Hi Kausar,

Yes, a task-based timeseries convolved with HRF would be an appropriate reference timeseries. This option is present in case the signals correlate particularly strongly with your task, in order to prevent variance being "stolen" by the correction. However, I don't fully understand its effects on the analysis, only that it will have an effect on the analysis.

I would only use this option if you run PESTICA and either the cardiac or respiratory noise coupling map looks like activation. The problem is identifying that kind of coupling. Any time you run a regression, some of the signal of interest would correlate with a signal of no interest, even if the no interest signal were random noise. If the covariance is stronger, then the amount of signal of interest thats removed will increase, and the noise coupling maps will look more and more like the t-map you'd get for your signal of interest. When do you decide one of the physio regressors is covarying too strongly with your task? I do not have a good answer beyond the almost useless answer "you know it when you see it". I recommend not using the reference protection option. I looked into objective methods, but they fell off my radar. This is an issue I am concerned with, because I see strong coupling (both to fMRI and connectivity) occasionally, so I will be revisiting this issue. A better method is to use first-pass when the coupling is concerningly like activation or connectivity.

Erik