Connectivity detection in application to spike-wave discharge study

Abstract

In our study, we compare three popular approaches to directed coupling analysis, in particular transfer entropy and two types of Granger causality, applied to real data from genetic absence epilepsy rats. We have chosen the channels for which the coupling architecture is already well known from previous studies. Recordings from 5 WAG/Rij rats of 8 hours duration with at least 28 spontaneous seizures of length not less than 6 s in each recording were studied. To test results for significance, surrogate signals based on series permutation technique were constructed. Connectivity development in time was investigated by considering six two-second intervals before, during and after the seizure. Our outcomes showed large differences between studied approaches, while all of them exploit the same general idea. Transfer entropy demonstrated the smallest number of significant couplings throughout all three considered measures, while the linear Granger causality showed the largest number of them. This indicates that transfer entropy is the most conservative measure and the least sensitive one. Its sensitivity is affected by insufficient series length. The linear Granger causality is likely to demonstrate insufficient specificity.