Is it correct to take means of maps of t-values between repetitions?

We have data from localizer experiments. Each localizer experiment was repeated 3 times, which was recommended for better results. The FSL first level analysis code we have run expected a single experimental run, so now we are left with 3 separate brain-shaped maps of t-values and are uncertain how to correctly combine them.

They were realigned during preprocessing. Could we simply average across runs to combine them (before applying thresholds)? Or wouldn’t this be correct?

What you need to do is to compute fixed effects estimate. A simple way to do it is to add separetely the parameters and variance estimates of the three blocks, and then derive a t-statistic from these summaries.
You can use for that Nilearn’s ‘compute_fixed_effects’ function

1 Like