ANTs registration (rigid) - *InverseWarp.nii.gz not found

[Cris Lanting]

Hi All,

In a NiPy-pipeline I've included two separate (ANTs-based) registration steps:

1. A rigid-body (6 DoF as far as I can see) registration (reg.inputs.transforms = ['Rigid']) to go get from functional data (EPI)-space to the individual T1-space

2. A non-linear step from the individual T1 to a template (reg.inputs.transforms = ['Rigid', 'Affine', 'SyN'])

With step 1 I've ran into a problem previously found and described elsewhere (https://github.com/binarybottle/mindboggle/issues/15#issuecomment-18250434):

FileNotFoundError: File/Directory '['/data/local_dir/tamac/Connectivity/tmpdir_FS2_smooth/resting/restpreproc/registration/_subject_id_n3875/CoregAnts/func2highres_0GenericAffine.mat', '/data/local_dir/tamac/Connectivity/tmpdir_FS2_smooth/resting/restpreproc/registration/_subject_id_n3875/CoregAnts/func2highres_1InverseWarp.nii.gz']' not found for Registration output 'reverse_transforms'.

Especially the *_1InverseWarp.nii.gz seems to be missing. This was acknowledged a bug (may 2013; see the link mentioned) but was wondering whether there is a work-around by now. I'm using Nipype v. 0.9.1 and Ants 1.9.x SVN 1789. What I don't understand is that it seems to work when I isolate the registration to the following  (now on OSX instead of linux but surely this is not the issue):

import os

from nipype.interfaces.ants import Registration

import nibabel as nb

from nipy.labs import viz

datadir = '/Users/cplanting/Projects/RestingState/n3806'

input_images=[

os.path.join(datadir,'FFE_brain.nii.gz'),

os.path.join(datadir,'mean_brain.nii.gz'),

]

# see https://github.com/binarybottle/mindboggle/issues/15

reg = Registration()

reg.inputs.fixed_image =  input_images[0]

reg.inputs.moving_image = input_images[1]

reg.inputs.num_threads = 2

reg.inputs.output_transform_prefix = 'output_'

reg.inputs.output_warped_image = 'output_warped_image.nii.gz'

reg.inputs.transforms = ['Rigid']

reg.inputs.transform_parameters = [(0.1,)]

reg.inputs.number_of_iterations = [[1000,500,250,100]]

reg.inputs.dimension = 3

reg.inputs.write_composite_transform = True

reg.inputs.collapse_output_transforms = False

reg.inputs.metric = ['MI']

reg.inputs.metric_weight = [1] # Default (value ignored currently by ANTs)

reg.inputs.radius_or_number_of_bins = [32]

reg.inputs.sampling_strategy = ['Regular'] 

reg.inputs.sampling_percentage = [0.25] 

reg.inputs.convergence_threshold = [1.e-8]

reg.inputs.convergence_window_size = [10] 

reg.inputs.smoothing_sigmas = [[3,2,1,0]]

reg.inputs.sigma_units = ['mm']*2

reg.inputs.shrink_factors = [[8,4,2,1]]

reg.inputs.use_estimate_learning_rate_once = [True]

reg.inputs.use_histogram_matching = [False]  # This is the default

reg.inputs.output_warped_image = True

reg.inputs.winsorize_lower_quantile = 0.01

reg.inputs.winsorize_upper_quantile = 0.99

reg.inputs.write_composite_transform = True

reg.run()

normed_img_fname = '/Users/cplanting/output__Warped.nii.gz'

normed_img = nb.load(normed_img_fname)

fixed_img_fname = input_images[0]

fixed_img = nb.load(fixed_img_fname)

slicer = viz.plot_anat(normed_img.get_data(), normed_img.get_affine(), dim=.2, black_bg=True)

slicer.edge_map(fixed_img.get_data(), fixed_img.get_affine())

Any ideas?

[Satrajit Ghosh]

hi chris,

this is puzzling. are your nipype and ants versions the same on both OSX and the other machine?

cheers,

satra

--

---
You received this message because you are subscribed to the Google Groups "NiPy Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nipy-user+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Cris Lanting]

Hi Satra,

With a bit of debugging I've found the culprit:

with:
reg.inputs.collapse_output_transforms = True

it generates the following output:
output_0GenericAffine.mat
output_Composite.h5
output_InverseComposite.h5
output__Warped.nii.gz

it the results in the reported error.

with:
reg.inputs.collapse_output_transforms = False
output_0Rigid.mat
output_Composite.h5
output_InverseComposite.h5
output__Warped.nii.gz

When you think of it a bit more it makes sense (sort of): since I ask for a rigid transform only, there is one output transform. Perhaps it needs two or more to be able to 'collapse' the ouput transforms?! I also noticed that with collapse_output_transforms = true it generates a 0GenericAffine.mat file wheres if it's false it results in a 0Rigid.mat output.

Cheers,

Cris

[Satrajit Ghosh]

hi chris,

that makes sense. however, we should improve the registration interface to ensure that it doesn't try to fill that output field when collapse_output_transforms is True and there is not non-linear transform selected.

the generic affine vs rigid is also understandable.

let's say you have as transforms: ['translation', 'rigid', 'affine']

these can all be represented by a 4x4 matrix, so when collapse is true these are combined into a genericaffine.mat file.

cheers,

satra

--

[Satrajit Ghosh]

issue filed here:

https://github.com/nipy/nipype/issues/883

cheers,

satra

[Cris Lanting]

Thanks. 

Cheers, Cris