Spatio-temporal correlations from fMRI time series based on the NN-ARx model

Area of Science:

• Neuroscience
• Data Science
• Statistical Modeling

Background:

• Functional magnetic resonance imaging (fMRI) generates high-dimensional time series data.
• Characterizing spatio-temporal correlations in brain activity is crucial for understanding brain function.
• Existing methods often involve static models that may oversimplify dynamic brain processes.

Purpose of the Study:

• To introduce a novel statistical approach for characterizing spatio-temporal correlation structures in fMRI data.
• To extend correlation analysis by incorporating dynamic modeling of fMRI time series.
• To develop methods for assessing brain connectivity based on dynamic correlation patterns.

Main Methods:

• Developed a Nearest-Neighbors AutoRegressive model with external inputs (NN-ARx) for data whitening.
• Defined measures of dependency based on correlations between model innovations at multiple time lags.
• Employed voxel-pair summarization of correlations, avoiding regional averaging to preserve information.

Main Results:

• The NN-ARx approach provides an extension to standard correlation methods by using dynamic mean-correction.
• Summarized correlations of innovations capture dependencies between brain regions without information loss from averaging.
• Applied the method to fMRI data from visual stimuli experiments, demonstrating its utility.

Conclusions:

• The NN-ARx approach offers a powerful tool for statistical characterization of brain functioning from fMRI data.
• This method enhances the analysis of spatio-temporal correlations and brain connectivity.
• Further investigation into the potential and limitations of the NN-ARx approach is warranted.