Thanks for the guidelines on creating these FIR regressors. I am working with a block design with conditions imported from SPM 1st level models. The task has jittered ITI fixation crosses followed by a 2sec instruction block, a 6sec emotion regulation block, and a 5 second rating period. I only modeled the emotion regulation blocks with 6sec duration and let the fixation, instruction and rating go to baseline. I followed the instructions for creating the FIR regressors (0:2:16 / 0 window) with sparse (no-HRF convolution) selected and moved the new conditions to the first level covariates. However, after proceeding through setup, the FIR covariates do not appear in the list of all effects to add as confounds.  

A couple of questions:

1) Should I be explicitly modeling the instruction and the rating periods for task-based connectivity confound regression?

2) Cole et al. suggested that the FIR regressors might be more effectively estimated after regressing out the HRF-convolved main effects of task for block designs. How might this be accomplished in conn? 

3) When performing PCA on ROIs such as GM/WM/CSF or seed ROIs, is there a way in the denoising settings to select only a selection of the components (e.g., components 4 & 5) for denoising? 

Thanks in advance, 
Patrick