Multi-Paradigm fMRI Fusion via Sparse Tensor Decomposition in Brain Functional Connectivity Study.

Area of Science:

• Neuroimaging
• Machine Learning
• Cognitive Neuroscience

Background:

• Functional magnetic resonance imaging (fMRI) is crucial for understanding individual differences in cognition.
• Integrating multiple fMRI datasets offers complementary information but faces data fusion challenges.
• Existing methods often analyze datasets separately, missing cross-modal interactions.

Purpose of the Study:

• To propose a novel sparse tensor decomposition method for fusing multiple task-stimulus fMRI data.
• To simultaneously consider relationships across subjects and modalities for improved data integration.
• To enhance the prediction of cognitive behaviors using integrated fMRI data.

Main Methods:

• Modeled functional network connectivity (FNC) from multi-paradigm fMRI data as a third-order tensor.
• Applied sparse tensor decomposition with L2,1-norm regularization to extract shared features across modalities.
• Validated the method on three-paradigm fMRI datasets from the Philadelphia Neurodevelopmental Cohort (PNC) study.

Main Results:

• The proposed method outperformed competing approaches in predicting cognitive behaviors (Wide Range Achievement Test - WRAT).
• Successfully identified functional network connectivity (FNC) associated with cognitive abilities.
• Discovered default mode network (DMN) connectivity patterns across paradigms and between DMN and visual domains during emotion tasks.

Conclusions:

• Sparse tensor decomposition offers an effective approach for multi-modal fMRI data fusion.
• This method enhances the prediction of cognitive traits by leveraging cross-subject and cross-modal information.
• The findings provide insights into the neural underpinnings of cognitive behaviors via FNC.