More efficient resampling

Development
,
1

Hi all,

I would like to make some function which perform real-time resampling depending on the transform matrix

I used simple itk resample function, however, this one quite slow (e.g, 1~2 s for one interpolation)

Is there more effcieint way to perform the resample ?
Below, this is my cord

 # Create a new transform
    transformNode = slicer.mrmlScene.AddNewNodeByClass("vtkMRMLTransformNode", "MyTransform")
    transNode.SetAndObserveTransformNodeID(transformNode.GetID())

    # Callback function that will be called each time the transform is modified

    def onMyTransformModified(caller, event):
      transformMatrix = vtk.vtkMatrix4x4()
      caller.GetMatrixTransformToWorld(transformMatrix)
      transformMatrix.Invert()
      matrix_arr = []
      translate_arr = []
      for i in range(0, 3):
        for j in range(0, 3):
          matrix_arr.append(transformMatrix.GetElement(i, j))
        translate_arr.append(transformMatrix.GetElement(i, 3))

      rigid_euler = sitk.Euler3DTransform()
      rigid_euler.SetMatrix(matrix_arr)  # rotate 
      rigid_euler.SetTranslation(translate_arr)  # move to 
      interpolator = sitk.sitkCosineWindowedSinc

      default_value = 0
      resampler = sitk.ResampleImageFilter()
      resampler.SetInterpolator(interpolator)
      resampler.SetReferenceImage(reference_image)
      resampler.SetOutputPixelType(sitk.sitkFloat32)
      resampler.SetDefaultPixelValue(default_value)
      resampler.SetTransform(rigid_euler)
      resampler.SetNumberOfThreads(10)
      resample_image = resampler.Execute(excute_image)

      sitkUtils.PushVolumeToSlicer(resample_image, targetNode=transNode)

    # Add an observer that will make the callback function called each time the transform is modified
    observationTag = transformNode.AddObserver(slicer.vtkMRMLTransformableNode.TransformModifiedEvent, onMyTransformModified)
2

A sophisticated filter, such as ITK’s windowed sinc implementation is quite slow, while you can get visually indistinguishable results with a much faster linear or cubic interpolation implemented in VTK. Transferring image between ITK and VTK (if you deep-copy the image and the image is large) and instantiating a filter each time the callback function is executed are also unnecessarily slow down things.

All these problems are solved if you simply use Volume Reslice Driver module (in SlicerIGT extension). It uses a fast interpolator and there is no overhead in recreating filters, unnecessary copies, etc. The resampling usually does not take more than a few ten milliseconds. Your code will be much simpler, too, as you can set up a parameter node and set the input image and transform nodes in a few lines of Python code.

1 Like