Multivariate dynamical systems-based estimation of causal brain interactions in fMRI: Group-level validation using...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Brain Imaging Analysis

Background:

• Causal estimation methods are increasingly used for functional brain network analysis in fMRI.
• Concerns exist regarding the validity of current fMRI causal estimation techniques.

Purpose of the Study:

• To validate Multivariate Dynamical Systems (MDS), a state-space method for estimating dynamic causal interactions in fMRI.
• To apply MDS to human connectome project (HCP) data for investigating a fronto-cingulate-parietal control network during working memory tasks.
• To develop and apply novel stability analysis for robust causal interaction identification in experimental fMRI data.

Main Methods:

• Validation of MDS using benchmark simulations and a realistic stochastic neurophysiological model.
• Application of MDS to human connectome project (HCP) fMRI data.
• Conducting novel stability analysis to ensure robustness of identified causal interactions.

Main Results:

• MDS demonstrated high accuracy in recovering dynamic causal interactions, achieving an Area Under the ROC Curve (AUC) > 0.7 for benchmark datasets and > 0.9 for neurophysiological model datasets.
• Bootstrap procedures on experimental fMRI data revealed a stable causal influence pattern from the anterior insula to other nodes within the fronto-cingulate-parietal network.
• MDS proved effective in estimating dynamic causal interactions, showing favorable AUC, sensitivity, and false positive rates compared to existing methods.

Conclusions:

• MDS accurately estimates causal interactions in fMRI data, supported by simulation and stability analysis.
• Neurophysiological models and stability analysis offer a robust framework for validating computational methods in fMRI causal interaction estimation.
• The right anterior insula was identified as a causal hub during working memory tasks within the studied network.