Classification of meaningful and meaningless visual objects: a graph similarity approach

TL;DR

This study uses EEG data and network similarity algorithms to distinguish meaningful from meaningless visual objects, achieving 76% accuracy in categorization based on brain network states.

Contribution

It introduces a novel approach combining brain network reconfiguration analysis with network similarity metrics to classify visual object categories.

Findings

Dynamic brain networks can be segmented into states reflecting processing stages.

High intra-category and low inter-category network similarity enable effective discrimination.

Achieved 76% classification accuracy across different brain states.

Abstract

Cognition involves dynamic reconfiguration of functional brain networks at sub-second time scale. A precise tracking of these reconfigurations to categorize visual objects remains elusive. Here, we use dense electroencephalography (EEG) data recorded during naming meaningful (tools, animals) and scrambled objects from 20 healthy subjects. We combine technique for identifying functional brain networks and recently developed algorithm for estimating networks similarity to discriminate between the two categories. First, we showed that dynamic networks of both categories can be segmented into several brain network states (times windows with consistent brain networks) reflecting sequential information processing from object representation to reaction time. Second, using a network similarity algorithm, results showed high intra-category and very low inter-category values. An average accuracy of 76% was obtained at different brain network states.