Intelligence Quotient Scores Prediction in rs-fMRI via Graph Convolutional Regression Network.

The intelligence quotient (IQ) scores prediction in resting-state functional magnetic resonance imaging (rs-fMRI) imagery is an essential biomarker in understanding autism spectrum disorder (ASD)' mechanisms and in diagnosing and treating the disease. However, existing intelligence quotient prediction methods often produce unsatisfactory results due to the complexity of brain functional connections and topology variations. Besides, the important brain regions which contribute most to IQ predictions are often neglected for priority extraction. In this paper, we propose a novel Graph Convolutional Regression Network for IQ prediction that consists of an attention branch and a global branch, which can effectively capture the topological information of the brain network. The attention branch can learn the brain regions' importance based on a self-attention mechanism and the global branch can learn representative features of each brain region in the brain by multilayer GCN layers. The proposed method is thoroughly validated using ASD subjects and neurotypical (NT) subjects for full-scale intelligence and verbal intelligence quotient prediction. The experimental results demonstrate that the proposed method generally outperforms other state-of-the-art algorithms for various metrics.