, electrode sets. More importantly, the actual optimized thresholds, while different for each patient, tend to cluster in a region which may allow a priori recognition of likely ictal electrodes from interictal, intraoperative data. This may improve surgical decision making from intraoperative corticography. Funding: This work is funded in part by NIH (1R01NS069696) .