Version 3.0 dwi2response gives nan

3

Just one further clarification:

The reason the behaviour has changed for you is because in MRtrix3 Release Candidate 1, dwi2response was altered to use the new amp2response command, as opposed to sh2response. The details of this command can be found here.

  • sh2response simply rotates the SH decomposition of each single-fibre voxel so that the fibre is aligned with the z-axis, then averages across voxels. In this situation, l>0, m=0 terms will be dominated by noise: the signal intensity shouldn’t be influenced by fibre orientation at b=0!

  • amp2response is encountering a fundamental issue in obtaining a solution. It uses the signals from all single-fibre voxels at once; but it’s facing a conflict between having a signal intensity that shouldn’t vary as a function of angle from the fibre orientation, and a constraint that forces monotonicity as a function of angle from the fibre orientation (details in the abstract linked above). Hence the NaN result.

So I’d argue that NaN is in fact the more sensible answer of the two, since it’s highlighted the error in your -lmax option :smiley: