MEGA-PREE analyzed by Osprey but got 0 in GABA

Hi everyone. Nice to have the chance being here and wish to have your support about the problems that I’m facing now.

I got my MEGA-PRESS data from 3T siemens with 352 averages (176 on + 176 off), TR/TE=1500/68ms. The voxel was 2.2 x 2.2 x 2.2 cm^3 and placed on the visual cortex. Before the formal scanning, I also acquired another 32 averages using the similar TR/TE but with no water suppresion as the reference.

The Osprey works well in doing the analysis. However, when I checked the quantified table, it showed that the GABA value is zero in all kinds of scaling.
Does anyone know where the problem is? Is it from the data quality or anything else? I’m happy to share the orignal scanning PDF and the MRS data. Many thanks!

I can only post one picture each time because I’m a new user :smile:.
As for the data quality, the SNR for GABA is 9.7, and the FWHM(tCR) is 7.74, the reference shift is -6.42 Hz.

The fitting doesn’t look good and no clear bumps of GABA can be seen. I wonder if there’s a way to exlude bad averages from Osprey, but it seems that the preprocessing is done automaticly and I’m not sure whether average exclusion are involved.

Thanks for your time and hope this question can be resolved soon.

Hi Melinna,

This is unfortunately not a good GABA-edited difference spectrum, it is riddled with subtraction artefacts and presumably substantial lipid contamination (which is not uncommon when scanning the visual cortex). This would certainly need to be discarded for any further analysis. Do all of your datasets look like this?


Hi Georg,

Thanks for your quick reply! I really appreciate it since a few people around are doing the MRS study, and I’m glad to have professional advice here.

I understand that the lipid contamination can be a big problem in GABA-edit sequence, so we delibrated placed several saturation bands along the skull to avoid lipid contamination and increased the averages. Do you have any ideas on how to lower the chance of being contaminated by lipids in MEGA-PRESS?

Currently, I only have one pilot subject and still have the chance to refine the scanning procedure. This pilot subject also underwent another PRESS squence scanning on him (TR/TE:3000/30, 99 averages, plus 16 water reference). The output of his PRESS is shown in below.

Do you think the PRESS scanning is less affected than the MEGA-PRESS? If we target two metabolites (GABA and Glx), is PRESS more reliable than MEGA-PRESS in terms of the current data quality?

Thank you again for your valuable help.


This doesn’t look catastrophic in terms of lipid contamination, but also not great (side note: more transients don’t help with lipid contamination). Can you show the output of the voxel placement? I generally recommend staying an extra millimeter or two away from the edge of the brain, and make sure that the outermost voxel face is roughly parallel to the curvature of the skull that it faces. You don’t want the voxel corners to point towards the skull, if you see what I mean. Sat bands need to be set up properly (i.e. the frequency set to -3.4 ppm in order to saturate lipids at 1.3 ppm) and should be at least 7 mm away from the voxel itself to avoid deleting your actual signal.

Couple more thoughts about the protocol:

  • 99 transients is a very odd choice. You usually want to go in multiples of 16 to complete a phase cycle, i.e., 96 or 128.
  • What’s your voxel size? For MEGA-PRESS, I would shoot for at least 20 cc3, everything below that will be noisy. Make sure that you acquire more transients for MEGA-PRESS (start with 320, work your way down if the data look good)
  • You don’t need 16 water transients, 1-4 should be enough, but make sure that the water delta frequency is set to 0 (for best colocalization) and acquire it with the water suppression set to “RF off”.
  • No, PRESS is not inherently less affected by lipids than MEGA-PRESS, it’s the same localization.
  • No, PRESS is certainly not reliable for GABA at all at 3T.


1 Like

Hi Georg,

Thanks for your kind help! I’ve included our voxel placement data below, and I would greatly value your feedback.

Both our team and the MRI center technician are new to this scanning process, so your insights are incredibly valuable to us.

Here are some key points:

  1. Voxel Size: Currently, we have chosen a voxel size of 2.2 x 2.2 x 2.2 cm^3. We made this decision based on a previous study that reported issues with a larger voxel size, which sometimes covered unwanted areas such as the cerebellum. However, after this pilot study, we are open to considering adjustments to the voxel size.

  2. Transients and Water: We plan to modify the 99 transients used in PRESS to a more appropriate number (96 as you suggested) and reduce the transients of water involved in the scan.

  3. Scanning Parameters: Please help review our scanning parameters provided in the attached document. I have some concerns about the water delta setting, which appears to be -2.80 ppm according to the second page. If I have misunderstood this parameter, kindly correct me.

scanning parameters:
SO_AC_049[Allen Cheong]64_4.pdf (56.3 KB)

Once again, thank you for your assistance.


  1. I would consider extending the voxel further along the calcarine sulcus. You currently have a 10-ml voxel, just slightly more than a third of the volume we recommend for making your first steps with MEGA-PRESS. Remember SNR scales linearly with voxel volume, but only with the square root of measurement duration, so you pay a hefty, hefty price for making the voxel small. I always say it’s easier to write a discussion about partial volume effects than to read tea leaves from noisy data.

  2. About the protocol:

  • Choose the lowest-possible value for edit bandwidth.
  • You don’t need 16 preparation scans; 2-4 should be enough.
  • You definitely need to acquire the water reference with water suppression set to “RF off” instead of “None”. Incorrect eddy-current correction may have contributed to the messed-up MEGA-PRESS spectrum that you posted.
  • We usually set the delta frequency to -1.7 ppm for GABA-edited MEGA-PRESS (which means that the 3-ppm signal gets localized as displayed). All water reference scans must have delta frequency set to 0.0 ppm.
  • The protocol printout didn’t contain the short-TE PRESS scans, so I can’t review them.

Hi Georg,

I’m so impressed by the high-quality comments you’ve provided! We contacted the MRI center and made the changes according to your suggestion immedialtely!

Majr changes are made:

Water reference:
1 set the water suppression to “RF off” instead of “None” and set the water suppression B.W to 35 Hz (we should make it consistent for both MEGA-PRESS and PRSS).
2 set the delta frequency for both MEGA-PRESS and PRESS water reference to 0 ppm.
3 decrease the numbers of water avearges and the preparation scans.

1 set the edit bandwidth to 47 Hz instead of 70 (this is the lowest avaliable option now)
2 set the delat frequency to -1.7 ppm.
3 increase the voxel size to cm^3 (almost 20 ml).

Additionally, I’ve uploaded the short TE PRESS sequence now and am really looking forward to receiving your feedback. The default delta frequency in PRESS is -2.3 ppm. I’m not sure should we change to.

SO_AC_049[Allen Cheong]64 _new.pdf (60.8 KB)

By the way, would it be possible for you to share a Siemens 3T scanning PDF for the GABA-edited MEGA-PRESS sequence? Although I’ve carefully marked all the messages you’ve mentioned, I’m concerned that I might overlook some other key points.

Thank you so much for your invaluable assistance, your time and your selflessness truly shines through. We are definitely planning to implement the revised protocol and will keep you updated on its performance.



  1. I would recommend using the same TR for both classes of scans (short-TE and MEGA-PRESS). 1.5 s is very short (heavy T1 weighting)… does your protocol time allow for all scans to have 3 s TR? 2 s is a compromise.
  2. Set water suppression bandwidth to be the same for both classes of scans.
  3. Set vector size and bandwidth to be the same for both classes of scans. 2048 pts / 2000 Hz seems as good a combination as any.
  4. PRESS delta frequency is commonly set to -2.3 ppm to localize glutamate as displayed. Honestly, I think this is perfectly fine.
  5. Use the same shim settings for both classes of scans. What options do you have available?
    There are some threads discussing Siemens shimming options ( How to measure metabolites in occipito-temporal cortex? - Data Acquisition - MRSHub Forum).

I don’t have a PDF printout ready at hand (we’re a mostly-Philips site) but I’d feel pretty comfortable with your protocol now.

Hi Georg,

Thank you for getting back to me so soon!

Initially, we decided to use a TR of 1.5 seconds because a similar study at the same MRI center had encountered issues with lipid contamination. A Siemens technician had suggested they try a TR of 1500 ms. However, after gaining some experience and reviewing relevant literature, I believe it would be better to revert to a TR of 2 seconds.

Regarding the shim tricks to improve SNR, I’m currently working on gaining a deeper understanding of their effects. I’ve checked some documents, and it appears that for B0 shim, the options include standard, advanced, and brain (while for B1 shim, TrueForm seems to be the primary choice, coilshim is always off too). Do you have any specific recommendations for shimming settings for MEGA-PRESS and PRESS sequences?

Your assistance is greatly appreciated. Wishing you a wonderful weekend.


Hi Georg,

I’m reaching out once more. Your previous feedback is highly appreciated, and today we gathered data using the revised MEGA-PRESS/PRESS sequences. The TR was set to 2 seconds, and all parameters like vector size, bandwidth, water suppression bandwidth, and delta frequency were adjusted accordingly. The raw data looks much better now, especially the high frequency noise is gone, compared to last pilot data. However, upon analysis with Osprey, the GABA value is still zero.

I’m feeling a bit confused at the moment, is the lipid contamination still that strong? Would you be willing to take a look and provide some insights? I’ll attach the voxel placement for reference.

Thank you for your assistance.

It seems that the PRESS sequence was not that bad. Please correct me if I’m wrong in interpretation.

Thank you so much!

I can’t see the difference spectrum plot here, but you generally want to look at GABAplus, not GABA. Osprey uses two separate basis functions for GABA and the co-edited macromolecules, but they are not reliably separated, so you can only ever report GABAplus, which is the combination of the two.

Also, which setting in the job file are you using for opts.SpecReg? I still see a bit of a subtraction artefact in the difference spectrum - you can try the other available options for this setting to see if it goes away.

Thank you for your response! After your explanation, I realized that since GABA+ is precisely what I want, I shouldn’t obsess too much over the GABA value itself. Besides, the setting of could greatly affect the pure GABA value after fitting.

I tried RobSpecReg and ProbSpecReg for opts.SpecReg, and it turns out the latter is indeed better in eliminating the substraction artefact. Thanks for the suggestion.

The below two images are processed by ' = ‘1to1GABAsoft’ (by this setting, GABA value is not zero anymore) and 'opts.SpecReg = ‘ProbSpecReg’. I think the GABA+ value falls into a reasonable range and could be happy with this scanning now.

During this long conversation, I’m thankful for learning about so many critical settings in high quality MEGA-PRESS data acquisition. I hope others can also benefit from this post too :grin: :grin: :grin:.

Thanks again.

Hi @admin,
what minimal edit bandwith do you reach with your Philips MR Scanner at TE 68ms? We have partially the problem to reach low values when the voxel is tilted (only around 90 Hz possible) with a MR7700 system, customer sequence and transmit body coil.


14 or 15 ms duration (with a sinc-Gaussian editing pulse), which is around 80 Hz FWHM.