Cognitive and personality trait prediction using activation maps and temporal dynamics derived from resting-state...

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Machine Learning

Background:

• Predicting behavioral and personality traits from neuroimaging data is challenging due to complex brain dynamics.
• Current trait prediction models using neurocomputations show moderate accuracy.
• Effective modeling of high-dimensional brain data is crucial for advancing neuroscience.

Purpose of the Study:

• To propose a unified framework for predicting cognitive and personality traits using resting-state functional MRI (rs-fMRI).
• To leverage both spatial and temporal information from rs-fMRI for enhanced trait prediction.
• To improve the accuracy of predicting traits like fluid intelligence and extraversion.

Main Methods:

• Developed a novel framework integrating spatial and temporal information from rs-fMRI.
• Estimated task-activation maps from rs-fMRI to capture spatial brain activity.
• Utilized MultiRocket for efficient time-series feature extraction to capture temporal dynamics.
• Fused spatial and temporal features into a unified representation for an ensemble prediction model.

Main Results:

• Achieved state-of-the-art prediction correlations up to 0.5284 on the HCP dataset.
• Demonstrated superior predictive performance for cognitive traits (reading ability, fluid intelligence, processing speed).
• Showcased effectiveness in predicting personality traits (openness to experience, extraversion).

Conclusions:

• The proposed spatio-temporal modeling framework significantly enhances trait prediction from rs-fMRI.
• Integrating spatial and temporal brain dynamics is vital for understanding brain-behavior relationships.
• This approach offers a promising direction for personalized neuroscience and mental health applications.