Canon diffusion gradient vectors

[neurolabusc]

Canon has recently acquired Toshiba. In the past, Toshiba used public tags to report diffusion properties. Specifically, DiffusionBValue (0018,9087) and DiffusionGradientOrientation (0018,9089). This is not the case with the latest Canon software release, where the tag DiffusionGradientOrientation is not generated. Instead, the gradient direction is stored as text in the ImageComments tag. While this usage is not documented, validation datasets suggest that the order of the vector component storage is [Y,X,-Z]. This solution seems robust for all images evaluated. However, note that all sample images use InPlanePhaseEncodingDirection (0018,1312) of COL, so it is unclear if this proprietary component order is correct when InPlanePhaseEncodingDirection is RAW.

The header looks like this:

 (0018,9087) FD 1500                                     #   8, 1 DiffusionBValue
 (0020,4000) LT [b=1500(0.445,0.000,0.895)]              #  26, 1 ImageComments

The recent release of dicm2nii and the developmental release of dcm2niix (v1.0.20200828) support (our interpretation of) this unusual nomenclature.

I would suggest users of Canon MRI should pay special attention to diffusion parameters. I would also urge users to lobby their Canon Research Collaboration Managers to return to using clearly specified public tags (0018,9089) for reporting this vital information. Clément Debacker deserves credit for providing a publicly sharable dataset that illustrates the issue and provides evidence for the solution.

[neurolabusc]

New sample dataset. Note that the DICOM header does not distinguish between phase encoding polarity (A>P vs P>A) not does it correctly remap the reported b-vectors. Therefore, there is no universal solution for converting the images. On may need to invert the y-component of the bvec file. Also, there are no examples of RL vs LR encoding, so these may not be correct. dsi-studio users should use the Check b-table option. AFNI users should use grad-flip-test. Users of other tools should use extreme caution. This reflects a limitation of the DICOM images, not dcm2niix. A robust solution will require intervention by the vendor.