Predicting States of Abstract Reasoning Using EEG Functional Connectivity Markers

High order abstract reasoning is impaired in patients suffering from mental disorders especially from schizophrenia. Thought and language disorders typical of schizophrenia are presumably connected with the aberrant ability to filter out irrelevant associations. We hypothesized that EEG biomarkers in healthy population could be detected, extracted and validated with regard to the ability to abstract a general principle underlying presented words while ignoring irrelevant associations and retaining only relevant ones. We developed three models of abstract reasoning: a direct generalization presented by nouns from the same semantic category, a latent association based on a loose relation between the presented words, and no associations introduced by non-related words. In the present EEG study 17 healthy participants solved tasks trying to figure out a general principle in a group of words. Subsequently, we carried out a functional connectivity analysis in order to restore synchronous neuronal interactions in the theta-alpha frequency range. We used the obtained spatial patters restored individually and relevant phase locking values (PLVs) as features for the Support Vector Machine classifier with Gaussian kernel. The accuracy rating validated on an independent sample made up 62.5% which is a promising result if inter-subject variability in cognitive processing is taken into account. Being validated on the same sample, the accuracy reached 82%. The results indicate that spatial patterns of functional connectivity and PLVs can be used as predictors of types of abstract reasoning.