Developing a multi-modal neuroimaging-based BrainAge model across childhood.

Area of Science:

• Neuroscience
• Developmental Biology
• Biomarkers

Background:

• BrainAge models are promising biomarkers for childhood developmental brain health.
• Accurate models must capture dynamic neurodevelopmental changes across childhood.
• Distinguishing normative variance from pathological divergence is crucial for early intervention.

Purpose of the Study:

• To develop and evaluate BrainAge models reflecting distinct childhood developmental phases.
• To compare phase-specific and full-span models using multi-modal neuroimaging data.
• To identify optimal neuroimaging features for predicting age and clinical status in children.

Main Methods:

• Utilized multi-modal neuroimaging data from three pediatric cohorts (ages 4-13 years, n=1005).
• Developed twelve sex-stratified BrainAge models (Full-Span and Phase-Specific) using structural and functional features.
• Compared model performance in age prediction and subgroup discrimination, benchmarking against existing models.

Main Results:

• Phase-specific models incorporating structural (cortical thickness, subcortical volumes) and functional (network integration) features demonstrated superior age prediction.
• The best-performing model effectively distinguished between healthy and symptomatic pediatric subgroups.
• Findings support the development of higher temporal resolution BrainAge models aligned with developmental phases.

Conclusions:

• Phase-specific, multi-modal BrainAge models offer improved accuracy for assessing childhood neurodevelopment.
• These models serve as a proof-of-concept for advanced developmental biomarkers.
• Optimized BrainAge models can enhance early detection and intervention for developmental brain health.