Incorrect bvecs for GE DTI studies with ROW phase encoding (dicom_series_to_nifti)

[samshipengs]

We noticed some of our tractography generated are really messy and obviously incorrect. We traced back the cause of this to the bvecs generated from dicom_series_to_nifti, we basically compared the bvecs to the one generated by using dicom2niix (from which the tractography looks correct).

What we realized so far is that, for vendor GE and DTI image, if it is ROW phase encoding (Tag(0x0018, 0x1312)), then the second column of bvecs (b1) should be negated and then first column (b0) and b1 should be flipped. Currently, the code in convert_ge does not handle it (it is correct when the phase encoding is COL, but it does not have a different handle for ROW)

Could you let me know if you guys think that is the case, if so if you guys have plan to fix it, and I'd be willing to create a PR on this as well.

[samcoveney]

I am seeing similar problems when converting a cardiac DTI dicom series. In order to get the tensor orientation correct with respect to the image, the second column of bvecs needs negating, otherwise all the tensors point the wrong way when plotted on the image

edit: this is when using reorient_nifti=False

edit 2: and indeed my data using row phase encoding