Functional connectivity: shrinkage estimation and randomization test.

Area of Science:

• Neuroscience
• Statistics
• Signal Processing

Background:

• Estimating functional connectivity in multivariate time series is crucial for understanding complex systems.
• Partial coherence is a key measure, but its estimation faces numerical instability with large, correlated datasets.
• Existing methods can yield highly variable partial coherence estimates.

Purpose of the Study:

• To develop robust statistical methods for estimating functional connectivity.
• To introduce a novel shrinkage-based estimator for improved numerical stability and accuracy.
• To create a randomization method for testing differences in functional connectivity across conditions.

Main Methods:

• Characterizing functional connectivity using partial coherence, estimated via the inverse spectral density matrix.
• Proposing a shrinkage-based estimator as a weighted average of periodogram and identity matrix estimators.
• Developing a frequency-specific weight for the shrinkage estimator to minimize mean-squared error.
• Implementing a randomization method for comparing functional connectivity networks.

Main Results:

• The proposed shrinkage estimator demonstrates greater computational stability compared to traditional smoothing methods.
• The shrinkage estimator achieves a lower mean squared error in partial coherence estimation.
• Numerical experiments confirm the effectiveness of the new methods.
• Analysis of EEG data during a hand movement task showcases practical application.

Conclusions:

• The developed statistical methods offer a more stable and accurate approach to functional connectivity analysis.
• The shrinkage estimator effectively addresses numerical instability issues in partial coherence estimation.
• These advancements have significant implications for analyzing complex time series data, including neuroimaging.