Applying LN2_PHASE_JOLT to multi-echo data

Software Support
, ,
1

Summary of what happened:

I have read the phase jolt in fMRI poster from OHBM 2024 and was interested in applying LN2_PHASE_JOLT to my data. My data are multi-echo (5 echoes), with magnitude+phase reconstruction, from a 3T scanner (see OpenNeuro). To clarify, these data aren’t suited for layer fMRI, I was more curious about using the phase jolt time series for denoising (i.e., phase regression).

@ofgulban, have you applied LN2_PHASE_JOLT to multi-echo data in the past? Do you know of a good way to combine the phase jolt time series across echoes (or, alternatively, combine phase time series across echoes before running LN2_PHASE_JOLT)?

1 Like
2

Glad you found this! One of the idea with phase jolt is to ignore the ugly phase problems - you go into jolt first, which makes life easier. Unless I am very mistaken, you should be able to average across echoes after performing the jolt calculation. Circular math/phase challenges have been removed.

It’s something I want to look into when I get to Maastricht, so you’ll hopefully hear more from me. First idea was to inspect a vein voxel phase jolt time series over echoes - should have similar shapes.

If you wanted to be very goofy you could use magnitude derived weights to average phase jolt images… @ofgulban, you may want to jump in here (and we can discuss in person very soon!).

2 Likes
3

Thanks @dowdlelt! I’m looking forward to seeing what you figure out.

The three weighting schemes I was torn between are (1) the T2* method from Posse et al. (1999), the PAID method from Poser et al. (2006), or the weighted regression based on magnitude weight used in warpkit, from Van et al. (preprint).

1 Like
4

Dear @tsalo , thanks for your interest in phase jolt and questions (thanks @dowdlelt for jumping in as well). Some answers/comments:
1- I have developed and used phase jolt on anatomical multi echo images. Then started looking at fMRI data. I have not yet used phase jolt on multi echo fMRI data.

2- As this is a very new method, I would keep echo combination very simple. I would actually advise you to first compute phase jolt on each echo, then simply average across echoes. I would actually advise you to not do weighted averaging on phase jolt images. This is because phase jolt values lie on a constrained range with noise centered on pi/2, and signal (brain) voxels will he close to zero. If you do weighting, you will disturb the number line properties in an unuseful way. There would be ways to overcome this but it would be wise to keep things as simple as possible at this very early stage.

3- I would advise you to run the LN2_PHASE_JOLT program with -phase_jump flag. This will generate the first derivative (rather than second) analogue of phase jolt (called phase jump). Currently, I think that phase jump has some advantages over phase jolt in the functional imaging context.

4- As far as I can see right now, these phase derivatives are not special to be used in high res / layer fMRI context. So I would not worry about applying it to 3 T data. One difference might be that with higher resolution data, you would be able to detect and regress out the macrovasculature dominant signals better (probably). I am in the process of turning these thoughts and insights into a paper so my advice might change over time :smiley:

Currently, I am traveling and it might take a day or two for me to reply, but I am very interested and excited that you picked up on this. Please do not hesitate to ask further questions :slight_smile:

1 Like