Comparison of spatial normalization strategies of diffusion MRI data for studying motor outcome in subacute-chronic...

A common means of studying motor recovery in stroke patients is to extract Diffusion Tensor Imaging (DTI) parameters from the corticospinal tract (CST) and correlate them with clinical outcome scores. To that purpose, conducting group-level analyses through spatial normalization has become a popular approach. However, the reliability of such analyses depends on the accuracy of the particular registration strategy employed. To date, most studies have employed scalar-based registration using either high-resolution T1 images or Fractional Anisotropy (FA) maps to warp diffusion data to a common space. However, more powerful registration algorithms exist for aligning major white matter structures, such as Fiber Orientation Distribution (FOD)-based registration. Regardless of the strategy chosen, automatic normalization algorithms are prone to distortions caused by stroke lesions. While lesion masking is a common means to lessen such distortions, the extent of its effect on tract-related DTI parameters and their correlation with motor outcome has yet to be determined. Here, we aimed to address these concerns by first investigating the effect of common T1 and FA-based registration as well as novel FOD-based registration algorithms with and without lesion masking on lesion load and DTI parameter extraction of the CST in datasets typically acquired for subacute-chronic and acute stroke patients. Second, we studied how differences in these procedures influenced correlation strength between CST damage (through DTI parameters) and motor outcome. Our results showed that, for high-quality subacute-chronic stroke data, FOD-based registration captured significantly higher lesion loads and significantly larger FA asymmetries in the CST. This was also associated with significantly stronger correlations in motor outcome with respect to T1 or FA-based registration methods. For acute data acquired in a clinical setting, there were few observed differences, suggesting that commonly employed FA-based registration is appropriate for group-level analyses.