How sensitive is functional connectivity to electrode resampling on intracranial EEG? Implications for personalized network models in drug-resistant epilepsy

Abstract

Focal epilepsy is a clinical condition arising from disordered brain networks. Network models hold promise to map these networks, localize seizure generators, and inform targeted interventions to control seizures. However, incomplete sampling of epileptic brain due to sparse placement of intracranial electrodes may profoundly affect model results. In this study, we evaluate the robustness of several published network measures applied to intracranial electrode recordings and propose an algorithm, using network resampling, to determine confidence in model results. We retrospectively subsampled intracranial EEG data from 28 patients who were implanted with grid, strip, and depth electrodes during evaluation for epilepsy surgery. We recalculated global and local network metrics after both randomly and systematically resampling subsets of intracranial EEG electrode contacts. We found that sensitivity to incomplete sampling varied significantly across network metrics, and that this sensitivity was independent of the distance of removed contacts from the seizure onset zone. We present an algorithm, using random resampling, to compute patient-specific confidence intervals for network localizations on both global and nodal network statistics. Our findings highlight the difference in robustness between commonly used network metrics and provide tools to assess confidence in intracranial network localization. We present these techniques as an important step toward assessing the accuracy of intracranial electrode implants and translating personalized network models of seizures into rigorous, quantitative approaches to invasive therapy.

The Interictal Suppression Hypothesis in Focal Epilepsy: Electrographic and Structural Evaluation

“…This could be due to an intrinsic difference in PZ connectivity between surgical responders and non-responders, a difference in SEEG implantation strategy between the patient groups that affects network observations, or a difference in ictal interpretation that affects node designation. 115–117 With this observation as motivation, we developed SVM models to classify SOZs, PZs, and non-involved regions. The models were able to classify SOZs with very high accuracy (range of 91.4-92.5% accuracy) but differed in ability to differentiate PZs from non-involved regions (range of 81.6-93.0% accuracy).…”

Interictal intracranial EEG for predicting surgical success: the importance of space and time

“…It is clear that such subnetworks do not necessarily have the same properties as the whole-brain network. Recent analyses demonstrated that even leaving out one node from iEEG functional networks can dramatically change their network properties (Conrad et al, 2019). The implication is that the property of the epileptogenic tissue/network may change depending on the subnetwork sampled, which may explain some conflicting results in the literature (Zijlmans et al, 2019).…”

“…between outcome groups Spatially undersampling networks can directly lead to changes in the estimated network properties (Conrad et al, 2019), and thus we investigated the impact of spatial sampling on our ability to distinguish outcome groups. Fig.…”