Background
Challenges with child neuroimaging constrain research into developing social processing.
Past developmental studies using naturalistic stimuli do not map content of the movie to responses in the brain.
We densely label a naturalistic movie and apply an encoding model to link its social content to neural activity in children as young as three.
Figure 2: Analysis overview. Social (pink) and motion energy (blue) features for each 2 second segment of Pixar’s ”Partly Cloudy” were linked neural activity in adults and 5 year olds who watched the movie. 80% of movie data was used to learn model beta weights. The learned weights were then used to predict fMRI activity in held out movie data. Predictions were correlated to actual activity at held out time points and averaged across subjects within each age group to measure prediction performance.
Methods
Data collection: 122 children aged 3-12, and 33 adult participants watched the Pixar short film “Partly Cloudy” while undergoing fMRI (Richardson et al., 2018).
Feature labeling: 3 human raters labeled faces, social interaction, theory of mind, valence, arousal for each TR. Motion energy features were also extracted for each TR.
Encoding model: Banded ridge regression (Nunez-Elizalde et al, 2019) was used to generate model beta weights for each voxel. Prediction for held out timepoints was correlated to true activity at those timepoints to generate prediction scores.
Figure 3: Feature prevalence and feature correlation. Faces and social interactions were present for the majority of the movie. Social features were correlated with each other, with the strongest correlation a negative correlation between valence and arousal. Motion features were not correlated with social features.
Results
Social features had greater predictive power in adults along pSTS. Motion features, meanwhile, showed comparable prediction performance across age groups.
Figure 4: Encoding model results and differences, projected onto the surface. Prediction performance were comparably high across age groups for motion features, with limited differences in the ventral stream (with the exception of ages 3-4). By contrast, for the social features, differences were strongest in the right pSTS for 3-4 year olds, 5 year olds, and 7 year olds.
Conclusions
An encoding model trained on social features of the movie has greater predictive power in adults than in children, particularly along the right pSTS.
Prediction performance is comparable across age groups for a model that uses only perceptual features of the stimulus.
These results link social content of the movie to neural activity, and show that differences across age groups cannot be explained by differences in data quality or attention alone.
References
Nunez-Elizalde, A. O., Huth, A. G., & Gallant, J. L. (2019). Voxelwise encoding models with non-spherical multivariate normal priors. neuroimage. NeuroImage, 197, 482–492.
Richardson, H., Lisandrelli, G., Riobueno-Naylor, A., & Saxe, R. R. (2018). Development of the social brain from age three to twelve years. Nature Communications, 9, 1027.
Developmental differences in social brain responses during movie viewing
Angira Shirahatti & Leyla Isik, Johns Hopkins University Department of Cognitive Science
Figure 1: Full poster, presented on August 27, 2022