Sample dta uploaded in ZIP file

In the current Slicer 05-19-2015
Load images:
preoperative.nrrd
intraoperative.nrrd

Load transforms:
AfBS2.h5 BSPline transform followed by linear transform
LT2.h5 The linear transform separate

Set intraoperative as background image and preoperative as foreground image in a viewer
Switch to Data module
Put preoperative under AfBS2.h5 , this should transform preoperative onto the intraoperative image by BSpline followed by a Linear Transform but only the BSpline transfroms the preoperative image.
Drag preoperative under LT2 and then drag preoperative to the AfBS2 and the preoperative is transformed onto the intraoperative image

I'm trying the steps but I'm not understanding the issue.

To start, the volumes do not overlap.

Dragging preoperative under LT2 makes them overlap.

Dragging the preoperative under AfBS2 makes them overlap and applies the BSpline.

If you look at the AfBS2 Information in the transforms it includes both steps (linear and nonlinear).

If you drag preoperative under LT2 and then drag LT2 under AfBS2 then the volumes do not overlap.

I believe this behavior is correct and Slicer 4.4.0 was incorrect.

I'm going to run this by Andras for his analysis.

The volume don't overlap to start. AfBS2 was made with BRAINSfit affine and BSpline registration modes checked to transform preoperative onto intraoperative with 1 transform object (containing 2 transforms).

We shouldn't need LT2.

Slicer 4.4.0 does 2 transforms transforming preoperative on intraoperative.

I made LT2 with only affine checked in BRAINSfit to test the linear transform separately from the BSpline transform.

The transforms look good, they seem to be applied correctly. However, there seems to be an issue with the red slice viewer update: if the foreground volume is set before the transform is applied to the foreground volume then the red slice viewer does not show the foreground volume (the yellow and green slice viewers are OK). After I switch the foreground/background volume in the red slice viewer it starts to behave correctly.

Diedrich, can you confirm that if you switch the foreground and background volume (intraop foreground, preop background) then it starts to work correctly?

I wonder if this is related to the layout manager issue that came up recently.

https://github.com/Slicer/Slicer/commit/2118883d08f404d5ec035c5271dbec1b7e83532b

After several hours of debugging I've finally found the root cause of the issue:

in int vtkImageReslice::RequestUpdateExtent() if this->ResliceTransform is not a vtkHomogeneousTransform then the hole input extent is updated and then the method returns. However, it does not touch the this->HitInputExtent flag. If the this->HitInputExtent was 0 before then the output of the filter remains always 0.

Probably the simplest fix is to set the HitInputExtentFlag to 1 if the reslice transform is not homogeneous:

VTKv6/Imaging/Core/vtkImageReslice.cxx:

@@ -592,6 +592,7 @@ int vtkImageReslice::RequestUpdateExtent(
{ // update the whole input extent if the transform is nonlinear
inInfo->Get(vtkStreamingDemandDrivenPipeline::WHOLE_EXTENT(), inExt);
inInfo->Set(vtkStreamingDemandDrivenPipeline::UPDATE_EXTENT(), inExt, 6);

• this->HitInputExtent = 1;
  return 1;
  }
  }

I've sent an email to David Gobbi to confirm that this is an issue in vtkImageReslice.

Wow, amazing detective work Andras! Shall we put that in github.com:Slicer/VTK and propose it to kitware:VTK/VTK?

I've sent a pull request to fix it in Slicer's VTK fork:
https://github.com/Slicer/VTK/pull/6

I've checked the process for sending patches to VTK and it seems to be very long/difficult and I don't have any more time to spend with this issue.

Understood - I merged the pull request and am testing on mac. If all is good I will update the version of VTK in slicer.

Updated slicer/vtk superbuild hash in r24272

Thanks Andras!

Karl, please test on your own build or tomorrow's nightly and close the issue if all is good.

David has integrated the patch into VTK:
https://gitlab.kitware.com/vtk/vtk/merge_requests/227

Nice. Looks like it all worked as intended.

Yes, it works with the fixed and moving images in both the foreground and background.