Functional covariance modes reveal aligned fetal and neonatal brain functional connectomes

Multiple lines of evidence suggest that that spatially distributed functional networks in the brain may first start emerging before birth. Reliably demonstrating this in utero using fMRI remains a very challenging problem due to a variety of MRI-adverse factors, including distant coil positioning and motion-induced magnetic field perturbations. Here, we introduce a novel approach to functional network analysis, called seed-based functional covariance modes (FCMs), which leverages inter-subject variability in connectivity between seed regions and the rest of the brain to infer whole-brain network configurations. We first applied the FCMs approach to neonatal data to benchmark it against group-level independent component factorisation - a standard in fMRI network analysis - and found a high degree of concordance between the results produced by the two methods. We then applied it to the fetal data, where the standard approach has consistently failed to reveal spatially distributed networks. For the first time, and despite fundamental differences in signal characteristics between fetal and neonatal data, we detected network-like patterns with high spatial correspondence to neonatal functional networks. In particular, the FCMs approach efficiently recovered interhemispheric connections, a landmark feature of neonatal functional networks. Systematic organisation of interhemispheric fetal networks was observed; they tended to cluster along the brain midline but also were present in lateral sensorimotor and temporal areas as well as cortical limbic territories in ventral orbitofrontal cortex and temporal pole. By aligning fetal and neonatal connectomes, this study represents a crucial step towards supporting the biological veracity of observations made using fetal fMRI. Meanwhile, the concordance between FCMs and independent component factorisation in neonates prompts a re-evaluation of how inter-individual variability contributes to network structure inference in methods that ostensibly emphasise shared correlation patterns across subjects.