I have some data that needs analysis which is from a block design fMRI with blocks of 90s, with inter block "rest" trial of 20s, and intro screens of around 10s (making around 120s).
The experiment was run in two sessions, with 6 blocks in each session, making sessions of around 720s . There were four block conditions, so that in one session, there could be only one example of a particular condition.
All this means that when modelling the data, the majority of the data exists under the SPM default 127s (0.0079 Hz) filter. Therefore, filtering the data in this way throws out almost all the experimental related data, and means you get no significant BOLD response! Reducing this filter does not help too much (although it does to a degree), as it increases the amount of experiment-related signal, but also the amount of noise.
I have been looking into alternatives to help analyse the data. One of which would be to remove the drift by some other means such as detrending (auto-detrending or spline detrending) or ICA/PCA.
Are these sensible options, or are there any other options? Also, how would this be implemented in SPM?
I know the experimental design is the cause of the problem, and a simple way to get round it would be to redo the experiment with smaller blocks. However, this is a patient study, and re-recruting may be difficult.
Thank you for the help.