Hi Mark, 

Thank you for your advices. 

Unfortunately, we are still having some issues with these analyses. 

Our scans are 10min long and include 5min of infusion and continuous ratings of ache and fatigue followed by 5min of passive movement and continuous ratings. We have decided for now to focus only on the activation associated with the passive movement cycles, which are 20sec long (10sec on, 10sec off). In the previous analyses I ran without HPF, with CSF-WM regressors as nuisance regressors and convolving with Flobs, we had some amazing activations, but a lot of noise for the contrast using the passive movement cycles.
We have tried to work with conservative HPFs as you suggested to remove the noise. So our analyses include the following now: 
1) Feat with only MCFLIRT and registration turned on
2) Extract CSF and WM regressors
3) Feat with the following parameters: HPF on, no slice timing correction, FieldMap correction, spatial smoothing at 5mm, registration, CSF and WM nuisance regressors, 4 regressors of no interest convolved with Flobs (ache during infusion, ache during passivement movement, fatigue during infusion, fatigue during passive movement), and one regressor of interest (on-off of the passive movement) convolved with Flobs.
We tried HPFs from 50s to 200s. But it removes everything (noise and signal) in most subjects. 
We thought that it might be due to the first 5 minutes during which we do not have the passive movement cycles, which are coded as 0. So we added a regressor of no interest, which codes the first part as 1 and the part with the passive movement as 0. It did not change the results. 

Another solution could be to cut the scan and run the analyses only on the passive movement part, although it should give us the same results as when we add a regressor coding for the first part, shouldn't it?

Is there anything we are missing and we could try?

I would truly appreciate any advice you could give me. 

Thanks a lot.
Sincerely,
Marie-Eve