People will argue all night long over what belongs in a basis set and what doesn’t… but I’m pretty sure there’ll be near-unanimous consensus on GABA.
To ask the other way around and make sure you’re not missing any big ones - which metabolites did you include in your MRSCloud basis set?
Thank you both so much for your help with this issue.
@hraum I added GABA to the basis set as you recommended, however this didn´t solve the problem, it appears that for the code starting in line 1036, we need to have the following metabolites listed:
metabNames = {‘GABA’,‘Glu’,‘Gln’,‘Glx’};
I added now Gln, which I didn´t have before, and now it runs without a problem.
So, if I understand correctly, there is a correction using GABA and Glx when calculating the final concentrations, independently of the sequence used: MEGA-PRESS or PRESS?
@admin and @hraum to clarify why I didn´t include in the first place GABA, is because I am using a PRESS sequence which I believe is not the most accurate way to measure GABA, therefore I omitted it. For my research question I am interested only in looking at Cr and Choline, and after some feedback rounds (with Georg as well, maybe you remember this is the anesthesia data) we thought about running my analysis with a more restricted basis set including only the metabolites I am interested in. Therefore in my basis set I chose originally: Glu, PCh, Cr, PCr, PCP and NAA. Currently I added GABA and Gln.
Thank you again,
Best,
Juliana
You should always include as many metabolites as are needed to achieve a fit that represents the whole spectrum well.
There is a line to walk between underfitting a spectrum (not including enough model parameters and/or basis functions, causing a substantial fit residual) and overfitting a spectrum (using too many model parameters and/or basis functions, causing unnecessary model complexity and high variance of the parameter estimates).
But specifically, you cannot just include the metabolites that you are interested in - omitting metabolites from the model will just result in an inappropriate model, and the estimation of its parameters will be invalid. As an example, just because short-TE PRESS isn’t the best way of estimating GABA (I agree), the GABA signal is still present in the spectrum, and omitting GABA from the analysis will just cause its signal to be incorrectly absorbed by other model components (say, Glu, Gln, or the baseline) or to end up in the fit residual.
There is, unfortunately, quite some disagreement among MRS experts which of the lower-concentration metabolites ought to be included in a fit. Most will say “it depends”, and they’re not wrong, but I’ve always felt like that’s an easy cop-out. We’ve recently published a paper that tries to use information criteria to determine the optimal basis set composition. The general approach works, but there’s still ambiguity in the detail.
Sticking with defaults (as the default Osprey basis set composition, for example) usually strikes a middle ground - it might not be perfectly optimal for each and every application, but it’s probably “reasonably good” for most applications.
Hi @admin
Thank you very much for your feedback, I really appreciate it!
I will stick to the defaults from Osprey as you recommend it.
Best,
Juliana