Comparing connectivity metrics in cortico-cortical evoked potentials using synthetic cortical response patterns

David Prime,Matthew Woolfe,David Rowlands,Steven O'Keefe,Sasha Dionisio

DOI: https://doi.org/10.1016/j.jneumeth.2019.108559

2020-01-09

Abstract:Background: Cortico-Cortical Evoked Potentials (CCEPs) are a novel low frequency stimulation method used for brain mapping during intracranial epilepsy investigations. Only a handful of metrics have been applied to CCEP data to infer connectivity, and no comparison as to which is best has been performed. New method: We implement a novel method which involved superimposing synthetic cortical responses onto stereoelectroencephalographic (SEEG) data, and use this to compare several metric's ability to detect the simulated patterns. In this we compare two commonly employed metrics currently used in CCEP analysis against eight time series similarity metrics (TSSMs), which have been widely used in machine learning and pattern matching applications. Results: Root Mean Square (RMS), a metric commonly employed in CCEP analysis, was sensitive to a wide variety of response patterns, but insensitive to simulated epileptiform patterns. Autoregressive (AR) coefficients calculated by Burg's method were also sensitive to a wide range of patterns, but were extremely sensitive to epileptiform patterns. Other metrics which employed elastic warping techniques were less sensitive to the simulated response patterns. Comparison with existing methods: Our study is the first to compare CCEP connectivity metrics against one-another. Our results found that RMS, which has been used in many CCEP studies previously, was the most sensitive metric across a wide range of patterns. Conclusions: Our novel method showed that RMS is a robust and sensitive measure, validating much of the findings of the SEEG-CCEP literature to date. Autoregressive coefficients may also be a useful metric to investigate epileptic networks.