On Stabilizing the Variance of Dynamic Functional Brain Connectivity Time Series

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Data Analysis

Background:

• Dynamic functional brain connectivity (dFC) using fMRI reveals brain network dynamics.
• Current dFC analysis often uses the Fisher transformation to stabilize correlation variance.
• The effectiveness of the Fisher transformation for fluctuating true correlations is unclear, impacting downstream analyses requiring stable variance or Gaussian distributions.

Purpose of the Study:

• To evaluate the efficacy of different variance stabilization strategies for dynamic functional connectivity time series.
• To compare the Fisher transformation, Box-Cox transformation, and a combined approach.
• To determine the optimal method for achieving stable variance and Gaussian-like distributions in dFC time series.

Main Methods:

• Simulations were used to model dynamic functional connectivity time series.
• Resting-state fMRI data was analyzed to assess real-world performance.
• The Fisher transformation, Box-Cox transformation, and a sequential Fisher-Box-Cox transformation were applied and compared.

Main Results:

• The Fisher transformation alone is suboptimal for stabilizing variance and can skew dFC time series away from a Gaussian distribution.
• Variance stabilization is crucial for analyses like clustering that assume or benefit from Gaussian distributions.
• A sequential application of the Box-Cox transformation after the Fisher transformation substantially improves variance stabilization and Gaussian properties of dFC time series.

Conclusions:

• The standard Fisher transformation is insufficient for preparing dynamic functional connectivity time series for many statistical analyses.
• Combining the Fisher transformation with the Box-Cox transformation offers a superior method for variance stabilization and achieving near-Gaussian distributions.
• This enhanced transformation strategy can improve the reliability and validity of downstream analyses in neuroimaging research.