METHODS FOR STATISTICAL EVALUATION OF CONNECTIVITY ESTIMATES IN EPILEPTIC BRAIN

Abstract

Connectivity analysis using modern approaches like Granger causality, partial directed coherence and transfer entropy always demands additional statistical evaluation of the obtained measures for significance. Although for very simple linear autoregressive processes and quasilinear oscillatory activities theoretical estimates are available, the real biological signals are too complex for application of analytical approaches and surrogate data come into use. When pathological activity like epileptic seizures is studied, the question can also rise in a somewhat different way: one asks whether the studied activity is different from the normal one rather than whether connectivity exists. The similar question is also valid if one compares connectivity in different physiological states like sleep and wakefulness. Here, we study two different approaches to statistical evaluation of transfer entropy estimates in application to the study of spike–wave discharges (SWDs), the main encephalographic manifestation of absence epilepsy, registered in local field potentials of WAG/Rij rats (genetic models). The first approach is to compare distributions of the estimators for the baseline and different stages of pathological activity using traditional measures like t-test with additional corrections for multiple testing. The second approach is to make surrogate data and test whether the achieved estimators differ for surrogate series and for real ones. To support our findings and to understand the methods better, the series simulated using simple oscillatory models of epileptic activity are evaluated in the same way as the experimental data. We show that the most pronounced phenomena like bidirectional increase in coupling between frontal and parietal cortical areas during SWDs in comparison to baseline activity are considered to be significant by both approaches. But when the less expressed coupling changes are under consideration, the approach base on surrogate data provides less false positives. These results confirm that the primary outcomes of connectivity analysis for absence epilepsy (and not only it) achieved previously are valid although the statistical evaluation of the connectivity estimators was suboptimal.