[vtkusers] Using vtkImageReslice

[David Gobbi]

Hi Andrius,

I haven't read through your entire email, but here are a few comments:

You shouldn't have to use more than one vtkImageReslice.  Instead, you
should incorporate all the rotations and translations into a single 
vtkTransform.

The _one_ vtkImageReslice that you use should have an OutputExtent that
is only one slice thick.  (That is, one vtkImageReslice per slice that
you want to view)

This way, everything will be done in one efficient step.  


Here ishow to put together the pipeline:  Let's say that your the desired
output slice has the following extent
 
reslice->SetOutputExtent(0, nVoxels[X]-1, 0, nVoxels[Y]-1, 0, 0, 0)

with voxel spacing the same as the input spacing

reslice->SetOutputSpacing(voxelSize[X], voxelSize[Y], voxelSize[Z])

and the origin set so that coordinate (0.0, 0.0, 0.0) lies at tha center
of the slice

reslice->SetOutputOrigin(-0.5*voxelSize[X]*(nVoxels[X]-1),
                         -0.5*voxelSize[Y]*(nVoxels[Y]-1),
                         0.0)

Now, you have to create a transformation that will place the center of
the slice at the desired position & orientation within the input volume.
As an example, this will place the slice at the center of the volume
and at a 30 degree oblique angle:


transform = vtkTransform::New();

transform.PostMultiply()
transform.RotateY(30.0)
transform.Translate(0.5*voxelSize[X]*(nVoxels[X]-1),
                    0.5*voxelSize[Y]*(nVoxels[Y]-1),
                    0.5*voxelSize[Z]*(nVoxels[Z]-1))

reslice.SetResliceTransform(transform)


That should be all that is required.  Just a single transformation,
and a single vtkImageReslice.

 - David



On Tue, 17 Oct 2000, Audrius Stundzia wrote:

> Hi fellow vtker's,
> 
> I would appreciate any insight on using vtkImageReslice to extract 2D image 
> slices from a 3D.
> 
> The catch (for me, anyways) is that to get the slices I want I first have to 
> perform a rotation R(1), followed by a translation T and a second rotation 
> R(2), ie., R(2) T R(1)
> 
> I've tested the rotation + translation tranform by extracting a surface and 
> comparing it center and orientation to a reference grid and it checks out 
> okay.
> 
> Step 1: Here I create and fill my initial vtkImageData object with origin at 
> (0,0,0)
> 
> ( nVoxels[X] = nVoxels[Y] and nVoxels[Z] < nVoxels[X] and
> 
>   voxelSize[X] = voxelSize[Y] and voxelSize[Z] > voxelSize[X] )
> 
>    imageDataVolume = vtkImageData::New();
>    imageDataVolume->PrepareForNewData();
>    imageDataVolume->SetDimensions( nVoxels[X]
>                                  , nVoxels[Y]
>                                  , nVoxels[Z] );
>    imageDataVolume->SetOrigin( -firstCenter[X]
>                              , -firstCenter[Y]
>                              , -firstCenter[Z] );
>    imageDataVolume->SetSpacing( voxelSize[X]
>                               , voxelSize[Y]
>                               , voxelSize[Z] );
>    imageDataVolume->SetExtent( 0
>                              , nVoxels[X] - 1
>                              , 0
>                              , nVoxels[Y] - 1
>                              , 0
>                              , nVoxels[Z] - 1 );
> 
>    imageDataVolume->SetScalarType( VTK_FLOAT);
>    imageDataVolume->SetNumberOfScalarComponents( 1);
>    imageDataVolume->AllocateScalars();
>    imageDataVolume->GetPointData()->SetScalars( imageVolumeScalars);
>    imageDataVolume->Update();
> 
> Step 2: First I reslice along the 1st rotation axes
> 
>    imageDataFirstRotationVolumeReslice = vtkImageReslice::New();
>    imageDataFirstRotationVolumeReslice->SetInput( imageDataVolume);
>    imageDataFirstRotationVolumeReslice
>    ->SetResliceAxes( firstRoatationMatrix);
>    imageDataFirstRotationVolumeReslice->WrapOff();
>    imageDataFirstRotationVolumeReslice->MirrorOff();
>         imageDataFirstRotationVolumeReslice
>    ->SetInterpolationModeToLinear();
>    imageDataFirstRotationVolumeReslice->InterpolateOn();
>    imageDataFirstRotationVolumeReslice->OptimizationOn();
>    imageDataFirstRotationVolumeReslice->ReleaseDataFlagOn();
>    imageDataFirstRotationVolumeReslice->SetBackgroundColor( 0.0
>                                                           , 0.0
>                                                           , 0.0
>                                                           , 0.0 );
> 
>    imageDataFirstRotationVolumeReslice
>    ->OutputAlwaysCenteredOnInputOff();
> 
>    imageDataFirstRotationVolumeReslice
>    ->SetOutputSpacing( voxelSize[X]
>                      , voxelSize[Y]
>                      , voxelSize[Z] );
>                                                                 
> imageDataFirstRotationVolumeReslice
>    ->SetOutputOrigin( -firstCenter[X]
>                     , -firstCenter[Y]
>                     , -firstCenter[Z] );
> 
>    imageDataFirstRotationVolumeReslice
>    ->SetOutputExtent( 0
>                     , ( nVoxels[X] - 1)
>                     , 0
>                     , ( nVoxels[Y] - 1)
>                     , 0
>                     , ( nVoxels[Z] - 1) );
> 
>    imageDataFirstRotationVolumeReslice->Update();
> 
> Question: as the SetOutputOrigin arguements are specified as (float) am I 
> correct in assuming that one uses physical coordinate (x,y,z) rather than 
> voxel coordinates (i,j,k)?
> 
> Step 3: Now I reslice along the 2nd rotation axes, taking the translation T 
> into account by changing the output origin function call SetOutputOrigin.
> 
> Questions: Do I need to first update the origin of the input for the second 
> reslice? Ie.,
> 
> imageDataFirstRotationVolumeReslice->GetOutput()
> ->SetOrigin( -secondCenter[X]
>            , -secondCenter[Y]
>            , -secondCenter[Z] );
> 
> to take the translation T into proper account?
> 
>    imageDataSecondRotationVolumeReslice = vtkImageReslice::New();
>    imageDataSecondRotationVolumeReslice
>    ->SetInput( imageDataFirstRotationVolumeReslice->GetOutput() );
>    imageDataSecondRotationVolumeReslice
>    ->SetResliceAxes( secondRotationMatrix);
> 
>    imageDataSecondRotationVolumeReslice
>    -> same set of function calls as in the first reslice;
> 
>    imageDataSecondRotationVolumeReslice
>    ->OutputAlwaysCenteredOnInputOff();
> 
>    imageDataSecondRotationVolumeReslice
>    ->SetOutputSpacing( voxelSize[X]
>                      , voxelSize[Y]
>                      , voxelSize[Z] );
> 
>    imageDataSecondRotationVolumeReslice
>    ->SetOutputOrigin( -secondCenter[X]
>                     , -secondCenter[Y]
>                     , -secondCenter[Z] );
> 
>    imageDataSecondRotationVolumeReslice
>    ->SetOutputExtent( 0
>                     , ( nVoxels[X] - 1)
>                     , 0
>                     , ( nVoxels[Y] - 1)
>                     , 0
>                     , ( nVoxels[Z] - 1) );
> 
>    imageDataSecondRotationVolumeReslice->Update();
> 
> Step 4: At this stage my reslicing should be done and I can go ahead and 
> extract my 2D image slices along the axes defined by an identity matrix. The 
> relevant parts
> 
>    imageXYPlaneReslice[iPlane]
>    ->SetInput( imageDataSecondRotationVolumeReslice->GetOutput() );
>    imageXYPlaneReslice[iPlane]->SetResliceAxes( identityMatrix);
> 
>    int zIndex
>    = (int) ( ( secondCentre[Z] + GetPlaneZ()[iPlane] )
>              / voxelSize[Z] );
> 
>    imageXYPlaneReslice[iPlane]->SetOutputExtent( 0
>                                                , ( nVoxels[X] - 1)
>                                                , 0
>                                                , ( nVoxels[Y] - 1)
>                                                , zIndex
>                                                , zIndex );
> 
> Three problems:
> 
> 1. Step 4 is slooooow. I assume this is what is meant by the  comment in the 
> online manual, "Warning: This filter is very inefficient if the output X 
> dimension is 1."
> 
> However, my extract plane voxel thickness needs to be 1 as I then need to 
> texture map the output.
> 
> Should I use vtkExtractVOI instead? Any yabuts?
> 
> 2. The translation was not done properly. It's off in all three axes.
> 
> 3. The corresponding XZ and YZ plane images look like they've been    left 
> out in the rain and have had their colours run along the Z axis.
> 
> Any insight would be most appreciated. Thanks.
> 
> Audrius