Reliability and comparability of human brain structural covariance networks

Area of Science:

• Neuroimaging
• Brain anatomy
• Structural MRI

Background:

• Structural covariance analysis (SCA) is a key neuroimaging technique.
• SCA characterizes morphological correlations between brain regions in healthy subjects.
• Comparability and reliability of SCA results across datasets and scan parameters are under-investigated.

Purpose of the Study:

• To assess the comparability of SCA results between different datasets of healthy adults.
• To evaluate the reliability of SCA results across repeated scans, varying image resolutions, and FreeSurfer versions.
• To identify optimal practices for robust structural covariance analysis.

Main Methods:

• Compared structural covariance matrices from different datasets of age- and sex-matched healthy adults.
• Assessed reliability using repeated scan sessions, different image resolutions, and FreeSurfer versions.
• Estimated relative measurement error for different morphological measures (cortical thickness, surface area, volume).
• Utilized simulated data to investigate measurement error influences.

Main Results:

• Substantial differences in structural covariance were found between datasets, persisting after site correction.
• Differences were exacerbated by small sample sizes and most pronounced with cortical thickness.
• Reliability issues were observed across repeated scans, resolutions, and FreeSurfer versions.
• Cortical thickness exhibited the largest relative measurement error, impacting reliability.

Conclusions:

• Combining data across sites with differing parameters is not recommended due to significant result variability.
• Surface area and volume are more reliable morphological measures than cortical thickness for SCA.
• Large sample sizes (n≫30) are crucial for reliable structural covariance estimation.
• Explicitly measuring and modeling error covariance is necessary when combining sites.