Voxel flipped L-R when using nifti T1 for GE data

MRS Software Osprey
1

Dear development team,
I have discovered that the voxel mask created in the coregistration step seems to be flipped left-right, see the attached pdf file. This will probably affect the obtained tissue fractions in the segmentation step and also the quantification based on water and tissue corrections but this needs to be checked carefully to make sure that the correct voxel position is used in the calculations.
/Greger

OspreyVoxelFlipped.pdf (519.7 KB)

1 Like
2

Hi Greger,

Thatā€™s indeed concerning, thank you for the bug report. Can you please let me know whether you used DICOM or NIfTI data for the co-registration, and what GE software version this is? Iā€™m quite sure we got it right for Siemens and Philips, but Helge and I personally havenā€™t done a lot of GE processing, and the feature to co-register straight to NIfTI instead of DICOM has only been added recently.

Thanks for helping us make Osprey better, and happy holidays,
Georg

3

Hi Georg,

I was collecting data on a GE MR750 running on DV26.0 R03.
For the coreg I used nifti data created from the DICOMā€™s with Chris Rordenā€™s dcm2nii.

Happy holidays to you and the rest of the development team,
/Greger

4

Hello,

Echoing Gregerā€™s finding. I also have found that the voxel mask is flipped left-right when using nifti T1 for Siemens data.

I too created nifti data from the DICOM using dcm2nii by Xiangrui Li (xiangruili/dicm2nii - File Exchange - MATLAB Central)

Kind regards,
Mike

5

Hi @Mike,

Thanks for your feedback.

Iā€™ve tested with some Siemens data that I had at hand. I used the following combinations with Siemens MEGA-PRESS WIP spectroscopy data from VE11C:

  • rda file with nifti generated in SPM12 and generated with option 1 and 5 from Xiangrui Liā€™s converter
  • twix files with the same combinations

They were all co-registered right compared to the reference image from the scanner and Gannet 3.1

Would you mind sending me your spectroscopy and DICOM data, so that I can do some further digging?

Best,
Helge

1 Like
6

Hi @Helge,

Sorry for the late reply. I have sent through email, a link to an example folder containing the DICOM and Nifti files.

Kind regards,
Mike

1 Like
7

Hi @Mike,

Is the WM voxel supposed to be in the left or right half? (See attachment )

sub-38_ses-01_WM_press_Voxel_1_OspreyCoregSeg.pdf (556.4 KB)

Best,
Helge

1 Like
8

Hello @Helge,

The WM voxel should be coming up in the left parietal white matter.

Kind regards,
Mike

9

Hi @Helge,

As an update, Iā€™ve tried using two other dcm-to-nii converters: MRIconvert (MRIConvert files ā€” LCNI) and Mango (Research Imaging Institute ā€” Mango).

With MRIconvert I seem to run into the same problem as before with Xiangrui Liā€™s converter (image 1).

With Mango, it seems to have shifted all coordinates (image 2).

I have also tried just flipping the image and saving it again (image 3). Seems to work but alignment is still a bit out.

CoregSeg_ConversionExamples.pdf (113.5 KB)

Essentially the voxel placements were aligned to the posterior gray matter and parietal white matter as indicated in:

Lin, A., Tran, T., Bluml, S., Merugumala, S., Liao, H. J., & Ross, B. D. (2012, September). Guidelines for acquiring and reporting clinical neurospectroscopy. In Seminars in neurology (Vol. 32, No. 04, pp. 432-453). Thieme Medical Publishers. DOI: 10.1055/s-0032-1331814

Would I need to specify the folder containing all of the associated dicom files? I have tried that but keep getting the error:

Error using spm_vol>spm_vol_hdr (line 80)
File ā€œH:\PhD\RA-MRS\Jobfiles\ErrorChecks\sub-43\ses-01\anatā€ does not exist.

Kind regards,
Mike

10

Hi @Mike,

thanks for the update. So just to clarify the current state. The PDF that I posted was still not right? I did test it with the nifit you send me and a nifti generate with MRIcroGL and the voxel position matched.

It is relatively hard for me to figure out how for off the positioning is, as I did not run the study or have a real reference. Do you have the voxel positioning images generated on the scanner ( It is something like save images and spectra in the spectroscopy tab) or can you go back to the scanner to generate them? It would be much easier if I could see them or a screenshot/photo of the on scanner positioning.

DCM input is currently only possible for GE data. You could look into the coreg_p function to see how this is set up, using SPM to generate the nifti.

Best,
Helge

11

Hi. Just in case anyone else stumbles across this thread wondering why their data are flipped L-Rā€¦ Osprey appears to be using NEUROLOGIC convention (L side of brain on L side of image) and NOT radiologic (L on R) convention.

I processed multiple datasets from Philips and GE, using T1s that were converted from dcm2nii using all of the same tools above (i mean all). And achieved the same result. Then tested this with data that I have with clear demarcation of L-R (e.g., unilateral lesions) and am consistently seeing things flipped L-R, but in the appropriate location relative to the anatomy.

Helge, Iā€™m curious, is there a flag I can use to flip things into radiological convention for those of us who prefer that (and already have enough R-L confusion :wink: ?

Jessica

12

Hey all!

Sorry to resurrect an old thread - but I am having a similar problem. I have run some data through both Gannet and Osprey, and it appears that the voxel is being flipped in Osprey.

Iā€™ve attached the coreg output from both Gannet and Osprey - the voxel should be left ventral visual, which it is in Gannet, but not Osprey. As @jwisnowski asks, is there a flag to easily flip things in Osprey?

Thanks so much!
mindy197_S78848_gabaleftventrovisual_vox1_coreg.pdf (831.1 KB)
mindy197_vox-LVV_type-gaba.svs_OspreyCoregSeg.pdf (940.3 KB)

13

Hi,

Adding to @mollysimmoniteā€™s observation - Iā€™m having weirdness with the CoReg function in Osprey on GE data, that has been converted to nifti (through Gannetā€™s routine for the most part I believe).

Pictured is the Gannet (blue) and Osprey (yellow) overlaid

Thanks!
Marilena

image
image1642Ɨ578 62.3 KB

14

Hi @marilena,

could you send me a dataset? And maybe a screenshot of the intended location. Iā€™ll try to figure out were it is going wrong.

Best,
Helge

15

Hi @mollysimmonite,

I hope that this is really just about an LR flip. Iā€™ll come up with an option to do this as this seems to be a common issue with GE MRS/imaging data depending on the NifTi conversion.

Best,
Helge

16

Hi all,

Iā€™ve just added a bug fix for the LR flip to the develop branch which solved the issue for the test datasets that I had at hand.

@mollysimmonite Please let me know if this solves the problem for you.

Best,
Helge

17

sorry for the delay @Helge! Just ran a participant through and the problem is fixed :smiley:

thanks so much!

1 Like