Impact of data processing varieties on DCM estimates of effective connectivity from task-fMRI

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Connectivity Analysis

Background:

• Effective connectivity (EC) describes directed influences between brain regions, often estimated using dynamic causal modeling (DCM) on fMRI data.
• The impact of preprocessing choices on task-evoked EC estimates remains underexplored, posing challenges for result reproducibility.

Purpose of the Study:

• To investigate how variations in data processing impact task-evoked EC estimates derived from DCM.
• To identify which processing steps most influence DCM results for brain networks involved in spatial conflict.

Main Methods:

• Examined the effects of global signal regression (GSR), general linear model (GLM) design (block/event-related), activation contrast type, and significance thresholding on DCM.
• Employed parametric empirical Bayes (PEB) and Bayesian data comparison (BDC) to analyze group-averaged EC differences across processing conditions.

Main Results:

• Data processing choices led to substantial variations in group-averaged task-evoked EC patterns.
• GLM design and activation contrast type significantly affected EC estimates and parameter certainty, as shown by PEB and BDC.
• GSR and significance thresholding had minimal impact on EC estimates.

Conclusions:

• Task-evoked EC estimates from DCM are sensitive to data processing decisions.
• Event-related GLM designs offer greater sensitivity to EC modulations but lower parameter certainty compared to block designs.
• Researchers should carefully select processing steps and consider parallel analyses to ensure robust and reliable EC findings.