Sliding window functional connectivity inference with nonstationary autocorrelations and cross-correlations

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Statistical Analysis

Background:

• Functional connectivity (FC) traditionally assumes stationarity, but brain activity is dynamic.
• Dynamic FC (dFC) captures time-varying brain network synchrony.
• Current dFC methods using sliding windows often overlook time-varying signal properties.

Purpose of the Study:

• To address the limitations in dynamic FC analysis by accounting for non-stationarity.
• To propose a novel variance estimation method for sliding window correlations in fMRI.
• To improve the statistical inference of dynamic brain connectivity.

Main Methods:

• Demonstrated non-stationarity in autocorrelation (ACF) and cross-correlation (XCF) functions using in vivo resting-state fMRI data.
• Developed a new variance estimation approach for sliding window correlations that incorporates dynamic ACF and XCF.
• Validated the method through simulations, comparing its performance against existing techniques.

Main Results:

• Confirmed significant non-stationarity in both ACF and XCF of fMRI time series.
• The proposed variance estimation method accurately captures dynamic connectivity changes.
• Simulations showed the proposed method's superiority in inferring dynamic connectivity compared to standard approaches.

Conclusions:

• Accurate variance estimation considering dynamic signal properties is crucial for reliable dFC analysis.
• The novel method enhances statistical inference, potentially reducing false positives and negatives in dFC studies.
• This work provides a more robust framework for understanding dynamic brain networks.