Hemispheric Fingerprinting: Identifying left and right hemispheres using functional connectivity

Abstract

Many studies have analyzed what organizational features distinguish the left and right hemispheres of the human brain, with differences typically being found in language areas and in fine motor control (i.e., handedness). In this analysis, we test whether supervised learning can categorize ("fingerprint") an unseen hemisphere as right or left based on functional connectivity. Using data from the Human Connectome Project, we find success to be extremely high (accuracies > .90) in models trained on right-handed participants (Edinburgh Handedness Inventory [EHI] > 0) and in models trained on left-handed participants (EHI [≤] 0). In a second analysis, we test whether the same can be done to identify handedness alongside hemisphere left/right sidedness. While individuals hemispheres are less distinct the more left-handed they are, hemiconnectomes cannot be reliably classified as belonging to a left- or right-handed person. Our approach can inform developmental and post-injury work on hemispheric organization and reorganization.