Combining Real-Time Neuroimaging With Machine Learning to Study Attention to Familiar Faces During Infancy: A Proof of Principle Study

Abstract

Looking at caregivers’ faces is important for early social development, and there is a concomitant increase in neural correlates of attention to familiar versus novel faces in the first 6 months. However, by 12 months of age brain responses may not differentiate between familiar and unfamiliar faces. Traditional group-based analyses do not examine whether these ‘null’ findings stem from a true lack of preference within individual infants, or whether groups of infants show individually strong but heterogeneous preferences for familiar versus unfamiliar faces. In a preregistered proof-of-principle study, we applied Neuroadaptive Bayesian Optimisation (NBO) to test how individual infants’ neural responses vary across faces differing in familiarity. Sixty-one 5–12-month-olds viewed faces resulting from gradually morphing a familiar (primary caregiver) into an unfamiliar face. Electroencephalography (EEG) data from fronto-central channels were analysed in real-time. After the presentation of each face, the Negative central (Nc) event-related potential (ERP) amplitude was calculated. A Bayesian Optimisation algorithm iteratively selected the next stimulus until it identified the stimulus eliciting the strongest Nc for that infant. Attrition (15%) was lower than in traditional studies (22%). Although there was no group-level Nc-difference between familiar versus unfamiliar faces, an optimum was predicted in 85% of the children, indicating individual-level attentional preferences. Traditional analyses based on infants’ predicted optimum confirmed NBO can identify subgroups based on brain activation. Optima were not related to age and social behaviour. NBO suggests the lack of overall familiar/unfamiliar-face attentional preference in middle infancy is explained by heterogeneous preferences, rather than a lack of preference within individual infants.