Comparing EEG background activity of epileptic patients and control subjects using spectral entropy and permutation-based statistical testing

Abstract

In this study we have analyzed the background activity of the electroencephalogram of patients with epilepsy and of control subjects. We have studied the recordings of 10 patients and 10 controls by means of their spectral entropy in various frequency bands to determine whether the groups could be differentiated by this feature alone. Contrary to other similar studies, we have examined the statistical significance of the sample by implementing a (nonparametric) cluster-based permutation statistical test.

This has the advantage that it provides a straightforward way to solve the multiple comparisons problem, in that it allows considering all available channels and time/frequency samples simultaneously, but requires no a-priori correction and results in a valid statistical test. In our experience, background activity from patients with epilepsy is enough to detect the condition when compared to a control group, i.e. there exists a significant difference between groups according to their spectral entropy in a specific frequency band. In fact, we have obtained statistical significance (at below 0.05 alpha-level) in 18 out of 19 channels between 5 and 11Hz and in an ever wider range for some channels.