Abstracts

Rationale: Epilepsy surgery remains an option for pediatric patients with intractable epilepsy, but the outcomes and potential morbidity of the procedure is not ideal. Extensive presurgical evaluation is required to identify the epileptic regions of cortex prior to any resection. Surgical decisions are made based on visual inspection of intracranial EEG (IEEG) recordings. Unfortunately, this approach is qualitative, labor intensive, and ultimately insufficient to render all patients seizure free. The ability of automated signal processing algorithms to extract features from the interictal EEG is a recent advancement based on improvements in the data sciences. One such feature that has been hypothesized to be important for understanding the ictal networks is the propagation of interictal spikes across subdural recording grids. Thus, we aimed to explore the clinical relevance of interictal spike propagation in intractable pediatric epilepsy surgery patients.Methods: This study is IRB approved. Intracranial IEEG was obtained from 18 pediatric patients (mean age = 10.3, range = 3-17 years) who underwent Phase II presurgical evaluation. Novel algorithms were developed to detect interictal spikes and quantify their spatiotemporal propagation. Channel pairs that frequently co-occurred within propagating discharges were identified in order to visualize spike connectivity networks. A graph theoretic approach was used to describe the topology, configuration, and spatial extent of spike networks across a wide range of threshold densities. Individual spike trajectories were compared using aggregating algorithms and clustered to uncover unique propagation patterns. The distribution and sequential activation of cluster events was examined within random EEG segments and during the hour surrounding ictal onset.Results: Reproducible propagation patterns were identified within interictal, pre-ictal, and post-ictal segments. Seizure-free patients and those with focal pathology demonstrated spike networks with decreased global integration (characteristic path length) and increased local interconnectedness (clustering coefficient) relative to comparison groups. Using these features as predictors, a linear-kernel support vector machine (SVM) learning algorithm accurately classified patients by postsurgical outcome (89%) and pathology class (72%). For a subset of seizures, propagation trajectories became more spatially confined and stereotyped during the post-ictal period while several other EEG measures (e.g., relative delta frequency power, root mean square amplitude) also varied with ictal state.Conclusions: These results suggest that network analysis of interictal spike propagation contributes useful insights into the spatial configuration and epileptogenic potential of EEG networks prior to surgery. Further, propagation patterns were ictal state-dependent and co-varied with spectral features of the EEG. Together, these findings implicate a significant role for spike propagation analyses in the presurgical evaluation while furthering our understanding of the pre-ictal and post-ictal states.