Data Quality in Osprey

Hi! I have run 2 data sets through Osprey, one is a PRESS (40ms) and the other MEGA-PRESS (68ms) for GABA. I am hoping to get some insights to if my data quality numbers are looking “okay”. I’ve read through some other posts and know that hard cut-offs are not being advised, but I want to make sure my data aren’t totally off. I’ve only found one pre-print paper so far that gives data quality measures from Osprey, and the SNR was ~150-160

NAA SNR for PRESS range: 82 - 668
NAA SNR for MEGA-PRESS range: 203-519

NAA FWHM for PRESS range: 3.3 - 16.7
NAA FWHM for MEGA-PRESS range: 2.9 -7.2

The same participant represents the lowest SNR and highest linewidth for PRESS.

Thanks for any insights!

That sounds pretty good. Can you post a bunch of screenshots? SNR and FWHM are two quality indicators, but there are others, too (lipid contamination, spurious echoes, frequency drift…). Getting a feel for what a good spectrum and a bad spectrum look like is an important part of learning the process, especially since fully automated QC criteria still aren’t really there yet.


Thank you for the quick reply! I’ve run into a few issues since yesterday. Osprey was crashing when I tried to view spectra for the screenshots, but I saw that a new version was released a few days ago so I downloaded it hoping it would fix the issue. I can get to the quantify section, but I am getting this error. Thoughts? Thank you!

Hi @AnnaKirk,

Thanks for reaching out. The problem is that we made substantial changes to the overall data structure since version 1.x and occasionally this causes errors as you see them now.

I think, that a simply changing
exclude = zeros(MRSCont.nDatasets,1);
exclude = zeros(MRSCont.nDatasets(1),1);
should fix the issue.

Let me know if this works.


Thank you - I will try that!

For the newest version GUI, I am unsure what to select for the Spectral Registration and Subspectra Aligment. I see that the “RobSpecReg” is the default for Spectral Registration - will that correspond with default settings from the 2.0.0 version? I also see that L2Norm is the default for the Subspectra Aligment, but in the readme file it notes this is for edited data. For a regular PRESS sequence, is there an option that would correspond to default from 2.0.0 version? Thank you!

Yes - RobustSpectralRegistration has always been the default option for the shot-to-shot alignment. Similarly, the SubspectraAlignment L2Norm was and still is the default option for the subspectra alignment. As you have mentioned, the subspectra alignment is only used for edited data and is not used for conventional/regular PRESS data. Therefore, you can safely ignore this option for this kind of data.

Awesome, thank you @Helge! And the change in code worked :slight_smile:

@admin I’m attaching some screenshots of individual participants at varying visits (V1= baseline, V2= medication or placebo, V4= opposite contingency of medication or placebo). I am also attaching the grand mean, model overview, and general overview. Let me know what else would be helpful!


i saw the title “Data quality” and I watched recently Ivan’s talk at ISMRM where a lot of details on data quality are provided - Quality of MRS/MRSI - YouTube

are here you can find some nice videos - Chapter 3 MRS | _main.knit (

best cristina