Hi all,
I’ve been having problems for some diffusion images when I run them through dwi2tensor and tensor2metric to create colour coded maps. The maps come out looking wrong:
I have this problem when I convert the dicom data to .mif using mrconvert and run it though dwi2tensor and tensor2metric, and also when I firstly convert the dicom data to nifti using dcm2niix and then run it through dwi2tensor and tensor2metric. However when I pass the niftis through FSL’s dtifit, the colour-coded maps look ok:
So there seems to be some specific problem with MRtrix reading the data. I’m not sure what could be causing this issue and I wondered if anyone could provide any advice on things to check or possible solutions. I received the dicom data from a different site. It was acquired with a Siemens Skyra. I wondered if there could be a problem with anonymising the dicom data as mentioned in this previous post (Diffusion Gradient Information Philips Acheiva). Since I received the dicom data from a different site I’m not sure what the process was to anonymise the dicom data.
If anyone is able to provide any help that would be greatly appreciated. Let me know if I can provide anymore information.
Many thanks,
Claire
They might not be wrong actually (though you’re entirely right to worry whether they are in this extreme case!). The colours in these maps, as well as the orientations of the eigenvectors themselves are defined with respect to the world/scanner coordinate system. If this is the explanation for this dataset, then the difference with FSL’s dtifit here and subsequent colour map in FSL is also sensible: in FSL, a lot (if not all) things are done with respect to the image (voxel) coordinate system.
We can check this directly, to make sure there’s nothing wrong otherwise. When you loaded the ev.mif image in the first screenshot, can you add these 2 things:
The orientation labels. I suppose you hid all label/info type of stuff to make the image more clear to to anonymise further, but the orientation labels specifically here are helpful to put the colours in context. After hiding all the other stuff with the space bar, you can show the orientation labels (only) again via the “View” menu, or directly by pressing the “O” key on your keyboard. If you imagine connecting the orientation labels by 2 lines (A with P, and L with R), you can see the world coordinate system. My guess / hope here is that it sits at a substantial angle. The colours are relative to that then: red is indeed left-right, but with respect to the L and R labels in world space.
Load the same ev.mif image also (additionally) via the Fixel plot tool. If the orientations of the vectors themselves align with the anatomy, this would also confirm that everything is actually correct, but indeed due to a world coordinate systems that sits at quite an angle.
I’ve recently emailed to someone else (very close to us ) who ran into the exact same scenario actually. I’ll forward and include you in the email conversation. Depending on what you need or want to generate here e.g. for visualisation, there’s a few different solutions. I’ll help you out offline; there’s a couple of tricks that might apply here.
) who ran into the exact same scenario actually. I’ll forward and include you in the email conversation. Depending on what you need or want to generate here e.g. for visualisation, there’s a few different solutions. I’ll help you out offline; there’s a couple of tricks that might apply here. 