Optimizing the measurement of sample entropy in resting-state fMRI data

Area of Science:

• Neuroimaging
• Complexity Science
• Biophysics

Background:

• Brain signal complexity, measured by sample entropy, is crucial for understanding brain-based disorders.
• fMRI-based sample entropy analysis faces challenges due to motion artifacts and parameter selection.
• Standard preprocessing methods like scrubbing may be unsuitable for entropy measurement.

Purpose of the Study:

• To develop and validate a novel windowing approach for fMRI data to accurately estimate sample entropy.
• To create user-friendly utilities for optimal parameter selection (matching length m, error tolerance r).
• To generate whole-brain dense entropy maps and assess reproducibility.

Main Methods:

• A novel windowing technique was developed to select and concatenate low-motion segments of fMRI data.
• Autoregressive models and grid search utilities were created for optimal selection of sample entropy parameters (m and r).
• The methods were applied to ABCD study data, and results were compared to standard approaches, assessing reproducibility via intraclass correlation.

Main Results:

• The windowing procedure effectively mitigated the inverse correlation between entropy and head motion.
• Sample entropy values calculated using the windowed approach demonstrated good reproducibility between early and later scan sessions.
• Regional variability in reproducibility was observed across the brain.

Conclusions:

• An optimized, adaptable method for fMRI sample entropy measurement has been developed, addressing motion and parameter selection challenges.
• The developed methods and associated R package ('powseR') offer a valuable tool for advancing neuroimaging research.
• Recommendations for fMRI data acquisition and analysis are provided to enhance sample entropy measurement.