Improving Phenotype Prediction using Long-Range Spatio-Temporal Dynamics of Functional Connectivity

The study of functional brain connectivity (FC) is important for\nunderstanding the underlying mechanisms of many psychiatric disorders. Many\nrecent analyses adopt graph convolutional networks, to study non-linear\ninteractions between functionally-correlated states. However, although patterns\nof brain activation are known to be hierarchically organised in both space and\ntime, many methods have failed to extract powerful spatio-temporal features. To\novercome those challenges, and improve understanding of long-range functional\ndynamics, we translate an approach, from the domain of skeleton-based action\nrecognition, designed to model interactions across space and time. We evaluate\nthis approach using the Human Connectome Project (HCP) dataset on sex\nclassification and fluid intelligence prediction. To account for subject\ntopographic variability of functional organisation, we modelled functional\nconnectomes using multi-resolution dual-regressed (subject-specific) ICA nodes.\nResults show a prediction accuracy of 94.4% for sex classification (an increase\nof 6.2% compared to other methods), and an improvement of correlation with\nfluid intelligence of 0.325 vs 0.144, relative to a baseline model that encodes\nspace and time separately. Results suggest that explicit encoding of\nspatio-temporal dynamics of brain functional activity may improve the precision\nwith which behavioural and cognitive phenotypes may be predicted in the future.\n