Modeling differences in neurodevelopmental maturity of the reading network using support vector regression on functional connectivity data

Abstract

AbstractThe construction of brain growth charts trained to predict age based on morphometric and/or functional properties of the brain (‘brain-age prediction’) has the potential to inform neuroscientists and clinicians alike about children’s neurodevelopmental trajectories. When applied to both typically- and atypically-developing populations – such as those with specific learning disorders – results may be informative as to whether a particular condition is associated with atypical maturation of specific brain networks. Here, we focus on the relationship between reading disorder (RD) and maturation of functional connectivity (FC) patterns in the prototypical reading/language network across development using a cross-sectional sample of N = 742 participants aged 6-21 years. A support vector regression model is trained to predict chronological age from FC data derived from (1) a whole-brain model, as well as (2) multiple ‘reduced’ models, which are trained on FC data generated from a successively smaller number of regions in the brain’s reading network. We hypothesized that the trained models would show systematic underestimation of brain network maturity (i.e., lower FC-based age predictions) for poor readers, particularly for the models trained with reading/language regions. Exploratory results demonstrated that the most important whole-brain ROIs and connections are derived from the dorsal attention and somatosensory motor networks. Comparisons of the different models’ predictions reveal that while the whole-brain model outperforms the others in terms of overall prediction accuracy, all models are effective at predicting brain maturity, including the one trained with the smallest amount of FC data. In addition, all models demonstrate some degree of moderation in the reliability of their age predictions as a function of reading ability, with predictions for both poor and exceptional readers being more accurate relative to those for typical readers.Key PointsA machine learning model trained to predict age from functional connectivity data has better predictive utility for children/adolescents with either low OR high reading ability, and performs less well for those with average reading abilityBoth impaired and exceptional reading may be characterized as having lower variance in connectivity patterns on the population level compared to typical readers (rather than having delayed/accelerated network maturation)Neural markers of reading and language are likely diffusely represented in the brain