Bruker Diffusion Directions

[neurolabusc]

I was recently sent a Bruker dataset where the correct conversion required reverting

if (!(d.manufacturer == kMANUFACTURER_BRUKER && d.isDiffusion) && (d.xyzDim[4] > 1) && (d.xyzDim[4] < kMaxDTI4D)) { //record variations in TE

to read

if ((d.xyzDim[4] > 1) && (d.xyzDim[4] < kMaxDTI4D)) { //record variations in TE
and recompiling dcm2niix

This change was a pull request made by isolovey to resolve issue 282
#282
In particular the pull request "Bruker 4D datasets: Fix parametric maps grouping"
8207877#diff-120bb215d28dcbddeca8ea56dbb97e237501901ac2dfa1fbe4182282a8dd2b75

This was an explicit change by a Bruker user that explicitly only influences Bruker diffusion output. Since I do not have access to their data, I must assume this change was required. The modification seems to suggest that at least at that time the Bruker DICOMs did not store bvec/bvalue variables in the order specified by the DimensionIndexValues 0020,9157 that orders the volumes and slices.

Perhaps the developer was in error, or perhaps Bruker realized that they were generating invalid DICOM files.

I have reverted the code in the development branch, but Bruker users should validate these solutions. .

The new data was acquired with Acquisition PV-7.0.0\ParaVision 7.0.0. If Bruker fixed an error with a specific software version, we could ensure robust conversion for both old and new images. Without any insight, I fear that either solution will fail with some users.

It is clear that Bruker has improved aspects of their DICOMs, for example providing 0018,9087 rather than the polarity-ambiguous 0018,9602..0018,9607. I aplaud these changes, but without having access to this moving target does make the job of dcm2niix difficult.

I strongly suggest that Bruker users Validate DTI Vectors for their datasets and report results as suggested in the manual. Validation diffusion datasets benefit from relatively low spatial resolution that provides strong single to noise for the major fiber bundles.