Individual variation underlying brain age estimates in typical development.

Area of Science:

• Neuroscience
• Developmental Psychology
• Computational Psychiatry

Background:

• Typical brain development is a protracted process crucial for cognitive function.
• Deviations in brain maturation are linked to neurodevelopmental and psychiatric disorders.
• Machine learning models estimating 'brain age' offer insights into developmental trajectories but can be complex and difficult to interpret.

Purpose of the Study:

• To investigate the specific cortical features contributing to brain age prediction errors on an individual level.
• To utilize model explanation techniques to enhance the interpretability of brain age models.
• To examine the consistency and variability of these features across different machine learning approaches.

Main Methods:

• A cohort of 768 typically developing children (aged 3-21 years) was analyzed.
• Three distinct machine learning approaches were used to model brain development.
• SHAP (SHapley Additive exPlanations), a model-agnostic technique, identified sample-specific feature importance for regional cortical metrics.

Main Results:

• Brain age prediction and contributing cortical features were consistent across model types, reflecting known developmental patterns.
• Despite overall consistency, significant between-subject variation was observed in the specific features driving individual brain age prediction errors.
• No association was found between brain age error and cognitive performance in this typically developing sample.

Conclusions:

• Brain age estimation based on cortical development is robust across different modeling strategies.
• Individual-level analysis reveals substantial variability in the neuroanatomical features underlying brain age prediction errors.
• Further research is needed to understand the implications of this individual variability in brain development.