Multivariate fNIRS response patterns to social information are increasingly discriminable from six to sixty months of age

Improving the geographic and economic inclusiveness of developmental neuroscience is an urgent concern, and functional near-infrared spectroscopy (fNIRS) is one tool extending the reach of neuroimaging beyond the communities closest to university laboratories. Where structural imaging facilities are unavailable, however, comparing fNIRS data across participants remains a challenge, especially as children’s head sizes vary widely with age. In this study, we describe a multivariate pattern analysis approach to describing fNIRS response patterns of infants and children (six- to sixty-months-old) from a low-income neighborhood of Dhaka, Bangladesh while they participated in a social cognition experiment (Perdue et al., 2019, Developmental Science). Instead of comparing the magnitude of hemodynamic responses in anatomical regions of interest, we use all channels simultaneously to compare changes in the discriminability between stimulus classes longitudinally over time (6-24 months, 36-60 months) and between groups (younger vs. older cohorts). From a sample of 53 to 74 children per age group, we find that a correlation-based, channel-space approach (Emberson et al., 2017, PLoS ONE) classifies fNIRS data more accurately with increasing age and is maximized by considering oxygenated and deoxygenated hemoglobin simultaneously. Using the brain response patterns from this sample to classify another, smaller sample of children (36-51 children per age group) from the same neighborhood, we achieve the same accuracy and age effects. These findings complement and extend the published univariate findings with finer-grained quantitative comparisons between ages, illustrating the power of multivariate approaches to understand developmental change without precise anatomical localization and in moderately sized samples.