Warping a template anatomy vs using individual mri - children/adolescents

Hello,

I would like to know how the deformation of the mri is performed when warping an anatomy template as described in (https://neuroimage.usc.edu/brainstorm/Tutorials/TutWarping).

How much does it affect the results? How likely do you recommend using an age-matched template (https://www.bic.mni.mcgill.ca/ServicesAtlases/NIHPD-obj1) and warp it instead of using individual MRIs when working on subjects aged between 5 and 18 years old. (I have 111 EEG sensors based on which warping may be done).

Regards,
Sahar

I would like to know how the deformation of the mri is performed when warping an anatomy template

A deformation field is computed between the two head surfaces (from the template head surface to the digitized head points), then applied to the other surfaces and the MRI. This is done by function bst_warp, if you are interested in technical details:
https://github.com/brainstorm-tools/brainstorm3/blob/master/toolbox/anatomy/bst_warp.m

How much does it affect the results? How likely do you recommend using an age-matched template (https://www.bic.mni.mcgill.ca/ServicesAtlases/NIHPD-obj1) and warp it instead of using individual MRIs when working on subjects aged between 5 and 18 years old.

If you have the individual MRIs, why would you not use them?

I don't have any experience about this, but this has been discussed multiple times already, other users might be able to help you:

@yuhanmc @christianoreilly @mclerc @Alexandre @John_Mosher @fleroy @tmedani @clairekabdebon @padibpour @ClaireK Any suggestion?

Thanks for your reply.

Actually, i'm working on a large dataset (444 subjects). Therefore, if the results won't be affected dramatically, I prefer being able to use one single template for all the subjects.

Sahar

Sorry for the late feedback.

I would agree with @Francois ("If you have the individual MRIs, why would you not use them?"). The computational load can be managed by running the reconstruction on a cluster or a similar infrastructure that is accessible to you (e.g., Compute Canada for Canadian researchers, a cluster at your university, etc.). You say "if the results won't be affected dramatically"... how do you assess that? You would first need to know what results you are expecting, how you would measure the "effect" of any distortion (displacement of a source peak? reduction of signal-to-noise ratio? reduction of some correlation between two variables? etc.) , and "how much" distortion you are ready to accept. You will probably not find any definitive answer in the literature on "how much using templates will affect your results as compared to using individual MRI" for your specific question. So you will most likely be ending up having to benchmark it yourself if you want to do things the right way. Since you do have the individual MRI, if you really want to use templates, I would first validate it on a smaller subset, i.e., run reconstruction on e.g. 20 subjects with templates and individual MRI and check how much the metric you are interested in is affected by using templates and then decide of whether it is acceptable or not... You'll have to justify it when you will want to publish your results anyway, so you'll need these numbers to convince the reviewers and the readers that it was an acceptable choice.

By the way, the atlas you are referring to is a "volumetric atlas". You will most likely still have to process it to create BEM head surfaces and cortical surfaces with FreeSurfer or similar software. There may be all kinds of unexpected surprises along the way. I just finished doing a similar job for the 0-2 yo range and it took me a significant amount of work and time. I have a database of EEG recordings from about 100 infants and, personally, if I had the individual MRIs, there is not a doubt in my mind that I would use them instead of templates. Template is meant to be a B plan in case the individual MRI are not available, but why settle for a B plan when you can go for the A plan.

Thank you!