regress_basis option

I am a relatively new user of and have a question about what regress_basis option that I should use for my analysis.

The fMRI data that I am analyzing is an slow-event-related design haptical task, which includes 8 types of events (e.g., openleftsmall graspleftsmall). For each type of event (e.g., openleftsmall ), the duration is the same across the whole task (e.g.,14s) and across all the participants (the stimulus onset times are the same across participants). Given this design, is it correct to use the “-regress_basis 'BLOCK(duration,1)” option? Any guidance would be greatly appreciated!!

The part of my script for the first-level GLM analysis (including the -regress_basis part) is detailed below:
-regress_basis ‘BLOCK(14,1)’


Here's actually a paper using for a haptic task design:

This might form a useful resource for setting up processing (and of course feel free to adapt any relevant features for your study). The AFNI Codex (Code Examples) page for this study, with scripts for and other code snippets used in the study, is here.
For example, that was a surface-based analysis, and you might want to do a volumetric one.

Looking at that, what you are suggesting seems reasonable.

I would probably also add this opt, for an additional censoring criterion, based on outlier fraction in mask:

-regress_censor_outliers 0.05  

For QC purposes, I might change update the following option to have 3 arguments, as here:

    -radial_correlate_blocks  tcat volreg regress    


Dear pt
Really thans for your answer! I will read this paper carefully. But actually, as a novice, I don't know the difference between surface-based analysis and volume-based anakysis, I just want to
make my preprocessing result better,cause now when I used GAM as regress basis, there are collinearity problems. Maybe I should use TENT as my regress basis or BLOCK(even though my experiment is a slow-event related design: 1s(cue)+3s(haptical event)+10s(the interval between events))?

Qianqian Wu


For getting a better sense of processing choices and surface vs volumetric analysis, here are a couple useful resources about and things to think about (actually, these kinds of considerations really apply when using any tool for setting up FMRI analyses):

To the question at hand: the initial question is mostly about regress block options within, which are fairly separate from volume- vs surface-based analysis considerations. I will let @Gang comment on the collinearity aspects. It might help to post the QC HTML images of the idealized stimulus regressors, as well as the full warnings.