Dhollander/msmt_csd - corrupted FOD images

Hi all,

I have been running dwi2response dhollander and dwi2fod msmt_csd to generate FOD maps for each subject in my data. However, when I check each individual’s voxels.mif/vf.mif data, some of data have corrupted/abnormal FOD maps. I couldn’t find what is common among those corrupted images, and I am not sure what kind of parameters I should check.

These are the codes I ran:

dwi2response dhollander ${dir}/${subname}/${subname}_unbiased.mif -fslgrad ${dir}/${subname}/${subname}_merged_dwi_gibbsCorrSubVoxShift_TED_synthB0.bvec ${dir}/${subname}/${subname}_merged_dwi_gibbsCorrSubVoxShift_TED_synthB0.bval ${dir}/${subname}/${subname}_response_wm.txt ${dir}/${subname}/${subname}_response_gm.txt ${dir}/${subname}/${subname}_response_csf.txt -voxels ${dir}/${subname}/${subname}_voxels.mif
dwi2fod msmt_csd ${dir}/${subname}/${subname}_unbiased.mif -fslgrad ${dir}/${subname}/${subname}_merged_dwi_gibbsCorrSubVoxShift_TED_synthB0.bvec ${dir}/${subname}/${subname}_merged_dwi_gibbsCorrSubVoxShift_TED_synthB0.bval ${dir}/${subname}/${subname}_response_wm.txt ${dir}/${subname}/wmfod.mif ${dir}/${subname}/${subname}_response_gm.txt ${dir}/${subname}/gmfod.mif  ${dir}/${subname}/${subname}_response_csf.txt ${dir}/${subname}/csffod.mif -mask ${dir}/${subname}/${subname}_masked.mif 
mrconvert -coord 3 0 ${dir}/${subname}/wmfod.mif - | mrcat ${dir}/${subname}/csffod.mif ${dir}/${subname}/gmfod.mif - ${dir}/${subname}/vf.mif

I would appreciate it a lot if you suggest me anything that might be causing this and how to resolve it.

I attached the images of a normal, and corrupted FOD maps.

Many thanks in advance!!

Rather than looking at / reporting the final result, please inspect/report the outcome of the individual steps, e.g.

  1. Inspect the input DWI data. Any obvious differences between the good and bad dataset?

  2. Inspect the output of dwi2response:

  • did the command output any warnings?
  • overlay voxels.mif on an anatomical scan to check if the exemplar voxels for each tissue type are in the right places
  • open the response functions (txt files) with a text editor and inspect the first column: for each tissue type, do you observe a similar b-value dependent decay in the corrupted vs good case?
  • inspect the response functions (txt files) with shview

Thanks a lot for your reply Ben.

  1. I couldn’t detect any obvious differences between those two datasets.
  2. There were not any warnings, however, csf does not seem right in the right place for bad data (as below). I am quite new at mrtrix, but I couldn’t see any obvious differences in the response function (txt files). On the other hand, csf is clearly getting less smaller in the bad data comapred to good data. I added the images for these as well, the elft one is good, and the right one is the bad data.