How to processing task data with very long block (120s)

Dear all,
Recently, I have got a bunch of fMRI data collected from an experiment using a block design. However, it has a very long task block (lasting for 120 seconds) and containing five different task conditions with only one block for each condition ! I have got some information from Friston’s book (statistical parametric mapping), which suggested to avoid block longer than 50 seconds and not to compare two conditions that separated far apart over time.

So, here is the question, is this dataset still useful? If so, which method is suitable to analyze it ?

Attachment is the diagram of that experimental design.

Any help would be appreciated.

Best Wishes!

The issue here is that MRI signal has a lot of low frequency drift, due to effects like gradient heating. Unfortunately, your expected frequency is 120s “up”, so a 240s sine-wave. Given the roll-off associated with filters, the standard rule fo thumb would be a 480s high pass filter, which will allow a lot of low frequency noise to leak through. I think you would want to choose your filter carefully. I would anticipate that you will have a null result, the amount of low frequency noise makes it unlikely your signal to noise will survive a statistical threshold. On the other hand, if your effect size is very large, you might find something.

What about something like aCompCor? The first component of a CSF/WM-based PCA would probably reflect drift, without the BOLD response from the task.