fMRS interest/working group incubator thread

Dear all,

At EDITINGSCHOOL2020, we found there is a lot of interest in functional MRS. Approximately 15 attendees expressed interest to participate in an ongoing exchange on best practices for fMRS protocol design, data acquisition, data analysis, and interpretation.

This thread is supposed to be an incubator to bring fMRS-curious people into the same virtual space. I hope it can serve as a vehicle to get to know others who do fMRS, develop best practices and consensus, and join forces to build tools that everyone can use.

I’m including @PGMM’s excellent fMRS primer (PG Mullins, Towards a theory of functional magnetic resonance spectroscopy (fMRS): A meta‐analysis and discussion of using MRS to measure changes in neurotransmitters in real time, Scand J Psychol 2018) as a reference here - feel free to suggest further useful reading so we can keep a running list of literature pinned.

Best wishes,


I think this is an excellent Idea Georg.
I hope to find some extra time in the next few weeks to engage more with this topic directly, and encourage all to do so. Perhaps set up a new Category (or sub category) as well? Of course posts will likely cross several categories (e.g Study and Experiment design, Data acquistion, Analysis/Modeling, software etc).


I agree! Excellent idea! Thanks for putting this together. I look forward to the fMRS discussions and development of best practices geared towards fMRS.



I’ve just added the first (to my knowledge) publicly available fMRS dataset from @arcj to the MRSHub data collection, see the separate thread for details.

Maybe this could be the first fMRS reproducibility challenge - and a way to exchange best practices on data processing.

1 Like

Hi All!
@arcj and I were again discussing dreams for our next adventure in fMRS. A question that keeps coming to my mind is: how would it be possible to perform quantification on a sliding window average spectrum such that each average spectrum contains one full phase cycle? I’m concerned that fitting spectra calculated with a sliding window that contain mostly the same information between each spectrum would just introduce a lot of fitting noise?
We have lately been working a lot with the Philips product implementation of semi-LASER which has a 32 step phase cycle scheme. With TR=5s, one whole phase cycle takes 2 min 40s, which is quite a poor temporal resolution. Being able to slide over one TR at a a time would be a big improvement.
I’m curious to hear what people in this forum would think? How would you approach fMRS experiments with 32 steps in each phase cycle?
Thanks so much!

1 Like

maybe you can find some answers here

i’m not an expert in the field but the talk is very interesting


Hi @erin.macmillan and @arcj,

Thanks for incubating :slight_smile: Just to clarify, and this might be trivial: if you choose the width of your sliding window to be exactly the length of the phase cycle (32 in your example), you can slide it across one TR at a time, and regardless of how far and where you slide it, it will always contain exactly one full cycle. That should work if your paradigm blocks are longer than a phase cycle (assuming that the metabolic response is much faster than the phase cycle and the sliding window, of course).

If you want to make the sliding window narrower, you’ll end up modeling different transients within a phase cycle - I’d think that’s probably fine as long as the crushers take care of most of the nuisance signals? One thing that I’ve observed (at least with the Philips excitation pulses) is a phase-cycle-dependent modulation of the MM/lipid region, which will probably be quite pronounced in cortical regions. I guess there would be ways to improve the modeling of that region by injecting prior knowledge about the phase cycle and chemical shift directions for the excitation gradient… what do you think?


1 Like


Thank you @cudalbu and @admin so much for your quick and helpful replies! Paul Mullins’ talk was very interesting with a lot of practical insights :smiley:

Perhaps I should clarify my questions a bit. I think that phase cycling is so important because we do often see fluctuations shot to shot caused by out of volume signals, and clearly these shot to shot variations are not related to any function that we’re trying to detect. As @admin points out, we also see phase-cycle-step dependent modulation of the MM/lipid region, but sometimes also in the area of glutamate (~2.3ppm) or closer to water (~4ppm). Also, with the increase use of semi-LASER, the phase cycling scheme is now 32 shots instead of 16, and semi-LASER may be more sensitive to out of volume signals since there are 2 additional RF pulses. So I’m curious about a few things:

  1. Can you fit glutamate in the spectrum from a single shot or with the number of shots less than a full phase cycle? Wouldn’t spectra with NSA < 16 have insufficient SNR to fit glutamate? (we only have a 3T for now…).
  2. We could average shots with a sliding window such that each spectrum we put into a fitting algorithm was the average of 32 shots, but each sliding window average spectrum has 31/32 = 97% the same information as the previous spectrum. Would fitting successive spectra that are 97% the same as each other just introduce fitting uncertainty such that the output glutamate levels would likely fluctuate more due to fit uncertainty than with any functional paradigm? … Maybe a pain stimulus would cause a strong enough glutamate change to overcome the fluctuations caused by fitting?

So I’m very curious how people have or would approach the issue of phase cycling to improve the temporal resolution of fMRS.
Thank you for your thoughts!

I’ve mainly been using semi-LASER for fMRS at 7T Philips, but my experience is that when I look at the individual averages that even when I do have large lipid artefacts, smaller phase cycling steps already cancel these out quite well, so you could check and make your moving average step smaller. I also do have the 3T Philips implementation, so I could have a look, I thought it was possible to change the phase cycling steps.
That being said, the solution that @admin provides of course works for your question. I do agree with your later point about the subsequent moving average blocks containing largely overlapping information, but you can of course play around with averaging smaller blocks and assessing its effect on the outcome - there’s likely a different optimum number of averages in terms of SNR and sensitivity to the neural effect you’re trying to measure for different field strenghts and voxel locations.


By the way, would there be any interest to set up an fMRS brainstorm/discussion/networking meeting at the ISMRM next week?


Very interested, @AnoukSchrantee. I’ll send you an e-mail.

Dear fMRS enthusiasts,

Hope you are all doing well, and looking forward to the ISMRM next week.

We thought it might be nice to organize a small get-together to discuss some topics in functional MRS, and to get to know each other.

Considering time zones and parallel sessions we would like to suggest a one-hour meeting on May 18th, 6pm UTC . Please DM or email us for the meeting link.

Although this will be an informal meeting, it is perhaps good to have a bit of an agenda:

  1. Welcome and introductions
  2. Opinions and current practices in fMRS data processing
  3. Suggestions other topics for discussion
  4. Discuss potential future meetings

If you can’t make it, but still would like to be involved, let us know. All suggestions and ideas are welcome, so please get in touch.

Hope to see you next week,

Georg Oeltzschner & Anouk Schrantee

1 Like

Dear Anouk and Georg,

I’m would also like to join the fMRS discussion and learn from this. Actually one of my collaborators is planning to start with a project in the coming weeks. Can you provide me with the meeting link?

Thanks in advance,

1 Like

Just forwarded you the link via email!