Self-Navigated, Retrospective, Data-Consistent Motion Correction for MPnRAGE

To extend and automate a data-consistent, self-navigated motion-correction method for 3D radial T1-weighted imaging.

This method incorporated rigid-body motion effects into the forward model, solving for parameters that maximize consistency with the data. The method was tested on five datasets with a range of motion types and severities. A separate collection of datasets was used to study the effect that the method has on the test-retest reliability of cortical thickness estimates.

Image quality was improved across a wide range of distinct motion types, including some cases that would have been unusable if left uncorrected. The error-based weighting scheme and the increased timing resolution afforded by the proposed method were especially useful in cases of extreme and rapid motions. Moreover, the method improved test-retest reliability of cortical thickness measures in pediatric subjects, decreasing the average coefficient of variation from $2.73\%\pm 1.75\%$ in uncorrected images (with freesurfer failing on one subject) down to $0.88\%\pm 0.21\%$ for images corrected at $\sim 2s$ timing resolution and $0.79\%\pm 0.16\%$ when corrected at faster temporal rates.

This method was found to be effective when used on T1-weighted radial data, both qualitatively and quantitatively. The fine-scale timing resolution and error-based weighting afforded by this technique will likely provide only a small benefit, unless one is investigating motion-prone populations or is searching for a very small effect size.