Fitting MEGA-PRESS data in Osprey


I tried to analyze my MEGA-PRESS GABA data (3T Siemens Prisma) in Osprey. Few important parameters of the scan data are - 64 averages, TE = 68ms, and TR = 2000 ms (TWIX data format). Osprey ran smoothly from loading to fitting the data. Though, the fit for the ‘off’ data looks good (the residual looks flat) but the fit for the ‘difference’ data doesn’t look good visually. The residual shows a big peak at 3 ppm/GABA position. I tried all the spectral registration methods to improve the fitting, but the fitting remains all most same in all spectral registration methods. What can I do to improve it? It will be really helpful if I can fit the GABA peak well.

I checked the same data in GANNET, the fitting is not bad. I attached screenshots of the fitting in the GANNET and Osprey LC model(RobustSpec Reg)OFF1_GABAGlx_vox1_fit.pdf (323.1 KB)

Also, another naive doubt, why does the GABA peak fit in the Gannet a unimodal lorenztian function, but, in the Osprey LC model or in the LC model, it looks like a bimodal shape function? Though they are at 3ppm, the same GABA proton.

Swagatameas_MID00026_FID22782_eja_svs_mpress_Gaba_1_Voxel_1_OspreyFit_diff1_diff1.pdf (393.7 KB)
meas_MID00026_FID22782_eja_svs_mpress_Gaba_1_Voxel_1_OspreyFit_off_off.pdf (422.9 KB)

1 Like

Hi Swagata,

Thanks for reaching out. Your suspicion about the differences in the Gannet and the Osprey model being the driver of the difference is right. You have to remeber that for the standard MEGA-PRESS (as you have performed in your experiment) you will have a considerable amount of coedited macromolecules at 3 ppm. Gannet uses a single Gaussian peak to model the signals at 3-ppm regardless of its origin (GABA or macromolecules). Osprey, on the other hand, uses a basis function to model the signals. You do have the option to add an additonal basis function to model the 3-ppm macormolecules in Osprey, which you are currently not doing (at least from what I can see in your fit plots). Therefore, you can see substantial residual at 3-ppm which will most likely be coedited macormolecule.

If you have dowloaded the latest release from GitHub you can choose between different options to model the macromoelcules in the differnce spectrum by adding the lines below:

% How do you want to model the co-edited macromolecules at 3 ppm for GABA-edited MRS?                                                     = '3to2MM';         % OPTIONS:    - {'3to2MM'} (default) 
                                                %             - {'3to2MMsoft'}
                                                %             - {'1to1GABA'} 
                                                %             - {'1to1GABAsoft'} 
                                                %             - {'freeGauss'} 
                                                %             - {'fixedGauss'}
                                                %             - {'none'}              = 14;

You can also look into the sdat example jobFile for this. Our paper about the advantages of the different models has just been accepted (Comparison of linear combination modeling strategies for GABA-edited MRS at 3T | bioRxiv).

You could start with the 3to2MM model or try one of the other options (1to1GABA is closest to the Gannet model)

Let me know if that improved your modelling.


Hi Helge,

Thanks for your suggestion! I read the paper, it really helped me to understand the need of MM incorporation in the LC model.

I tried the 3to2MM model, but I am not able to comment on the fitting because I think I am not doing the preprocessing right in OSPREY. My data shows one/two glitch (most likely subtraction artifact) at 3 ppm which affects the fitting as well. Because, in your paper or in the OSPREY paper, I see no such glitches at the MEGA-PRESS edited data.
I also do not see any glitch in the same data preprocessed in GANNET.

What can I do to improve this spectra in the preprocessing? I used all the default options in OSPREY job file for preprocessing.


Data in Osprey :
Fit - meas_MID00026_FID24639_eja_svs_mpress_Gaba_VIS_1_Voxel_1_OspreyFit_diff1_diff1.pdf (396.8 KB)

Preprocess data - meas_MID00026_FID24639_eja_svs_mpress_Gaba_VIS_1_Voxel_1_OspreyProcess_diff1.pdf (2.3 MB)

Same Data in GANNET - OFF1_GABAGlx_vox1_fit.pdf (321.9 KB)

Hi Swagata,

Great to hear that the paper did help you.

From what I can infer from your PDFs, it is indeed an alignment issue. Would you be able to share your data with me? The best case would be one dataset that is working fine and two with alignment issues.

You can send me a link to your data (


Hi Helge,

Yes sure, I can share my data. I tried OSPREY for four data sets, but this issue is present in all of them. So, I did not have a dataset where it did not appear. But I have data where this artifact is smaller than other data.
I have emailed it to you in this email id ( Please let me know if you need anything else. Thank you for the help!