Dynamic Granger causality based on Kalman filter for evaluation of functional network connectivity in fMRI data

Area of Science:

• Neuroscience
• Signal Processing
• Computational Biology

Background:

• Understanding dynamic brain neural network interactions is crucial for neuroscience.
• Standard Granger causality methods using multivariate autoregressive (MAR) modeling are limited by the stationarity assumption.
• Functional magnetic resonance imaging (fMRI) time series are often non-stationary, violating this assumption.

Purpose of the Study:

• To propose a dynamic Granger causality approach in the frequency domain for functional network connectivity analysis in fMRI data.
• To address the limitations of time-invariant MAR models with non-stationary fMRI data.
• To improve the accuracy and robustness of brain connectivity assessments.

Main Methods:

• Utilized independent component analysis (ICA) to detect functional networks.
• Applied a dynamic Granger causality measure, generalized partial directed coherence (GPD), in the frequency domain.
• Combined forward and backward Kalman filters to enhance MAR model estimates for dynamic analysis.

Main Results:

• The dynamic approach demonstrated improved effectiveness and robustness compared to standard time-invariant MAR modeling.
• The frequency-domain analysis allowed for the identification of causal relations specific to different frequency components.
• The method was successfully applied to simulated data and real fMRI datasets from auditory sensorimotor and oddball discrimination tasks.

Conclusions:

• The proposed dynamic Granger causality method offers a more robust and accurate way to assess functional network connectivity in non-stationary fMRI data.
• This frequency-domain approach provides valuable insights into the dynamic interactions within brain networks.
• The method successfully distinguishes frequency-specific causal relationships relevant to experimental paradigms.