The best thing I can think of is that you can average the two images to achieve better signal to noise.  They would need registering together first though.  Probably the most accurate approach for rigid-body alignment of such data using SPM12 would involve using the Longitudinal toolbox.  You can make it do within-modality rigid alignment by setting the warping regularisation to [Inf Inf Inf Inf Inf].  For this, it doesn't matter what time intervals you specify, as these only enter into the nonlinear warping part.

One of the outputs should be a weighted average of the two aligned images.  One thing to watch out for is that this average is bigger than the original images, and it will contain some regions where there is missing data.  These regions can potentially interfere with other processing steps.

I can't comment so much on interesting questions to ask of the data.  In regards to quality of combined 1.5T versus 3T, this will depend on the imaging sequences used.  Without further software development related to correcting the additional artifacts in 3T data, I suspect that you may actually get more out of group comparisons using 1.5T data. 

Best regards,

-John

On 26 July 2017 at 18:11, Sven Mueller <[log in to unmask]> wrote:

Dear SPMers,

for some reason (without going into the details) I find myself with the luxury of having 2 back-to-back acquired (in one session) structural scans (MPRAGES) on a 1.5T scanner. Our study sample was comprised of 4 different groups (2 patient, 2 control). I am now wondering how to exploit this unusual situation. We used the ADNI protocol and they suggest that one option could be to combine both images (after spatial registration) and thus increase SNR. Thus, one option would be to do this and then conduct a VBM analysis and simply compare the 4 groups. But, one silly question, are 2 combined 1.5T as good (or better) than one 3T scan?

However, I am wondering if people out there also have some other ideas/suggestions. I thought of using one of the two scans, do the VBM and then use it to see whether we can predict which group the person belongs to in the second anatomical. I have seen papers do this functionally so I have no idea whether this approach would make sense for structural analysis or not and would appreciate any advice on this. Third, are there other options/possibilities of what could be done with such data?

Many thanks in advance for any suggestions,

Sven