Temporal resolution in event timing 1D file

Hello,

May I ask for your recommendation on the appropriate time resolution to use when preparing a task timing file?

In our case, event onsets and durations are defined based on concurrent EEG recordings, and their time resolution is rounded to 0.1 seconds. Should I use this resolution directly when preparing the timing 1D file, or is it better to align them to the TR grid? My understanding is that, in AFNI the basis functions are treated as continuous. Therefore, the task regressor is sampled at the TR resolution after convolving the event and basis function at a much higher resolution.

If this is correct and TR=2 seconds, is it still better to use BOLCK(0.2, 1) instead of BOLCK(1, 1), even if the difference of the final generated regressors may be negligible?

If I would like to use duration modulation dmUBLOCK(x) and x is calculated by the median of all durations, I guess it is not necessary to round x?

Thanks,
Zhengchen

Hi Zhengchen,

That is right, the timing does not need to be rounded, so using the 0.1 s resolution sounds good.

Yes, there will not be much difference between BLOCK(0.2,1) and BLOCK(1,1), though the peak of the latter will be close to that 0.8 s difference.

What is a typical range of durations?

That is right, if the median duration is about 0.2s, using dmUBLOCK(-0.2) would be reasonable, but it could also be -0.2000173. Personally, I tend to round the time. It look nicer, and it does not have to change every time you add a subject. Either way seems okay.

Note that those times (e.g., 0.2 in dmUBLOCK(-0.2)) should be constant across subjects.

  • rick

Hi Rick,

Thank you very much.

What is a typical range of durations?
The range of durations varies from subject to subject since the events are observational, I don’t have a summary of it right now. It goes from 0.2s to ~5s, but we group events based on some domain knowledge, and usually events in one type have similar durations.

Personally, I tend to round the time.
I agree, I just round it to have the same number of decimals in TR, so 0.01s.

Note that those times (e.g., 0.2 in dmUBLOCK(-0.2)) should be constant across subjects.
This is a good point, I guess otherwise the effect in % change may not be comparable across subjects since the expected responses are not scaled in the same way.

Thanks,
Zhengchen