Applications of Granger Causality Calculations to Brain Networks for Efficient and Accurate Seizure Focus Identification via iEEGs

Abstract

Epilepsy patients can manage their seizures with medicines, but surgery remains the only cure. A challenge presented by such a treatment is accurate and efficient determi- nation of the seizure onset zone (SOZ), as the current standard of care requires week- long patient hospital stays for monitoring. However, Granger Causality (GC) analysis of interictal iEEG data shows promise in offering information about the epileptic net- work, potentially offering a quicker, alternate route to resection surgery planning [12]. This project builds off of those results by subjecting the matrix output of GC analysis to further algorithmic inquiry with a goal of improving predictive capability. A retrospective evaluation of 25 patients was conducted to evaluate effectiveness of each approach. Sampling approaches yielded up to 23.1% more patients with statistically sig- nificant predictions about the ultimate resection zone than previously published GC matrix analyses, yet were inconsistent due to the stochastic nature of the algorithm. Despite interest drawn from its unique application, PageRank was unable to improve in the number of patients for whom it offered significant SOZ and resection zone predictions. Centrality algorithms yielded 23% more patients than GC alone while providing precision of results not offered by sampling methods. Finally, clustering approaches provided higher fidelity resection recommendations than all other algorithms, but for a smaller number of patients.