When constructing a structural connectivity matrix using MRtrix3, if the T1-weighted structural image is not normalized to the MNI152 space, but instead the parcellation template in the standard space is inversely registered to the individual native space for matrix construction, can the resulting matrix be directly used for statistical analysis?
Hi,
If I understand correctly, what you are doing is registering the subject to the MNI and then apply the inverse transformation to move the labels to the T1w, am I right? If that’s the case that’s one of the common ways to compute the connectome, so those matrices can be safely used.
Best regards,
Manuel
Thank you for your reply.
I do not want to register the subject’s images to the MNI standard space. Instead, I intend to perform white matter tractography using the original T1-weighted structural images throughout the entire MRtrix3 processing pipeline. Finally, I will register the parcellation template in the standard space to the individual subject space for constructing the connectome matrix. Is this approach feasible?
And can the generated connectivity matrix be directly input into GRETNA for statistical analysis?Thanks.
Firstly it’s vital here to disambiguate between “register” and “transform”. You should absolutely not “register” a parcellation to a subject’s image data (or vice versa), as the contents of such parcellation images do not contain intensities that are suitable for driving the registration algorithm according to the minimisation of some image similarity cost function. Registration should involve the subject’s image data and the template image data, with the outcomes of that registration process being used to transform the parcellation information from the template to the subject.
Transforming template-based parcellations to subject space is probably more common than the converse for structural connectome construction. For fMRI the opposite may well be true.
Whether the resulting connectivity matrices can be immediately fed into some software for performing statistical comparisons depends on what you are computing from those matrices. Different global graph-theoretical measures, or indeed per-edge connectivity measures, are sensitive vs. insensitive to global scaling of the magnitude of the connectivity matrix, which has consequences for how you might want to deal with differences in that global scaling across individuals.
https://doi.org/10.1002/jmri.28631
Regards
Rob
1 Like