Clarifications on input structure for most functions

[gdevenyi]

When working with our various quantitative imaging components, we follow a BIDSified format where all our inputs are separate files. It appears various functions in QUIT require images with stacked-dimensionality of some kind to store multiple image inputs for processing (ex #37)

Could you please provide some additional documentation on the expected input format for the various functions? Thanks!

[spinicist]

All the QUIT commands require the main inputs to be "stacked" along the 4th / time dimension in the input nifti file. My own qMRI sequences on GE write out DICOMs that will convert to the corresponding niftis, and I find the format to be far more convenient to work with than separate files. For instance, CEST experiments usually have >20 offset frequencies, it's much easier to have one file than 20 separate ones. It's also the convention for diffusion imaging.

This is one of several reasons I have ignored BIDS as much as possible.

[gdevenyi]

Can you please enhance the documentation regarding the expected order of these stacked 4D files where possible? Thanks.

[spinicist]

What documentation do you want to see and where do you want it to go?

I do not mean to be obtuse - as I started working with fsl, and this is the way all the fsl tools work, I've taken this format to be self-evident.

[gdevenyi]

Hello.

I'm back again with some issues with function input.

I'm trying to prepare data for qi ssfp_bands

Based on prior reading and experience I think the function a 4D image, and I have the complex reconsruction, so I'm trying to use that:

fslcomplex -complex real_000deg.nii.gz imaginary_000deg.nii.gz complex_000deg.nii.gz
etc

However, now trying to make a 4D input:

> fslmerge -t ssfp.nii.gz complex_180.nii.gz complex_090.nii.gz complex_000.nii.gz complex_270.nii.gz
COMPLEX not supported as an independent type

Am I misinterpreting the inputs or does a different tool need to be used?

I've taken this format to be self-evident.

I have never used FSL before. I come from MINC-land.

[spinicist]

I suggest using qi planet instead of ssfp_bands. In essence the PLANET method finds the SSFP signal ellipse parameters, so you get 3 images with interesting contrast, whereas ssfp_bands only finds a single point..

To answer your actual question: I think the trick is to run fslmerge on the real and imaginary parts first, then use fslcomplex or qi complex. I don't have an equivalent to fslmerge in QUIT (there is qi select when going the other way).

[gdevenyi]

Thanks for the information. I've managed to construct inputs that work in ssfp_bands, and we're now investigating the planet reconstruction, however I've run into another thing I need to resolve, and it looks like QUIT has the feature.

My real and imaginary data is actually two different coils, stored as a "time-series" in each of the NIFTI files after reconstruction (real image 2 coils, 2 timepoints, imaginary image 2 coils, 2 timepoints). I think I can use coil_combine with the hammond method to combine them, but I'm again blocked by not undersatanding the input format.

I've tried this:
qi complex -r real_timeseries_two_volumes.nii.gz -i imaginary_timeseries_two_volumes.nii.gz --COMPLEX=complex.nii.gz

But coil_combine doesn't like it:

qi coil_combine  complex.nii.gz
terminate called after throwing an instance of 'itk::ExceptionObject'
  what():  /home/runner/work/QUIT/QUIT/build/vcpkg_installed/x64-linux/include/ITK-5.3/itkVectorImage.hxx:52:
ITK ERROR: VectorImage(0x55b03fdefd90): Cannot allocate VectorImage with VectorLength = 0
Aborted (core dumped)

[spinicist]

Coil combination is a total nightmare. It's a very long time since I've looked at the code in coil_combine. The best way to combine channels is with SENSE maps, but I assume you don't have those? For ssfp_bands you can take the root-sum-squares after you've run ssfp_bands. You could do that for the G output with planet, but not the shape parameters (a and b), I think.

Is this pre-clinical data? 2 channels is very uncommon (I've seen data from the 4 channel Bruker coil, I don't think I've ever seen only 2). Can you either share the data or run qi hdr on your real, imaginary, and the output complex data please?

[gdevenyi]

The best way to combine channels is with SENSE maps.

I do not in this case since the data was already collected. In general I don't think I can do this because as I understand it I need a volume-transmit coil to do this, which I don't have for this application

For ssfp_bands you can take the root-sum-squares after you've run ssfp_bands.

Thanks, that indeed worked.

Is this pre-clinical data? 2 channels is very uncommon (I've seen data from the 4 channel Bruker coil, I don't think I've ever seen only 2).

Here's how the two ssfp_bands outputs look before combination, you can see the two coil sensitivities on either side of the sample:

Indeed it is. We have a 2-channel transceve cryo-coil for a Bruker. Its 2-channel because 1. we started early with cryo and 2. we didn't have enough channels available in our hardware for 4.

[spinicist]

I do not in this case since the data was already collected. In general I don't think I can do this because as I understand it I need a volume-transmit coil to do this, which I don't have for this application

You can use the root-sum-squares of the channel images as an approximate reference instead of the body coil. This works well at 3T for clinical, and starts to get progressively worse at higher field strength as the wavelength decreases. Because you have pre-clinical data, you don't have to care about the wavelength.

Thanks, that indeed worked.

Excellent! I'm happy you have a solution. I'd be interested to know if you get planet working as well. Thinking about it, if you did the same root-sum-squares trick for G, you could then pick the a and b values voxel-wise from the image with the greater G. I think that would work.