A multicohort geometric deep learning study of age dependent cortical and subcortical morphologic interactions for...

Area of Science:

• Neuroscience
• Cognitive Science
• Artificial Intelligence

Background:

• The relationship between human brain structure and cognitive function, particularly fluid intelligence (Gf), is complex and not fully understood across developmental stages.
• Existing hypotheses often link Gf to prefrontal and parietal cortex, but a comprehensive understanding is lacking.

Purpose of the Study:

• To develop and validate a novel graph convolutional neural network (gCNN) for analyzing localized brain anatomy and predicting fluid intelligence (Gf).
• To investigate how brain structure-Gf relationships differ between childhood and adulthood using two independent cohorts.

Main Methods:

• Extracted morphologic information from T1-weighted MRIs of cortical and subcortical structures in children (ABCD study) and adults (HCP study).
• Utilized a novel graph convolutional neural network (gCNN) model to predict Gf based on extracted brain morphology.
• Compared the predictive performance of the gCNN model against state-of-the-art methods using brain measures.

Main Results:

• The gCNN model combining cortical and subcortical surfaces achieved the highest Gf prediction accuracy in both the ABCD (R=0.314) and HCP (R=0.454) datasets.
• This predictive performance surpassed existing state-of-the-art methods for Gf prediction from brain measures.
• Key brain regions driving Gf prediction included the amygdala, hippocampus, nucleus accumbens, and temporal, parietal, and cingulate cortex.

Conclusions:

• The findings suggest a broader neuroanatomic basis for fluid intelligence (Gf) than previously assumed, extending beyond the prefrontal and parietal cortex.
• The results indicate that brain regions involved in emotion, motivation, and decision-making play a significant role in Gf.
• This study reframes the understanding of brain morphology-Gf relationships by incorporating subcortical and limbic structures, with implications for both developmental neuroscience and artificial intelligence applications.