is it common practice to discard the first N spectra (eg first 10 or 20 )? I know it is recommended when analyzing fMRI data to discard the first ~ 5-10 sec of data to allow the signals to reach a steady state. I’ve also noticed in my data that the first several averages look worse than the subsequent ones (in isolation, not because of improvement via accumulation/averaging).
A lot of sequences do have
dummy scans that are full sequence cycles but don’t have a readout even (no data recorded). They usually 1 to around 4 of these.
From a simple MR physics point of view, with a spoiled sequence and a 90 degree flip-angle (most single voxel sequence) steady state should be reached on the second transient (see Experiment 1 in the figure below).
However there might be another reason you are suspicious of the first transients, for example you might expect more participant motion, in which case it’s perfectly reasonable to discard these, if not frequently done. Note that you probably want to discard a whole phase cycle of transients.