SIFT2 different outputs

1

Dear experts,

I’m running SIFT2 in my datase and I found that with the same command, the output weights are different, however the mu is not. Is this possible?

The differences is detected using the command diff in ubuntu.

Best regards,

Manuel

2

Hi,

I think I found the solution, after running some experiments, when I added the flag -nthreads 1 The results are in fact identical, so the only differences are due to the multi-threading, am I right?

Best regards,

Manuel

3

Hi Manuel,

Ideally the calculated streamline weights should not be dependent on the number of threads. I am getting some fuzzy recollection of having identified an issue with the relevant multi-threading code; I’ll have to scan through my various work systems to try to reconstruct whether I addressed it somewhere but failed to upload. I’ll keep you posted.

Rob

4

Hi @rsmith,

Thanks for you reply. I tested with 2 versions of mrtrix3 that I have in my computer:

  • MRtrix 3.0.0 tcksift2 May 13 2020
  • MRtrix 3.0_RC3-15-g9494da8d tcksift2 May 31 2018

And the same happens in both versions. I also nocited that the mu value is slightly different between both versions.

I don’t know how quantify the magnitud of the discrepancy, but the different SIFT2-TDI maps look identical, the difference between images is really small.

Best regards,

Manuel

5

Hi @rsmith,

Did you have the chance to look at this for the RC3.0.2?

Best regards,

Manuel