Bids Naming Convention for Changing Voltage Scan and B1+ map

[Kevancic]

We’re trying to organize our ex vivo MRI data into BIDS-compliant format, and are facing issues with our B1+ maps.

We are not using any of the standard B1+ mapping protocols listed in the qMRI appendix and are not sure how to best organize this data. To measure B1, we acquire FLASH images with many different flip angles and a long TR, and fit the bloch equation using a look-up table.

I see three potential ways of organizing this data:

Create a new TB1 file collection (TB1VFA?), but this does not seem to be BIDS-compatible Use one of the standard TB1 suffixes. This feels very wrong. Add them as anatomical VFA data. But I am worried it will confuse users, who will have to differentiate between the actual high-resolution anatomical VFAs, and the low-resolution VFAs used for B1 mapping. Is there a good way to do this?

Thanks!

[lukeje]

Hi @Kevancic,

I noticed that you made a similar post a while back, and have also used information from there in my answer below. As a disclaimer, this is just my opinion as I did not work on the original qMRI-BIDS standard (though I have worked on refactoring the B1 mapping code in the hMRI toolbox).

The file collection you describe is very similar to the double angle mapping (TB1DAM) approach, however in that case only two flip angles are acquired and it is expected that the flip angles have the specific relationship FA2 = 2*FA1, as that simplifies the arithmetic. A "quick and dirty" hack would be to use the TB1DAM suffix but in addition to _flip-1 and _flip-2 also have a _flip-3 to _flip-16 in the fmap directory. This might pass the BIDS validator if it doesn't check for the number of flip angles. I would expect that software packages would expect the first two angles to have the DAM relationship, and so you may want to order your data in that way if you have e.g. a 30° and a 60° case in your dataset.

However you are correct that the proper way to put this into BIDS is to make a pull request with a new TB1 file collection. This file collection would then include TB1DAM as a special case. This would require you to work with one of the qMRI-BIDS developers to get it into the standard.

You mentioned in the previous thread that you couldn't find literature on this sort of acquisition. One source could be

Bouhrara, M, Spencer, RG. Steady-state double-angle method for rapid B1 mapping. Magn Reson Med. 2019; 82: 189– 201. https://doi.org/10.1002/mrm.27708

as they include the full steady state equation and a look-up-table rather than assuming complete relaxation between excitations, though they still only use two flip angles.

As an aside (i.e. just out of curiosity, as I don't think these details are relevant to answer your question) there are some aspects of your method that are still unclear to me, specifically:

1. Why do you use the Bloch equations rather than one of the standard steady state results for your fitting?
2. How do you account for the sign ambiguity from the magnitude operation for flip angles above 90°? Do you phase your data first? Or just take the magnitude of your Bloch results?
3. Do you try to fit a rough estimate of T1 at the same time? Or assume a fixed T1?
4. Is collecting lots of flip angles really more optimal than collecting multiple sets of DAM pairs across the dynamic range of the B1 field?

Best, Luke.

[balbasty]

Hi @lukeje !

Thanks, making it look like a double-angle dataset is smart, we'll try that.

For your extra questions:

1. Sorry our phrasing was unclear: we do use the steady state signal equation (no Bloch simulations). And we do not account for imperfect spoiling.
2. We do not account for the sign and work with the magnitude. When fitting all flips together, there is a single optimum in the range of realistic B1+ values. We did have issue with noise floor in flips that result in very low signal though (we ended up not using them).
3. We do fit a rough T1 at the same time.
4. In our ex vivo scans (which use a legacy custom made ex vivo coil), we get very inhomogeneous B1+ (much more than you'd typically get in vivo), so we do need a bunch of flips to properly estimate it. We probably don't need all 16 flips, but since ex vivo time is cheaper, there's no loss acquiring the full circle. How would you put a map together from a series of double-angle fits, though?

Thanks again!

Yael

[lukeje]

Hi Yaël,

Thanks for the further information; that all makes sense.

Regarding pairs of angles, I was rather wondering what the most optimal way to sample the full dynamic range is. Is equispaced sampling optimal, or would it be better to choose e.g. FA1 = 15°, FA2 = 30°, FA3 = 60°, etc. which simplifies the structure of the estimation problem by reducing it to the estimation of a polynomial in FA1? As you say, though, since you have lots of time ex vivo, there is less need to evaluate what the optimal protocol would be.