Hi,
I have new fmri dataset and I wanted to check whether I replicated my results by applying the same contrast than a previous experiment. This was the case for one task but not for another. Thus I started to check image quality by running MRIQC. I found that the snr and tsnr measures were different between the two experiments. Indeed we change some acquisition parameters, reduction of TR and voxel size. I was wondering whether increasing the sample size may help to solve this SNR issue ?
Best whishes,
Simon
Hi there,
(t)SNR is not something that will be modified by adding more subjects. As you’ve already discovered, the SNR is a (run-level) property of the acquisition. You could adjust your SNR by changing your acquisition parameters.
If you collect more subjects, you will instead be increasing your statistical power to detect between-subject effects. Depending on the particular analyses you’re replicating, here, this may indeed be what you’re looking for ! This paper provides a nice overview of statistical power.
Hope that helps !
Elizabeth
Hi,
Thanks for your help !
Indeed, I noticed that the SNR is a property of the acquisition.
If I reformulate my question, I am wondering whether increasing my sample size is worthy to overcome the low SNR ? But following your recommandation it seems that YES.
Is there a tool that may help to estimate the sample that I need, taking into consideration the SNR change from acquisition to another ?
The last thing is that I ran univariate analyses, I planned to run multivariate analyses on this second dataset. Maybe my question would not be relevant because I don’t know how to clearly formulate it, but I am going to try. Thus, I am wondering whether the low SNR may affect also the multivariate results (low statistical power) ? Or as my data are too noisy, the pattern analysis would have few chance to be consistent from one subject to another.
Best,
Simon