Thank you for your help so far in helping me process our spinal cord data. We have written a paper and have received reviewer comments about the PAM50 template registration. Below is the reviewer comments:
How exactly was the PAM50 template registration performed? Linear or nonlinear registration? Was the spinal cord internal structure accounted for registration? Since the MTR and FA images were registered in the direction from their native space to the PAM50 template space, what data interpolation algorithm was used? For better fidelity of the MTR and FA metrics, would it make more sense to perform the registration in the opposite direction from PAM50 template to the native space? That way, data interpolation would not be needed.
Do you have any thoughts or ways to explain these comments in the response to the reviewers?
How exactly was the PAM50 template registration performed?
I suggest you add your analysis script as supplementary material or upload it on Github, so your analysis is fully transparent.
Linear or nonlinear registration?
A mix of both. The spinal cord was straightened as described in [De Leener, J Magn Reson Imaging 2017], followed by inferior-superior affine alignment based on vertebral levels. Then, the spinal cord centerline was aligned between the subject and the template using the center of mass of the spinal cord segmentation, which was followed by a non-linear within-plane BSplineSyN registration [Tustison et al].
Was the spinal cord internal structure accounted for registration?
The spinal cord internal structure was accounted for by assuming a linear deformation based on the outer shape of the spinal cord. Pros and cons of this strategy are discussed in [De Leener, Neuroimage 2017].
Since the MTR and FA images were registered in the direction from their native space to the PAM50 template space, what data interpolation algorithm was used? For better fidelity of the MTR and FA metrics, would it make more sense to perform the registration in the opposite direction from PAM50 template to the native space? That way, data interpolation would not be needed.
@sebake I need to know what you did. Both are possible. As stated above, if you publish your analysis script, I can tell you what you did and what to answer here.
thanks, so, based on your script, here is the answer:
Since the MTR and FA images were registered in the direction from their native space to the PAM50 template space, what data interpolation algorithm was used? For better fidelity of the MTR and FA metrics, would it make more sense to perform the registration in the opposite direction from PAM50 template to the native space? That way, data interpolation would not be needed.
Actually it was the other way around: the PAM50 was registered to the MTR and FA images, as can be seen in the analysis script (line 97 for MT data, line 146 for diffusion data). But regardless, the registration is bijective, meaning that the generated forward and backward warping fields can be respectively used to bring the template to the native space, and the native space to the template.
The command sct_warp_template, which comes after registration, brings the PAM50 template and atlases into the native space of the MTR and FA data, allowing to extract metrics without interpolation-related issue, as the reviewer rightfully pointed out.