ENH: set Siemens MeasurementFrame

[ihnorton]

Per this page:

https://www.nitrc.org/plugins/mwiki/index.php/dcm2nii:MainPage#Diffusion_Tensor_Imaging

Siemens and Philips report gradient direction with respect to the scanner bore.

DWIConvert does already account for this in the Philips converter, so I believe we should do so for Siemens as well.

I tested this change with one of the large rotation test datasets linked there,
specifically, the 'dtitest_Siemens_GSU_GT/ep2d_DTI_30_ax_roll_4' dataset.

Example DTI image in Slicer, with direction-coded colors (left: current, right: with this patch):

[ihnorton]

By the way, this makes the output bvec file in FSL mode match well with output from dcm2niix:

julia> vec_dwicp = readdlm("test_dwic_patched.bvec")
3×31 Array{Float64,2}:
 0.0  -0.558845  -0.207158  -0.0752197  -0.589165  -0.269029  -0.756038  -0.831669   -0.783142  …   0.078812   0.115386   0.328008   0.44147    0.735267  0.990763     0.915539
 0.0   0.54394    0.529593   0.179027    0.756425   0.953483   0.551316   0.554387    0.209761      0.938754   0.727798   0.16311    0.502128   0.490006  0.1354       0.138558
 0.0  -0.625957  -0.822567  -0.980964   -0.284088  -0.135989   0.352784  -0.0313337  -0.585397     -0.335455  -0.676015  -0.930487  -0.743621  -0.46827   0.00749746  -0.37761

julia> vec_dcm2nii = readdlm("__ep2d_DTI_30_ax_roll_20110331083655_4.bvec")
3×31 Array{Float64,2}:
 0.0  -0.558845  -0.207158  -0.0752197  -0.589165  -0.269029  -0.756038  -0.831669   -0.783142  …   0.0788119   0.115386   0.328008   0.44147    0.735267  0.990763     0.915539
 0.0   0.54394    0.529593   0.179027    0.756425   0.953483   0.551316   0.554387    0.209761      0.938754    0.727798   0.16311    0.502128   0.490006  0.1354       0.138558
 0.0  -0.625957  -0.822567  -0.980965   -0.284088  -0.135989   0.352784  -0.0313337  -0.585397     -0.335455   -0.676015  -0.930487  -0.743621  -0.46827   0.00749746  -0.37761

julia> norm(vec_dwicp - vec_dcm2nii)
2.724273932718676e-6

[hjmjohnson]

@ihnorton Please be careful here. I am not convinced that this is correct. My recollection is that Siemens and Phillips are quite different. Perhaps is scanner version or scanner model dependent.

I would like to see more testing before this is committed.

[ihnorton]

I'm going to close this. After closely reading through Gordon's document, this figure and related pages on the NA-MIC wiki -- I agree the current behavior is correct. (which makes sense, of course... someone would have noticed it in all these years)

I've been working a lot with FSL Nifti<->NRRD, and have had issues with incorrect glyphs from rotated acquisitions. A source of confusion was my misunderstanding that measurement frame (going to NRRD) of FSL-style nifti+bvec should be identity. Sylvain here pointed out that is wrong, and making the change fixed issues with some rotated-acquisition test datasets.

Currently DWIConvert makes the same assumption, which I think should be changed -- e.g. the code behavior for FSLToNRRD matches this comment in issue 342 and sets the output NHDR measurement frame to identity:

Due to the fact that the BVEC and BVAL information is implicitly in an LAS corrdinate system (and the fact that the measurement frame is implicitly an identity matrix

I'll close this and re-open that issue to discuss.

[ihnorton]

Also thanks for pushing back on this.

[hjmjohnson]

@ihnorton Thank you for working through this. I greatly appreciate your careful evaluation.

Hans