Task Classification Using Topological Graph Features for Functional M/EEG Brain Connectomics

Abstract

In the last few years the research community has striven to achieve a thorough understanding of the brain activity when the subject under analysis undertakes both mechanical tasks and purely mental exercises. One of the most avant-garde approaches in this regard is the discovery of connectivity patterns among different parts of the human brain unveiled by very diverse sources of information (e.g. magneto- or electro-encephalography – M/EEG, functional and structural Magnetic Resonance Imaging – fMRI and sMRI, or positron emission tomography – PET), coining the so-called brain connectomics discipline. Surprisingly, even though contributions related to the brain connectome abound in the literature, far too little attention has been paid to the exploitation of such complex spatial-temporal patterns to classify the task performed by the subject while brain signals are being registered. This manuscript covers this research niche by elaborating on the extraction of topological features from the graph modeling the brain connectivity under different tasks. By resorting to public information from the Human Connectome Project, the work will show that a selected subset of topological predictors from M/EEG connectomes suffices for accurately predicting (with average accuracy scores of up to 95%) the task performed by the subject at hand, further insights given on their predictive power when the M/EEG connectivity is inferred over different frequency bands.