Replication and Refinement of Brain Age Model for adolescent development.

Area of Science:

• Neuroscience
• Developmental Psychology
• Machine Learning

Background:

• The brain age gap, the difference between chronological and estimated brain age, is a potential biomarker for neurodevelopment and neuropsychiatric conditions.
• Accurate brain age estimation is crucial, but the generalizability of existing models requires further testing, especially in developmental cohorts.

Approach:

• Replicated existing brain age models on the Adolescent Brain Cognitive Development (ABCD) cohort across baseline and year-two follow-up.
• Developed and validated a refined machine learning model combining broad and granular age prediction for improved robustness.
• Evaluated model performance using mean absolute error and assessed the association of the brain age gap with cognitive measures.

Key Points:

• Existing pre-trained brain age models struggled to precisely estimate age variation within a narrow developmental range in the ABCD cohort.
• The refined model achieved high accuracy, with mean absolute errors of 0.49 and 0.48 years on baseline and year-two data, respectively.
• The refined model's brain age gap significantly correlated with information processing speed and verbal comprehension in adolescents.

Conclusions:

• A refined machine learning approach enhances the robustness and accuracy of brain age prediction in developing individuals.
• The brain age gap, as estimated by the refined model, serves as a meaningful indicator of cognitive function during adolescence.
• This study highlights the potential of advanced machine learning for understanding neurodevelopmental trajectories and associated cognitive outcomes.