Less is more: balancing noise reduction and data retention in fMRI with data-driven scrubbing

Area of Science:

• Neuroimaging
• Neuroscience
• Data Analysis

Background:

• Functional MRI (fMRI) data are susceptible to artifacts from head motion, respiration, and scanner noise.
• These artifacts compromise data quality, affecting statistical analysis accuracy and power.
• Current scrubbing techniques, particularly motion scrubbing, have limitations like arbitrary thresholding and high data exclusion rates.

Purpose of the Study:

• To introduce and evaluate "projection scrubbing", a novel data-driven artifact removal method for fMRI.
• To compare the performance of projection scrubbing against traditional motion scrubbing and DVARS.
• To assess scrubbing methods based on data retention and impact on functional connectivity validity, reliability, and identifiability.

Main Methods:

• Developed "projection scrubbing", a data-driven method using statistical outlier detection and dimension reduction (e.g., ICA).
• Compared projection scrubbing, DVARS, and motion scrubbing on fMRI datasets.
• Evaluated methods using metrics of data retention, functional connectivity validity, reliability, and fingerprinting.

Main Results:

• Stringent motion scrubbing degraded data validity and reliability while offering minimal improvements in fingerprinting.
• Data-driven methods, including projection scrubbing, improved fingerprinting without compromising validity or reliability.
• Projection scrubbing and DVARS excluded significantly fewer data volumes/sessions than motion scrubbing.

Conclusions:

• Data-driven scrubbing methods enhance fMRI data retention and quality without negatively impacting downstream analyses.
• Projection scrubbing offers a superior alternative to motion scrubbing for artifact removal in fMRI.
• Improved data retention via data-driven scrubbing has significant implications for increasing sample sizes in population neuroscience studies.