Hi, I am using the new version of Osprey (2.10) because I had different results analysing the data with different computers and I saw in a post here (Inconsistent Fit results from Osprey) that the latest version addressed this problem. However I am confused about the output of the new version which seems to include a lot of files none of which is a .csv. Is that normal?
I was using the following files: diff1_tCr_Voxel_1.csv, diff1_AlphaCorrWaterScaled_Voxel_1.csv
Should i assume that the equivalent in the new version are diff1_tCr_Voxel_1_Basis_1.tsv and diff1_AlphaCorrWaterScaled_Voxel_1_Basis_1.tsv?
The .tsv files are functionally equivalent to the .csv files - the only real difference is whether a tab or a comma is used to delimit the data, which has minor implications for reading the data into other software (e.g., R or Python) but no implications for what can be stored in the file.
The Osprey output should be totally unchanged. Hope this helps!
Hi, I wanted to ask about some changes I’ve noticed in the Osprey output since using Osprey 2.10.
The quality metrics used to show NAA_FWHM and NAA_SNR, etc. and have now been replaced with Cr_FWHM and Cr_SNR. I have not changed my code at all.
More importantly, and more concerning, is that GABA quantification has changed drastically. I am now obtaining values of 0 for the majority of my data, where in previous Osprey versions using the same exact pipeline there were more variable levels of GABA rather than a flat 0 for majority of data.
Has something changed behind the scenes for GABA, or overall metabolite, quantification and quality metrics in the Osprey version updates?
With regards to the SNR and FHWM estimates. This was changed in 2022 with the release v.2.2.0 (Release Osprey v2.2.0 · schorschinho/osprey · GitHub see #424). The reason for this was the integration into a larger infant study where NAA changes drastically compared to adult brains. Therefore, we decided to default to tCr instead of tNAA. You can change it back in the code if you want to.
With regards to the modeling. The LCM algorithm in v.2.10 is drastically different compared to the previous implementation. Therefore, I would also expect differences in the quantification. Which opts.fit.coMM3 model are you using? What are the estimates for GABAplus in the output? If the majority of the data shows a GABA = 0. I would recommend using the 1to1GABA model instead.
To respond to your questions, the estimates for GABAplus are more in line with our previous results and we are currently using the following settings:
opts.fit.coMM3 = ‘1to1GABAsoft’;
opts.fit.FWHMcoMM3 = 14;
We also have a few questions:
i) where in the code can we change the snr and fwhm back to tNAA?
ii) can Osprey provide snr and fwhm for GABA specifically?
I’ve included data from three different pre-processing runs. One done in Osprey 1.1.0 using 1to1GABAsoft, and another done in Osprey 2.10 also using 1to1GABAsoft. As you can see, the values for GABA are hugely different. Following your response to Valentina, I ran another in Osprey 2.10 using 1to1GABA, and the results are closer to the original from Osprey 1.1.0 using 1to1GABAsoft.
In addition to Valentina’s above questions, could you please clarify why these differences are happening? It is concerning that the data changes so much and with little understanding as to why that is, and is something we will need to be able to explain and justify in future papers if we continue to use Osprey.
Please let me know if you have any questions or need any further clarifications.