MCAN: Multimodal Causal Adversarial Networks for Dynamic Effective Connectivity Learning From fMRI and EEG Data.

Area of Science:

• Neuroinformatics
• Computational Neuroscience
• Brain Network Analysis

Background:

• Dynamic effective connectivity (DEC) describes continuous brain activity over time.
• Learning DEC from multimodal neuroimaging data (fMRI, EEG) is a growing research area.
• Existing methods struggle to bridge the modality gap between fMRI and EEG for precise DEC learning.

Purpose of the Study:

• To propose a novel multimodal causal adversarial network (MCAN) for improved DEC learning.
• To address the limitations of current methods in integrating fMRI and EEG data for DEC analysis.
• To enhance the accuracy of brain state estimation and the identification of abnormal brain activity patterns.

Main Methods:

• Developed a multimodal causal adversarial network (MCAN) comprising a generator and a discriminator.
• Utilized an attention-guided multimodal causal generator to produce posterior signals and DEC networks.
• Employed an unsupervised multimodal causal discriminator to guide network updates via joint gradient calculation.

Main Results:

• MCAN demonstrated superior performance over state-of-the-art methods in learning DEC network structure on simulated data.
• The method effectively estimated brain states in simulated datasets.
• MCAN revealed abnormal brain activity patterns more effectively on real datasets, indicating strong application potential.

Conclusions:

• MCAN offers a robust framework for learning dynamic effective connectivity from multimodal neuroimaging data.
• The proposed approach effectively bridges the fMRI-EEG modality gap for more precise brain network analysis.
• MCAN shows significant potential for clinical applications in identifying and understanding abnormal brain activity.