Rationale:
There are no surrogate biomarkers that reliably delineate epileptic networks. High-frequency oscillations (HFO) have been associated with epileptogenesis and ictogenesis. There is mounting evidence that resection of areas with high rates of HFO in fast ripple range overriding spikes is associated with a favorable outcome. Recent studies have explored spatiotemporal organization of HFO into networks and their potential role in defining epileptic networks and mapping seizure onset zones.Methods:
We conducted a retrospective analysis of 5 patients with refractory focal epilepsy undergoing intracranial monitoring for surgical evaluation. We selected 30-minute epochs for each patient corresponding to the immediate post-operative phase, least anti-seizure medications preceding spontaneous seizures, highest anti-seizure medications following spontaneous seizures, and triggered seizures during direct cortical stimulation.
All patients underwent automated computational analysis using MNI detector script via MATLAB, outputting ripple and fast ripple events. Only two had concomitant visual analysis to compare with automated detections. The HFO detector results were mapped to visual annotations, and further normalization statistical methods were utilized to validate those results. Artifacts were removed and, after being validated, the detections were run through a novel spatio-temporal network flow analysis and optimization algorithm, which identified parcels on the Yale Brain Atlas with the greatest flow of HFOs. These areas, if resected, would significantly decrease the flow of HFOs through the brain network. The HFO flow output was mapped onto the Yale Brain Atlas (see figure). Lastly, we compared the parcels with the greatest HFO flow with the seizure onset zones.
Results:
For the 5 patients, an average of 14209 HFO events were detected by the automated detector, 344 of which were artifacts. The spatiotemporal network flow and optimization algorithm identified an average of 7 clusters, from which YBA parcels with greatest HFO flow output identified clusters with high flow that are distinct from those with high HFO count (see fig 2). Our analysis confirms that the clusters with greatest flow correspond with the seizure onset zones. Those with concordant greatest flow and seizure onset zone had seizure free outcome. Those with discordant results had poor outcomes.
For example, for Patient 1 (see figs), 13389 HFO events were detected by the automated detector, 314 of which were artifacts. 3226 HFO events were detected by visual analysis (fig 1). 4 clusters were identified, from which 10 parcels were identified, overlapping with the ictal onset zone, and correctly predicting surgery success. The outputs of the analysis can be seen in the figure.
Conclusions:
Our analysis confirms that HFO have clustered spatiotemporal organization, with potential role in mapping seizure onset zones and determining surgical outcome. While our sample size is small, this observation necessitates further investigation with a larger sample size. We are currently working on 50 additional patients, which will be ready by AES 2024.Funding: Funded by Yale College Dean's Fellowship