ReadingVectorValuedNRRD

Introduction

This shows how to read vector-valued NRRD files using ITK, for example the GK's DTI datasets.

Details

Make sure that you have a correct "kind" field. In Gordon's helix dataset, this is missing. You should add:

kind: vector space space space

(I got segfaults without it.)

Because of "sizes: 7 38 39 40", you should get a 38x39x40 volume with 7-component vectors using the code below:

# if the "kind" field in your NRRD is correct, e.g. kind: vector space space space
# this should work.

# this is a variable length vector in a 3D image
vitf3 = itk.VectorImage[itk.F,3]
# then make the reader for it
fr2 = itk.ImageFileReader[vitf3].New()
fr2.SetFileName('c:/tmp/dti/dt-helix.nhdr')
fr2.Update()

# the output is a 3-D image of variable length vectors
o = fr2.GetOutput()

# now write out as multi-component metaimage file
fw = itk.ImageFileWriter.VIF3.New()
fw.SetInput(fr2.GetOutput())
fw.SetFileName('c:/tmp/dti/dt-helix.mha')
fw.Write()

Notes

• This is not the best way to read and process these things in ITK+VTK, because VTK wants full 3x3 tensors (although they are symmetric!) and you'd have to code that up (in Python). See AddingVTKTeem for details on how to use vtkTeem with DeVIDE, this has built-in support for reading NRRD tensor files into VTK tensor representation.
• If the kind is 3D-masked-symmetric-matrix (instead of vector), the NRRD code gives you only the 6 unique components of the diffusion tensor, and not the confidence / mask (the first component).