Hello,

I am not sure what to use for the Procrusted template for the ParaToSPHARMMesh step of the pipeline.

For the flipTemplate for cases after the first, I used the _SPHARM.coef file computed from ParaToSPHARM. Does it matter if it is the _SPHARM.coef or the SPHARM_ellalign.coef?

For the regTemplate, I am not sure what mesh file to use for Procrustes alignment. I attempted using the triangular surface mesh computed from GenParaMesh (_surf.vtk) but it gave me this error:

vnl_error_matrix_dimension:vnl_tag_mul: Dimensions [3,32062] and [4002,3] do not match

For the Procrustes, do I only apply it to the 2nd case and after? Also, just to confirm, do I need to convert the .vtk to .meta to use as template in the command line?

As for ShapeMANCOVA, would it be possible to compare several images individually of the same item (imaged at sequential times) to each other instead of comparing groups?

Thanks,
Sylvia

Hi Sylvia,


You want to be consistent in your analysis, and so you should select the same surfSPHARM.coef file for all your population. Usually we use the surfSPHARM in the original space (not the ellalign one). I would use the one that is in your Template folder. If you dont have a Template folder, just use the first one.
Here you need to use the surfSPHARM.vtk as well, that is why there is discordance with the points. Your surf.vtk still does not have correspondence, your surfSPHARM.vtk does. I will use the one that is the Template folder, if you dont have it just use the first one you compute.

With regard to whether or not you apply to the 2nd case and after, i would always compute the first individually and then making a script to compute the rest (including the first with the first). That makes it much easier.
If you are using ParaToSPHARMMesh and not ParaToSPHARMMeshCLP you have to convert to meta. However, I would recommend to use the CLP versions of all that commands because are the ones we mantain and debug. The rest are obsolete.
Of course! The fact you set correspondence in the whole population doesnt mean you have to analyze them altogether. You decide how you write your csv file.


I hope that helps.
Regards,

BEATRIZ

Hi Beatriz!

Thank you for all your help. Using the right SPHARM.coef and .vtk files for templates (from the first object), I was able to compute the original, ellipsoid-aligned, and procrustes-aligned mesh.

Just for checking, I used KWMeshVisu to take a look at the first case of original, ellipsoid-aligned, and procrusted and compared it to the second case that utilized the templates from the first. Without doing any mouse rotation, is the procrustes-aligned object of the first supposed to be oriented the same as the second? How do I know if my objects will be aligned properly for statistical analysis?

If I am comparing objects between Group A and Group B, should I use Group A's flipTemplate and regTemplate when computing the SPHARM of Group B?

Cheers,
Sylvia

Hi Sylvia,


Congratulations for being able to compute your data! I am happy I was able to help somehow.
Checking alignment with KWMeshVisu is very difficult because it only allows one mesh loaded at each time. It would be possible to do some sort of alignment checking by using the "Get Camera" functionality in the File menu, but that would be non-trivial and very time consuming. My suggestion here is that you use Slicer3 for that (www.slicer.org). Is a free software program for image analysis that you can just use for visualization purposes.

You can load as many meshes as you want using the "Models" module, and in that way you can see how well they overlap.

When comparing objects in your population, it depends if you want to make analysis in the original space because you have some sort of pre-aligned input data (just use *surfSPHARM.vtk files), or if you want to make analysis in the aligned space created by SPHARM (either *surfSPHARM_ellaligned.vtk or *surfSPHARM_procalign.vtk, I usually choose procalign space since ellaligned is more a pose correction than a registration per se).

Regards,

BEATRIZ