Unexpectedly Low tSNR After Using nilearn.signal.clean with Detrending

Has anyone experienced issues when using nilearn.signal.clean to deconfound resting-state fMRI time series?

By default, the function applies detrending. When I include detrending, the resulting residual time series values are much smaller in magnitude—so much so that they appear nearly mean-centered (though I’m not explicitly doing that and have set standardize=False). I understand that detrending is often beneficial, but in this case, it’s drastically reducing the amplitude of my residuals.

This becomes a problem when I calculate temporal SNR. The tSNR values drop into the single digits, which seems implausible and is a direct consequence of the low residual amplitudes.

I have two related questions:

  1. Has anyone else observed such a dramatic change in signal amplitude or tSNR when using signal.clean with detrending?
  2. If so, how do you interpret or adjust your tSNR metrics afterward—do you scale them in some way to make them meaningful?

Currently banging my head against a wall, so any insights would be appreciated!