Is Aligning an EPI straight to a template in MNI space a Hard No?



So this one is kind of a philosophical question that I’ve been wondering about for awhile that doesn’t have to do with any particular software.

As I’ve done multiple preprocessing workflows I’ve noticed in certain situations that using alignment algorithms (e.g. AntsRegistration, 3dAllineate) using a completely unaligned EPI as the moving image and either the subject anatomy in MNI space or an MNI template as the fixed image gives me better alignment than the common practice of first aligning the EPI to the anatomy then using the ant->MNI transform on the new EPI to get the EPI into standard space. In fact I don’t think I’ve ever seen going straight from EPI->standard space mentioned in the hundreds of tutorials and articles I’ve read.

I understand the conceptual logic for the EPI->anat->MNI tactic in that it’s supposed to give you better alignment because the original EPI and anat are supposed to be close together so aligning the EPI->anat first should give you a fairly good alignment which should then persist if you use the anat->MNI transformation matrix on the new EPI but is there any reason to NOT use a straight EPI->MNI alignment if it gives you better results because isn’t that the point of the process?

I’m actually in this exact situation right now but also it’s something I’ve always been curious about so I wanted to ask people who would know much better than me.


If there is enough contrast in the EPI images this strategy can be very successful. Further discussion and relevant references can be found at:


Given that this article is from 2017 I’m super mad I’ve literally never known about this. Thanks for the great resource. I’m trying this with the SPM EPI template and my data right now but I assume the same logic of this being ‘ok’ goes for using other (read. T1) templates in standard space as references as long as it works.

Thanks Chris!


We’ve also found that bypassing anatomy to align high-quality EPI was a good thing:

It’s particularly true of EPI images that have strong distortion, as these distortions are hard to correct.

EDIT: in the github discussion mentioned by Chris, our paper was already listed (good literature review!). Leaving my reply here, although it’s partly redundant.