I’m dealing with MEGA-PRESS data from a Philips scan with both raw_act and raw_ref data.
I’ve followed the pipeline suggested by Gannet for the preprocessing:
For each of the step I obtained the output, but I’m not sure about the quality of my experiment.
Here attached an example of GannetLoad’s output.
Could anyone give me some rule of thumb to make a sort of quality check of my data, or some general feedbacks about them?
9 ml is a very small voxel size to do any kind of MRS with, let alone spectral editing. We do not see a clearly resolved 3-ppm edited signal, so I would not recommend proceeding with this protocol.
We usually recommend starting at 27 ml and 320 transients, and then slowly working your way down from there.
I also see what looks like a systematic frequency offset (water should be at 4.68 ppm) and non-negligible frequency drift over the course of the experiments, both of which look like they’re throwing off the targeted editing pulse frequency. We’d expect the Glx signal at 2.25 ppm to be larger than it is.
We’re talking about a VOI that includes part of the left DLPFC/ACC. We usually perform SV PRESS MRS in the same VOI, this is the reason of the size.
I also noticed the drift in the water signal too, but I was unsure about the final conclusion.
Here’s the output of GannetFit: it seems that, besides its poor quality observed in the output in GannetLoad output, the error of the fitting is relatively low (1.23% and 1.76% for GABA).
How is it possible?
No, they’re not valid at all. The Gannet fit error is only relative - it compares the standard deviation of the residual to the estimated amplitude.
But the Gannet model is not particularly smart; all it does is try and fit a Gaussian with a baseline underneath to the signal region, without any constraints. If the edited signal does not look like the expected GABA+ peak at all, the model will not be appropriate, and come up with completely unreasonable amplitude estimates.
Your estimated amplitude is gigantic (approximately 100x as big as we would expect) - look at the concentration estimates, you can’t have ~130 mM of GABA! Only because of this absurdly highly estimated amplitude, the ratio of the standard deviation of the residual to the estimated amplitude (ie the fit error) is tiny.
This is what modeling failure looks like, and I would consider neither the data nor the concentration estimates as valid.