Spectral alignment GABA issue

[georg]

Hello!

I’m trying to better understand how the different spectral alignment options affect my GABA peak. For some datasets, the spectra look reasonable when no alignment is applied, but when I use “RestrSpecReg”, the GABA signal develops an unusual shape that persists regardless of the frequency range I specify. This makes me wonder whether I may be applying RestrSpecReg incorrectly, or if there is something specific about these datasets that makes this method problematic. We use the HERMES sequence on 3T, and our files are in TWIX format.

Has anyone encountered a similar distortion of the GABA peak when using RestrSpecReg, or have suggestions for diagnosing what might be causing this?

1. opts.SpecReg = ‘none’;
   opts.SubSpecAlignment.mets = ‘L2norm’;
   GABA peak looks okay
   

   

2. opts.SpecReg = ‘RestrSpecReg’;
   opts.SpecRegRange = [1.5 3.5];
   opts.SubSpecAlignment.mets = ‘L2norm’;
   

   

3. opts.SpecReg = ‘RestrSpecReg’;
   opts.SpecRegRange = [2.5 3.5];
   opts.SubSpecAlignment.mets = ‘L2norm’;
   

   

[mmikkel]

The spectral registration algorithms used in Osprey (and Gannet) occasionally make alignment worse. In this example case, substantially worse. This is likely due to low SNR as I don’t see any major signal artifacts in the data.

You could try the robust spec reg option (RobSpecReg), which should use the whole spectrum/signal. And if there’s enough residual water, will hopefully do a good job. There’s also the probablistic variant (ref) (ProbSpecReg) that I’ve seen work better than the robust approach.

Failing that, since there’s not much frequency drift or motion-related frequency spikes in that example, I personally would be fine applying no alignment. The frequency and phase errors will average out (to an extent) and would likely be much less impactful on your end results than the inherent measurement noise of in vivo HERMES.

Mark