FSL(fsleyes) displays image space issues (spec2nii GE MRSI)

[doctorjia]

I am a PhD student in neurosurgery, currently using FSL to process magnetic resonance spectral data. I used spec2nii to convert P-files to NIFTI format and then processed them with FSL-MRS, but the results were completely unreliable. Then, I opened the T2 sequence and spectral data with FSLEYES and found that the image orientation was incorrect; I’m unsure if this is the cause.

I hope someone can help me figure out where the problem is.

Below are the results and detailed information after opening the file using FSLEYES.

[wclarke]

Hi @doctorjia,

Sorry that this didn’t work out of the box. Could you tell us a bit more about the sequence and what you expect to see? What should the orientation be? Is the data reconstructed? I can see some signal right at the centre of the slice in a checkerboard pattern

What does the spectrum look like there?

[doctorjia]

@wclarke Thank you for your reply.Below is my processing flow and the issues I encountered. I am very willing to provide you with the original data if necessary.

1.data information

The GE P file is converted to Nifti format using spec2nii to generate the header JSON information.
P87040.txt (971 Bytes)

mrs_tools info raw_data/P87040.nii.gz

mrs_tools vis raw_data/P87040.nii.gz

2.preprocessing

According to the FSL-MRS documentation(Processing — FSL-MRS 2.4.13 documentation; Quick Start Guide — FSL-MRS 2.4.13 documentation), it was recommended to perform coil-combined and repetitions averaged before fitting, along with specialized preprocessing tools such as inter voxel alignment, phase correction, and lipid removal.

Therefore, I performed coil-combined before fitting only.

Mrs_tools info processed_data/combined/P87040.nii.gz

3.Create Basis Spectra

For the metabolite base set, I used MRSCloud on the MRICloud platform to generate it (https://braingps.mricloud.org/mrs-cloud). The metabolite selection and parameter settings are shown in the figure.

parameter settings
Localization:PRESS,sLASER(I choose PRESS)
Vendor:GE
Editing:unedited,MEGA,HERMES,HERCULES(I choose ubedited)

mrs_tools vis GE_PRESS_3T_144ms_fsl

4.Tissue Segmentation

Because the T2 sequence was used as the anatomical reference when performing MRS, I performed tissue segmentation on the T2 sequence.

5.Fitting

I fitted the coil-combined MRSI data with tissue segmented by T2 sequences.

fsl_mrsi --data processed_data/combined/P87040.nii.gz --basis GE_PRESS_3T_144ms_fsl --output results/P87040 --tissue_frac mrsi_seg_wm.nii.gz mrsi_seg_gm.nii.gz mrsi_seg_csf.nii.gz --internal_ref Cr --report

6.Visualise

fsleyes -smrs processed_data/combined/P87040.nii.gz raw_data/OAx_T2_Propeller.nii.gz &

When I open fsleyes separately and then load the T2 sequence and MRSI, the positions are displayed correctly, but this is not the case when opening these two files using the command line.

7.My questions.

Is there a problem with my processing for this fitting result?

Is it a problem with the metabolite baseline set parameter settings?

Is it a problem with the MRSI preprocessing?

I’m completely clueless.

I even tried adding several separate preprocessing steps based on MRSI recommendations, but the results were still unsatisfactory.

I also tried T1 sequences for tissue segmentation, but the fitting result was the same.

I would be extremely grateful if someone could offer me some guidance.

[wclarke]

Hi @doctorjia ,

Sorry for taking a few days to reply. I think that there are a couple of things going on here.

1. You need a nifti-formatted mask to select just the voxels that have signal, i.e. those within the sLASER VOI. You can do that using FSLeyes (see below) or loading the NIfTI-MRS in python (see fslpy to use a simple thresholding approach.
2. You need to perform some phase and frequency correction. Try using fsl_mrs_proc mrsi-align to do this. take a look at the help (--help) to see how to run it. If this doesn’t work, then let me know and we can refine/use a new tool we have.
3. Then after alignment you can try fitting again, but also provide the mask. The alginment and the mask will should result in a much higher SNR average spectrum being constructed for the initialisation.

Regarding fsleyes, I think what you see is just an artefact of the loading order, which will depend on the order of the files you enter ont he command line. This has two efffects. The first is to ‘hide’ one image behind the other, which image is hidden depends on the order specified. You can change this using the up/down arrows int he bottom left.

The second effect is that the image is oriented in the display space of the first (I think, could be the last). Simply click the spanner icon int he top left, find the display space dropdown menu in the window that opens, and select the T2.