Neonatal multi-modal cortical profiles predict 18-month developmental outcomes

Abstract

Developmental delays in infanthood often persist, turning into life-long difficulties, and coming at great cost for the individual and community. By examining the developing brain and its relation to developmental outcomes we can start to elucidate how the emergence of brain circuits is manifested in variability of infant motor, cognitive and behavioural capacities. In this study, we examined if cortical structural covariance at birth, indexing coordinated development, is related to later infant behaviour. We included 193 healthy term-born infants from the Developing Human Connectome Project (dHCP). An individual cortical connectivity matrix derived from morphological and microstructural features was computed for each subject (morphometric similarity networks, MSNs) and was used as input for the prediction of behavioural scores at 18 months using Connectome-Based Predictive Modeling (CPM). Neonatal MSNs successfully predicted social-emotional performance. Predictive edges were distributed between and within known functional cortical divisions with a specific important role for primary and posterior cortical regions. These results reveal that multi-modal neonatal cortical profiles showing coordinated maturation are related to developmental outcomes and that network organization at birth provides an early infrastructure for future functional skills.

Keywords: Brain development; Infant development; Morphometric similarity networks; Neonatal neuroimaging; Perinatal.

Fig. 1

Pipeline Morphometric Similarity Networks construction and behavioural analysis. a. Regions are defined using Voronoi tessellation of the cortical surface; b. A feature vector of averaged normalized values of cortical thickness (CT), mean curvature (MC), myelin index (MI), surface area (SA), fractional anisotropy (FA), mean diffusivity (MD), neurite density index (NDI) and orientation dispersion index (ODI) is derived for each region; c. Each pair of regions is correlated using Pearson's r, resulting in an individual similarity-based connectivity matrix. d. Network strength at a whole-network level and single-edges level is related to behavioural measures by means of association and prediction respectively.

Fig. 2

Prediction of Q-CHAT scores from neonatal MSNs using CPM- CPM. Plots of significant correlation between predicted and observed Q-CHAT scores (left) and results of null r values with permutation testing (right) using Connectome-based Predictive Modeling (CPM).

Fig. 3

Proportion of within-cluster edges involved in social-emotional networks. Proportion of edges included in successful prediction model of social-emotional outcomes connecting nodes within each of the seven clusters.

Fig. 4

Proportion of between-cluster edges involved in social-emotional networks. Proportion of edges included in Q-CHAT prediction model connecting nodes between clusters. On the left the positive network is shown and, on the right, the negative network.

Fig. 5

Scatterplots of whole-network average and behaviour. Plots of average MSN strength across the cortex against language, motor, cognitive and social-emotional measurements 18 months.

Fig. 6

Scatterplots of significant associations between within-cluster average network strength and developmental outcomes. Plots of significant associations between within-cluster average MSN strength and social-emotional, language and cognitive measures at 18 months. Top: insular & medial frontal cluster, bottom: somatosensory & auditory cluster.