Adding physiological regressors to fmri confounds file

Hello experts,

This is a general question about denoising with a specific follow-up. We are currently working with task-fMRI but I am also curious about how this issue applies to resting state.

I have seen users on this forum recommend against using physiological (heart rate/breathing) signals to generate regressors (e.g. with RETROICOR or physIO), since the 2007 Behzadi et al. paper found that these are largely accounted for by aCompCor. However, a recent 2023 publication by Krentz et al. suggests that RETROICOR can still account for significant signal variability, especially in certain individuals. I am wondering if it is still recommended to stick with aCompCor for the moment, or whether I should try and integrate our physiological recordings and somehow add them to the confound regressors file generated by fmriprep to use them in a later GLM.

IF adding these physiological regressors is now recommended, has anyone done this and could they let me know of any pitfalls to look out for? We have Siemens .phys and .puls files.