MRtrix 3.0 `dwi2fod` single output error

[jguillon]

Hi, I got this error when executing mrtrix3.EstimateFOD interface with the latest MRtrix version (3.0_RC2-117-gf098f097) installed on macOS Mojave 10.14.2 through a brew install mrtrix3:

dwi2fod: [ERROR] CSD algorithm expects a single input response function and single output FOD image

It seems that MRtrix now expects "pairs of input tissue response and output ODF images" according to the documentation. So I guess that the following lines need to be updated:

nipype/nipype/interfaces/mrtrix3/reconst.py

Lines 95 to 109 in 35b58cd

	wm_txt = File(
	argstr='%s', position=-6, mandatory=True, desc='WM response text file')
	wm_odf = File(
	'wm.mif',
	argstr='%s',
	position=-5,
	usedefault=True,
	mandatory=True,
	desc='output WM ODF')
	gm_txt = File(argstr='%s', position=-4, desc='GM response text file')
	gm_odf = File('gm.mif', usedefault=True, argstr='%s',
	position=-3, desc='output GM ODF')
	csf_txt = File(argstr='%s', position=-2, desc='CSF response text file')
	csf_odf = File('csf.mif', usedefault=True, argstr='%s',
	position=-1, desc='output CSF ODF')

[matteomancini]

The issue is that dwi2fod needs to be able to handle both CSD (where the only output is the white matter FOD) and multi-tissue CSD (where there is one file for each tissue). To solve the issue, gm_odf and csf_odf need to have the useDefault field equal to False. I have tested the solution and it works, I already fixed the code in my own fork of Nipype, if it is ok I can make a pull request. A related issue that I fixed there is the useDefault=True for the max_sh argument for mrtrix3.EstimateFOD, which causes troubles for multi-shell data. I already run the make check-before-commit and the code I worked on is fine.

[effigies]

Hi @jguillon, sorry for how long this took. The 1.5.0 release will have a ConstrainedSphericalDeconvolution interface that should work correctly. In case you didn't see other discussion, the concern with fixing EstimateFOD is that people using msmt_csd may have been depending on the bug in their workflows.

The new interface is identical, but without the bug.