Recently, I increased the strength of the spoiler gradients for the editing pulses of my MEGA-PRESS sequence from 46 µTs/m to 138 µTs/m. To assure data quality, I tested the sequence with both gradient strengths in a phantom and observed a SNR loss of about 10 %. This surprises me, as the spoiler gradients should not influence the signal of my volume of interest and therefore also not the SNR. I checked, if diffusion could be the reason, but as the b-values are very small (0.1 and 1.5 s/mm², respectively) this should be responsible for less than 1% SNR loss. Does anyone have an idea, what could be responsible for this SNR loss?
One way in which spoiler gradients can affect your edited signals is through eddy currents. We found in an old paper of mine (I spent a couple nights doing frequency-offset series on a phantom - 0/10, would not recommend that kinda work) that eddy currents can induce temporary B_{0} shifts that can throw off the editing pulse by enough Hz that your edited signal is affected. We figured out eventually (about a year later) that our (custom-built) MEGA sequence wasn’t applying the internal eddy-current correction model (which predicts the B_{0} offset over time and shifts the RF pulses accordingly) to the editing pulses. Once we fixed that, the problem went away. Did you build that sequence yourself, is this product, or are you using a patch from someone else? Philips/Siemens/GE?
I also want to raise the possibility that this might simply be a one-off artefact of your SNR measurement procedure - how exactly do you measure your SNR? On the NAA methyl singlet or on GABA? Do you fit the data or eyeball the maximum signal? Measuring SNR is not that straightforward - it’s difficult to get right on coupled signals like GABA, and especially on (phantom-narrow) singlets you can obtain different results depending on ever-so-small frequency shifts. To alleviate that, I’d recommend doing a lot of zero-filling before measuring the S in your SNR.
Thank you very much for your answer . I did not think of eddy currents, but it makes sense that they can spoil some of the desired signal. In my data, there is an eddy current artefact, which is unsurprisingly stronger with the larger gradients, estimated by looking at the water reference.
The sequence I am using is built by my group based on a sequence provided by Siemens. The comparison of the sequence to the product sequence did not show differences in spectral quality, so I am fairly confident, that our sequence is doing fine.
Regarding the measurement of SNR, I use the op_getSNR function of FID-A and use the region of -1 to -2 ppm as the noise region. So the SNR is calculated from the absolute spectrum and I calculated it for the NAA and the GABA peak. I did not zero-fill my data, though, so thanks for the tip for future analysis of phantom data. To accomodate for the variability in SNR measurement, I did the measurements for the low and high gradient strengths 20 times each and averaged those results (10 for low gradient strength, then 20 for high and then 10 for low again). A t-test between the SNR sets for NAA of the Edit Off spectra gives a p-value of 0.0004, so I do not think, that the SNR measurement procedure is the reason for the SNR loss.
OK, my next suggestion would have been to repeat the experiment a bunch of times - I would agree that your observation is not an artefact.
Did you do a Klose-style eddy-current correction of your spectra with the water reference?
Just to clarify, the loss of signal will not be due to the odd lineshape (which you can correct for), but due to the fact that your editing pulses will be off-target (if the sequence doesn’t adjust the editing pulse frequencies based on an eddy-current model).
I did the eddy-current correction with the op_ecc function in FID-A using the water reference, where I am not sure if that is a Klose-style eddy-current correction. The result looks good, though.
. I did not think of eddy currents, but it makes sense that they can spoil some of the desired signal. In my data, there is an eddy current artefact, which is unsurprisingly stronger with the larger gradients, estimated by looking at the water reference.