Insight and inference for DVARS.

Area of Science:

• Neuroimaging
• Biophysics
• Data Science

Background:

• Resting-state functional Magnetic Resonance Imaging (rs-fMRI) is sensitive to artifacts and nuisance variations.
• Current diagnostic measures like DVARS lack concrete interpretation for distinguishing good from bad scans.
• Preprocessing and artifact identification are crucial for reliable rs-fMRI analysis.

Purpose of the Study:

• To develop a novel method for decomposing rs-fMRI data variability.
• To provide a more interpretable and robust diagnostic framework for rs-fMRI quality control.
• To enable better comparisons of rs-fMRI datasets across different preprocessing steps and subjects.

Main Methods:

• Introduced a sum of squares decomposition of the 4D rs-fMRI dataset into D-var, S-var, and E-var components.
• Extended the partitioning to global and non-global signals, creating a rs-fMRI DSE table.
• Developed a null sampling distribution for DVARS-squared and robust estimation methods for computing DVARS p-values.

Main Results:

• DVARS was shown to be one component (D-var) of a larger variability decomposition.
• The DSE table decomposes total and global rs-fMRI variability into fast, slow, and edge components.
• Expected values for each component were determined under nominal models, revealing scaling relationships and the impact of temporal autocorrelation.

Conclusions:

• The proposed DSE table, diagnostic time series, images, and p-values offer a succinct summary of rs-fMRI dataset quality.
• This framework enhances the interpretability of DVARS and provides robust methods for artifact detection.
• The findings support improved quality assessment and comparability of rs-fMRI data.