Sub-graph entropy based network approaches for classifying adolescent obsessive-compulsive disorder from resting-state functional MRI

Highlights • The proposed classification scheme improves the best-known adolescent OCD vs. healthy controls classification accuracy from 78% to 89%.• This paper validates the efficacy of sub-graph entropy for classifying OCD vs. healthy groups.• The proposed technique identifies a predictive sub-network that partially consists of regions from well-known cortico-striato-thalamic-cortical (CSTC) network.• For both predictive sub-network and CSTC sub-network, sub-graph entropy is significantly lower in OCD patients compared with healthy controls.• Sub-graph entropy using 1-hop neighborhood is effective in the classification task. Sub-graph entropy using 2-hop neighborhood has a lower classification accuracy.