Multimodal neuroimaging data boosts the prediction of multifaceted cognition

Area of Science:

• Neuroscience
• Computational Biology
• Psychology

Background:

• Predictive modeling linking brain patterns to behavior is growing.
• Multimodal neuroimaging data integration aims to enhance behavioral prediction accuracy.
• The utility of multimodal integration in this context remains debated due to mixed findings.

Purpose of the Study:

• To systematically evaluate the necessity and benefit of multimodal integration for brain-based behavior prediction.
• To determine optimal levels of data integration across different age groups and behavioral domains.
• To identify specific neuroimaging features that contribute most to predictive accuracy.

Main Methods:

• Utilized 3 large datasets spanning diverse age ranges.
• Employed 25-33 feature types from various neuroimaging modalities (e.g., functional, structural, diffusion).
• Assessed 21 behavioral measures across different domains, building predictive models with increasing multimodal integration.

Main Results:

• Prediction performance saturates after integrating a limited number of feature types.
• Complex cognitive scores generally require higher levels of multimodal integration.
• Functional connectivity features are often sufficient, particularly in young adults, but aging may necessitate structural and diffusion data.

Conclusions:

• Multimodal integration offers benefits but has diminishing returns; focusing on key features is efficient.
• Functional connectivity is broadly applicable, but effective connectivity is crucial for aging populations.
• Exploring alternatives beyond functional imaging, especially for aging, is vital for advancing behavior prediction.