Five different task conditions in one single task block is not that well designed. Problem here is, BOLD responses of every task will contaminate each other and it will be hard to distinguish, which belongs to which.

Nevertheless, if the five tasks are not "new" in a sense, nobody ever tried them and you have references in the common literature in regards to location of activations, you could still try to model the time courses for each task and have a look at the data. Perhaps a shorter duration of each task could help, instead of using a duration of 24s for each task, just use around 12s task duration to avoid the overlap of the old and the new BOLD response.

Good luck!

Martin Domin

Thanks for answering!

Indeed, it is not a well-designed experiment. However, there is only one task condition in a 120s block, among which has 30*4s trials in total. Thus, it seems won't be contamination for each task but facing a problem when filtering the low-frequency noise during 1st-level analysis (there is a 128-second high-pass cutoff to remove slow signal drifts in SPM ).

Anyway, thank you once again!

Ok, now I see! A misunderstanding, I should have looked at the task diagram first.

What kind of tasks are you trying to compare? How many subjects? You could throw away parts of each block (e.g. the last 40s), because if the trials during one task are similar and perhaps monotonous a signal decrease over time will happen in any case.

I would look at the data first using a standard processing and modulate the preprocessing parameters iteratively, in a well-documented manner, of course. If there are enough subjects, effect size could be high enough to compensate for the bad design choices.