The question about scan sequences

Hi all,

There has two fMRI data set that was performed same task. But the scan sequences were different, the first sequence covers the whole brain with the 1.5mm voxel size, and the second sequence just covers the temporal lobe and part of the frontal lobe with the 2mm voxel size. I wonder if I can merge these two data set and do the group analysis.
Thanks for your help。

For this you should
0. In general, I would be very cautious and rather avoid that: it will be hard to convince people that the statistical analysis is rigorous.
If you want nevertheless,

  1. make sure that there is no dramatic distribution shift between the two data, e.g. by comparing histograms.
  2. make sure that the sequence (binary indicator) is not correlated to your analysis variable: for instance, if you want to compare two groups, the group information should be orthogonal wrt the sequence. Otherwise you’ll never be able to know whether you’re detecting sequence of group differences.
  3. If point 1 and 2 are OK, may give a try, but you should still model the sequence by a nuisance variable.


1 Like

Great comments. Thanks!

If I may ask, what would be the appropriate way to model a sequence as a nuisance? For example, what if we had 3 instead of 2 sequences, and we believed that 2 of them were more “similar” than the other?

In the first place, I would not consider modeling the similarity across sequences, and simply model the sequence with a one-hot encoding model.
Note that the effect of this regressor is just to remove any systematic additive sequence effect that would otherwise increase the variance estimate and reduce sensitivity in statistical analysis.

Great comments. Thanks a lot!