Dwi2fod produces corrupted voxels

[mlyon]

Hi MRtrix experts,

Using the recommended pipeline for multi-shell FBA produced a couple of strange artefacts in wmfod.mif when performing dwi2fod.

These artifacts (as seen below) only appeared on 2 of the DWI images being processed, and are quite different to each other. Strangely, upon inspecting the input image dwi_denoised_preproc_bias.mif there seems to be nothing wrong (though I could have missed something). The same artefacts are present in the corresponding gm.mif and csf.mif outputs too.

The commandline I used to run dwi2fod was as follows:
dwi2fod msmt_csd dwi_denoised_preproc_bias.mif ../group_average_response_wm.txt wmfod.mif ../group_average_response_gm.txt gm.mif ../group_average_response_csf.txt csf.mif -mask dwi_mask.mif -force

If you would like the image data to reproduce the error let me know.

Help much appreciated as always,

Matt

dwi2fod processed image1

dwi2fod processed image2

dwi_denoised_preproc_bias1.mif

dwi_denoised_preproc_bias2.mif

[jdtournier]

I’ve not seen anything like this before… The processing in dwi2fod is purely voxel-wise, so if it works OK in most of the brain, but not in these few voxels, there’s most likely something odd about those voxels. The values you’re showing particularly in the sagittal section (7e+27) is clearly a long way off reasonable, are you sure your data don’t have funny values for any of the volumes in these regions? All it would take is one volume with ridiculous values in these regions to cause that kind of issue.

Otherwise, try forcing single-threaded operation (-thread 0) just in case it might have something to do with multi-threading - but I seriously doubt that, given the appearance of these artefacts… If it does make a difference, we can try to investigate further.