See inline response below.

Best Regards,

Donald McLaren, PhD

On Fri, May 6, 2016 at 4:18 AM, Samantha Brooks <[log in to unmask]> wrote:

Hi SPM experts,

We are currently analysing a block design fMRI working memory task, 1-back versus 0-back, and we have, during the previous quality control step discovered that we had quite a lot of signal drop out, mainly in the second half of a 12 minute scan session (12 block, 6 x 1-back, 6 x 0-back alternating).

Typically signal drop out means that you are having acquisitions issues. However, from the remaining part of your email, "signal drop out" is being used to mean decreases in signal changes between event types. I would be very careful not to mix up signal acquisition issues from decreased changes in the BOLD signal.

We decided to use only half of the session for our analyses, but rather than cut the second half scans completely, where the signal drop out occurred, we have kept in the scans but only modeled in our design matrix the first half for analysis.

Is this the correct way to use only half the session, particularly if one wanted to use the second half in some cases?

Yes. This is problematic because you have not properly modeled the data. All events in the second half of the run are now contributing to the implicit baseline signal. This is very bad and will impact the estimates of the trials from the first half of the run. If the data is bad for some reason - acquisition is bad, motion, etc. then you could drop the second half of the run.

E.g. is habituation an issue here?

Seems like habituation might be an issue and that habituation might explain the lack of signal changes between event types. If you have habituation, then you don't have signal drop-out.

Also, if we were to use a mixture of first and second half sessions, is this wise and can we simply add a regressor to model whether we use first or second half for that subject?

I would figure out why you have habituation and try to model it. If you keep data in some subjects and not others and the reason you discarded data was because a qualitative assessment of habituation, then it will bias the results.

Finally, is it possible that due to statistical power in some instances at 1st level, single subject specification there are no significant voxels (when for e.g. testing 1-back > 0-back)?

Unlikely. I would check the task timing. With a block design, you should see the blocks. Also, its good to add a fixation baseline to the task to verify activity in the visual cortex changing when the task blocks are present versus absent.

Advice much appreciated.

Best,

Samantha and co.
_______________
Dr Samantha Brooks, Ph.D (Download Website)
Dept. of Psychiatry,
J2 Building, Groote Schuur Hospital
Anzio Road
Observatory
Cape Town