Infant Cognitive Growth Prediction via Centile-based Brain Development and Meta Matching

The infant brain undergoes heterogeneous, nonlinear development across multiple phases, with cognitive progression exhibiting diverse subdomain patterns. Traditional regression and artificial intelligence (AI)-based models linking brain development to cognitive outcomes often overlook the dynamic nature of brain maturation and cognitive progression. Other challenges arise from limited sample sizes and unmeasured early lifestyle and environmental exposures. In our prior work, we developed a computationally efficient, easily interpretable centile-based explainable-AI toolbox, which has been proven effective in the early diagnosis and prognosis of knee osteoarthritis. In this work, we integrate this centile-based AI with the recently developed cross-cohort meta-matching techniques to predict infant cognitive growth. We will leverage four infant brain multimodal magnetic resonance imaging (MRI) databases to derive individual centile score maps from structural and resting-state functional MRIs, separating influences into anatomical and functional domains. Additionally, we will incorporate latent neural signatures (LNS) associated with “unseen” non-infant, non-imaging phenotypes—such as adolescent fluid intelligence, social behavior, or PM2.5 pollution exposure—mapped from two large-scale children and adolescent datasets. By identifying key biomarkers, we aim not only to elucidate the specific anatomical and functional brain regions influencing distinct infant cognitive subdomains but also to determine whether LNS metrics mapped from childhood/adolescence can serve as proxies for underlying cognitive, behavioral, or environmental exposures impacting infant cognitive development. This meta-matching approach is a novel application with significant potential to uncover previously hidden implications, linking insights from children and adolescent data to infant brain measures. Two specific aims guide our project. Aim 1: construct a computationally efficient, easily interpretable centile score-based explainable- AI toolbox and assess the improvement in infant cognitive prediction with centile score-based imaging biomarkers. Specifically, we fit developmental trajectories and obtain individual centile score maps for all brain imaging measures, evaluating them using various machine learning approaches to improve predictions of infant cognitive development. Aim 2: Assess whether integrating unseen non–brain-imaging phenotypes via meta-matching—including adverse childhood experiences, latent lifestyle and environmental exposures and psychiatric disorders---can further enhance cognitive outcome predictions. With the transparency of explainable-AI, this project will not only provide a holistic understanding of infant cognitive development in relation to brain structure and function but also establish a novel framework to link brain regions activated by childhood and adolescent cognitive, behavioral and environmental factors, as well as psychiatric disorders, to infant cognitive outcomes. Our expert team, with diverse specializations in neuroimaging, infant brain development, advanced statistical modeling and machine learning, is committed to sharing findings and methodologies with the broader scientific community. This dissemination will enhance replicability, foster innovation, and expand the applicability of insights in early brain and cognitive development. Furthermore, the psychiatric disorders incorporated in meta-matching in Aim 2 hold clinical significance, potentially aiding in early intervention strategies for neurodevelopmental conditions. Project Number: 1R21HD120911-01 | Fiscal Year: 2025 | NIH Institute/Center: Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) | Principal Investigator: Tengfei Li | Institution: UNIV OF NORTH CAROLINA CHAPEL HILL, CHAPEL HILL, NC | Award Amount: $413,509 | Activity Code: R21 | Study Section: Analytics and Statistics for Population Research Panel A Study Section[ASPA] View on NIH RePORTER: https://reporter.nih.gov/project-details/1R21HD12091101