Gannet v. Osprey Results

Hi All,

First of all, thank you for such a great resource here! It has been very helpful.

My question is about differences in GABA concentration quantification between Gannet and Osprey software. I am analyzing MEGA-PRESS data for the first time (Siemens PRISMA 3T, TE 68) for GABA, and ran all the data in both Gannet and Osprey using the standard pipelines (Separated and with macromolecules/lipid basis functions added to fit in Osprey). I am comparing the GABA levels in the dACC between a medication and placebo (scans separated by about 20 days) using the “GABA_ConcIU_TissCorr” output from Gannet and the " diff1_TissCorrWaterScaled_Voxel_1_Basis_1" GABA levels from the Osprey output.

When I run the stats using the exact same code, I get a significant effect of medication using the Osprey (p=0.005) data and a non-significant effect when using the Gannet data (p=0.300).

My general question is: any thoughts as to why am I getting different overall results between Osprey and Gannet? And is there a way to feel confident in picking one software over the other when writing up the findings? I have read the papers for both Gannet and Osprey, but since I am new to this work I may be missing something important that causes this type of discrepancy. Forgive me if this is a naive question, but having different outcomes with different software is making me nervous! Thanks!

Hi @AnnaKirk,

(Full disclosure: I am Gannet’s lead developer so obviously my opinion will be biased.)

I highly suggest reading our paper where we looked at this very issue but across seven different MRS analysis tools. The short answer is that estimated GABA+ levels across these packages only moderately agree. The ICC of GABA+ levels measured using Gannet and Osprey were poor (depending on how you define what a “good” ICC is…) (see Fig. 5).

There are a number of reasons explaining these findings, but it boils down to each package having its own processing pipeline, use of prior knowledge, and modeling approach, sometimes these differ within packages as users may be able to modify the analysis parameters.

Personally, I don’t believe there is a way to pick the “right” software tool. All have their own advantages and disadvantages. I think the most important and fairest thing to do is to report as fully as possible how you acquired the data, how you analyzed the data (stating what the pipeline was used and if it was modified from the default), and what the quality of your data was (see the MRSinMRS paper for consensus recommendations on this).

We in the MRS field are well-aware that the lack of standardized analysis approaches may very well lead to conflicting results, so I would not be discouraged by what you are seeing with your own data.

Finally, should you have any technical concerns with your Gannet results, please do reach out for support.

I hope this helps.



Thank you SO much! This is extremely helpful.