Capturing brain-cognition relationship: Integrating task-based fMRI across tasks markedly boosts prediction and...

Area of Science:

• Neuroscience
• Cognitive Neuroscience
• Neuroimaging

Background:

• Estimating cognitive abilities using brain MRI has historically faced challenges in prediction accuracy and reliability.
• Individual differences in cognition are crucial for understanding human neuroscience.
• Previous research has shown limitations in predicting cognitive abilities from MRI data.

Purpose of the Study:

• To investigate if integrating whole-brain MRI signals across multiple modalities, including task-based functional MRI (tfMRI) and non-task MRI, can improve the prediction and reliability of cognitive abilities.
• To compare the predictive power of different sets of MRI modalities and integration approaches.
• To challenge the notion that tfMRI is unreliable for capturing individual cognitive differences.

Main Methods:

• Utilized data from the Human Connectome Project (n=873) including structural MRI, resting-state functional connectivity, and tfMRI from seven different tasks.
• Employed two multimodal MRI integration approaches: stacked and flat models, combined with 16 machine-learning algorithms.
• Directly compared predictive models using different combinations of MRI modalities.

Main Results:

• The stacked model integrating all modalities using Elastic Net achieved the highest prediction (r=0.57) and excellent test-retest reliability (ICC≈0.85) for general cognitive abilities.
• A stacked model integrating tfMRI across tasks significantly outperformed a model using only non-task modalities (r=0.56 vs. r=0.27), with high reliability (ICC≈0.83).
• The tfMRI-based model highlighted frontal and parietal regions and cognition-related tasks (working memory, relational processing, language), aligning with the parieto-frontal integration theory of intelligence.

Conclusions:

• Integrating tfMRI across tasks and combining it with other MRI modalities offers a powerful and reliable method for predicting individual cognitive abilities.
• The findings contradict the idea that tfMRI is unreliable, demonstrating its utility when analyzed appropriately across the whole brain and multiple tasks.
• This multimodal approach provides a more accurate and dependable estimation of cognitive abilities compared to non-task MRI modalities alone.