Leveraging the adolescent brain cognitive development study to improve behavioral prediction from neuroimaging in...

Area of Science:

• Neuroscience
• Cognitive Science
• Developmental Psychology

Background:

• Neuroimaging is widely used to connect brain structure and function with human traits.
• Recent studies question the reproducibility of predicting cognitive traits using structural and resting-state functional MRI.
• Establishing reproducible brain-behavior associations is crucial for neurodevelopmental research.

Purpose of the Study:

• To determine the necessary replication sample sizes for detecting reproducible brain-behavior associations using neuroimaging data.
• To evaluate the effectiveness of univariate and multivariate methods across different imaging modalities.
• To leverage large-scale data from the Adolescent Brain Cognitive Development (ABCD) Study.

Main Methods:

• Utilized baseline data from thousands of children in the ABCD Study.
• Applied univariate and multivariate statistical methods to structural MRI, resting-state fMRI, and task-based fMRI data.
• Calculated required sample sizes for replication across different neuroimaging techniques and analytical approaches.

Main Results:

• Multivariate methods applied to high-dimensional neuroimaging data identified reproducible, lower-dimensional patterns of brain architecture.
• Robust correlations were found between these brain patterns and cognitive phenotypes.
• Replication required as few as 41 individuals for working memory task fMRI and approximately 100 for structural and resting-state MRI.
• Even with limited discovery samples, adequate power for replication was achieved with 66 subjects for multivariate prediction of cognition using working memory fMRI.

Conclusions:

• Neuroimaging, particularly using multivariate approaches, plays a vital role in translational neurodevelopmental research.
• Large-scale neuroimaging studies can effectively inform the design of smaller, reproducible studies.
• Findings highlight the potential for robust brain-behavior association discovery even with limited replication sample sizes.