Mrinfo with DICOM

Hi, I was hoping if someone can help me make sense of this. I want to read the diffusion gradient directions from the raw DTI DICOMs (obtained from a Philips scanner) using mrinfo. I know that DICOM uses a LPS coordinate system (ie from R->L, A->P and I->S). So, mrinfo is showing strides [-1 -2 3 4], which is correct. However, I found out that the x and y axes of the gradient directions were flipped. Is this correct? I have attached 2 screenshots here. One showing the gradient directions using mrinfo. The other showing the directions using dicominfo in MATLAB (I got the same results using dcminfo in mrtrix).


Hi @brc0520,

This is simply down to the fact that MRtrix stores directions in real/scanner space RAS (same as the NIfTI standard frame), whereas DICOM uses LPS (as you noted) ā€“ so the only difference is in the inversion of the x & y axes.

This contrasts with the bvecs/bvals format which stores directions relative to the image axes.

I hope that answers your question?

1 Like

Thanks Donald. Yes, this is very helpful. Thanks for clarifying the bvecs/bvals as well. Does this means the gradient directions will always be stored (as dw_scheme) with respect to the scanner space with a RAS coordinate system regardless of the strides/format shown by mrinfo? Please correct me if Iā€™m wrong.

I am suspecting that the raw gradients in my DICOMs were stored with respect to the image space (or at least differently). This suspicion was due to the fact that the raw directions between participants were different despite that they were scanned with the same protocol and the fibre tracts were also implausible. Is there a way to tell mrtrix to treat those raw directions as relative to the image axes? Or is there a way to transform those directions back to the scanner space? (I know dwgradcheck is a great tool to address this but I think it may not work for our application about muscle fibres).

Hi Donald, I have another question relating to this topic. I have just acquired another DTI scan and export them in DICOM and PAR/REG. I found that the x, y and z axes of the raw gradient directions were inverted when i run mrinfo on PAR/REGs as compared to the raw gradient directions viewed with mrinfo on DICOMs. See image 1 (mrinfo on DICOM) and 2 (mrinfo on PAR/REG) for details. Could it be a potential bug?


If all elements of the gradient vector are inverted, as is the case here, then they are essentially equivalent. You should think of the gradient vectors as orientations (where the sense does not matter) as opposed to directions (where the sense does matter). So I think everything is fine.

I see. Yes, that make sense. Thanks Ben.