AbstractBackgroundBrain neuromaturation can be indexed using brain predicted age difference (BrainPAD), a metric derived by the application of machine learning (ML) algorithms to neuroimaging. Previous studies in youth have been limited to a single type of imaging data, single ML approach, or specific psychiatric condition. Here, we use multimodal neuroimaging and an ensemble ML algorithm to estimate BrainPAD and examine its relationship with broad measures of symptoms and functioning in youth.MethodsWe used neuroimaging from eligible participants in the Healthy Brain Network (HBN, N = 498). Participants with a Child Behavior Checklist Total Problem T-Score < 60 were split into training (N=215) and test sets (N=48). Morphometry estimates (from structural MRI), white matter connectomes (from diffusion MRI), or both were fed to an automated ML pipeline to develop BrainPAD models. The most accurate model was applied to a held-out evaluation set (N=249), and the association with several psychometrics was estimated.ResultsModels using morphometry and connectomes together had a mean absolute error of 1.16 years, outperforming unimodal models. After dividing participants into positive, normal, and negative BrainPAD groups, negative BrainPAD values were associated with more symptoms on the Child Behavior Checklist (negative=71.6, normal 59.0, p=0.011) and lower functioning on the Children’s Global Assessment Scale (negative=49.3, normal=58.3, p=0.002). Higher scores were associated with better performance on the Flanker task (positive=62.4, normal=52.5, p=0.006).ConclusionThese findings suggest that a multimodal approach, in combination with an ensemble method, yields a robust biomarker correlated with clinically relevant measures in youth.
Maturity of brain structures and white matter connectomes, and their relationship with psychiatric symptoms in youth
