Difference between revisions of "Documentation/Nightly/Modules/Transforms"
m (Text replacement - "\[http:\/\/www\.slicer\.org\/slicerWiki\/index\.php\/([^ ]+) ([^]]+)]" to "$2")
|Line 55:||Line 55:|
<!-- ---------------------------- -->
<!-- ---------------------------- -->
* Please use tutorial about [
* Please use tutorial about [Documentation/4.0/Trainingloading and viewing data].
<!-- ---------------------------- -->
<!-- ---------------------------- -->
Revision as of 02:36, 27 November 2019Home < Documentation < Nightly < Modules < Transforms
For the stable Slicer documentation, visit the 4.10 page.
Introduction and Acknowledgements
This module is used for creating and editing transformation matrices.
Transformation nodes are used in Slicer to define spacial relationships between different nodes (such as volumes, models, fiducials, ROI's, or other Transform nodes) or between the nodes and the global RAS space.
You can establish these relations by moving nodes from the Transformable list to the Transformed list or by dragging the nodes under the Transformation nodes in the Data module.
- Short video demonstrating the main features: http://screencast.com/t/Z6dQVjK3m
- Support non-linear transforms in the Transforms module: allow Apply, Harden, Invert transform
- Transform information displayed in the Transforms module (type of transform, basic properties)
- Transform visualization:
- Built into the Transforms module
- Three main modes: Glyphs (show an array of arrows, cones, spheres), Grid (show a deformed grid), or Contour (show isolines/isosurfaces for specified displacement magnitude values)
- All transform types are supported (chains of transforms as well)
- Visualization in the slice viewers
- Visualization in the 3D viewers, in the specified region (region can be a slice viewer, a volume, or a ROI widget)
- Real-time update: if the transform (or any visualization parameter) is changed then the visualization is updated immediately (interactive visualization while editing the transform)
- Built-in colormap editor
- MetaImage (mha), NIFTI (nii) vector volumes can be loaded as displacement field (grid) transform
- Saving and loading to/from file: transforms are saved using ITK file readers/writers. However, ITK does not know how to interpret inverted transforms. Therefore if an inverted transform is saved to file then standard ITK applications may not be able to use the transform file. A workaround is to export the transform to a displacement field and save that to file.
Most frequently Transform module is used for these scenarios:
- Manual Registration: You can insert a transformation node into your scene, and in the Data module drag a volume or a model under it, making them children of the transformation node. After that any changes to the transformation matrix of this node will be applied to the display of children volumes and models.
- Visualize the displacement that transforms specify: Transforms can be visualized in both 2D and 3D views, as glyphs representing the displacement vectors as arrows, cones, or spheres; regular grids that are deformed by the transform; or contours that represent lines or surfaces where the displacement magnitude has a specific value.
- Apply transforms: You can dynamically transform a node by selecting them in the Transformable list and clicking the right arrow button. Whenever the transform changes, the transformed nodes are updated accordingly. The Harden Transform (its button is below the left arrow button) can be used for applying the transform to nodes permanently. Transforms themselves can be transformed, therefore chain of transforms can be constructed. Non-linear transforms can be concatenated, too. If non-linear transform is hardened on a volume then the volume is resampled using the same spacing and axis directions as the original volume (using linear interpolation). Extents are updated to fully contain the transformed volume. If image resampling modules are used then output image geometry and interpolator can be chosen.
- Please use tutorial about loading and viewing data.
Panels and their use
Transform editing and application
- Active Transform: Select the transform node to display, control and edit.
- Display - Edit:
- Transform Matrix: 4x4 matrix. Each element is editable on double click. Type Enter to validate change, Escape to cancel or Tab to edit the next element.
- Translation: Apply LR, PA, and IS translational components of the transformation matrix in the RAS space (in mm). Min and Max control the lower and upper bounds of the sliders.
- Rotation: Apply LR, PA, and IS rotation angles (degrees) in the RAS space. Rotations are concatenated.
- Coordinate Reference: Switches between global RAS space transformation and a local one, relative to the current position and orientation.
- Identity: Resets transformation matrix to identity matrix.
- Invert: Inverts the transformation matrix.
- Transformed nodes: Control what nodes uses the current transformation.
- Transformable: List the nodes in the scene that DON'T use the transform node.
- Transformed: List the nodes in the scene that use the transform node.
- Right arrow: Apply the current transform node to the selected nodes in Transformable list.
- Left arrow: Remove the current transform node from the selected nodes in the Transformed list.
Transform display options
Open Transforms module / Display section / Colors section. If you click on a small circle then above the color bar you can see the small color swatch. On its left side is the points index (an integer that tells which point is being edited and that can be used to jump to the previous/next point), and on its right side is the mm value corresponding to that color.
The default colormap is:
- 1mm (or below) = gray
- 2mm = green
- 5mm = yellow
- 10mm (or above) = red
You can drag-and-drop any of the small circles or modify the mm value in the editbox. You can also add more color values by clicking on the color bar. Then, you can assign a color and/or adjust the mm value. If you click on a circle and press the DEL key then the color value is deleted.
If you need to know accurate displacement values at specific positions then switch to contour mode and in the “Levels” list enter all the mm values that you are interested in. For example, if you enter only a single value “3” in the Levels field you will see a curve going through the points where the displacement is exactly 3 mm; on one side of the curve the displacements are smaller, on the other side the displacements are larger.
You can show both contours and grid or glyph representations by loading the same transform twice and choosing a different representation for each.
Information for Developers
Key MRML nodes
- vtkMRMLTransformableNode: any node that can be transformed
- vtkMRMLTransformNode: it can store any linear or deformable transform or composite of multiple transforms
- vtkMRMLLinearTransformNode: Deprecated. The transform does exactly the same as vtkMRMLTransformNode but has a different class name, which are still used for showing only certain transform types in node selectors. In the future this class will be removed. A vtkMRMLLinearTransformNode may contain non-linear components after a non-linear transform is hardened on it. Therefore, to check linearity of a transform the vtkMRMLTransformNode::IsLinear() and vtkMRMLTransformNode::IsTransformToWorldLinear() and vtkMRMLTransformNode::IsTransformToNodeLinear() methods must be used instead of using vtkMRMLLinearTransformNode::SafeDownCast(transform)!=NULL.
- vtkMRMLBSplineTransformNode: Deprecated. The transform does exactly the same as vtkMRMLTransformNode but has a different class name, which are still used for showing only certain transform types in node selectors. In the future this class will be removed.
- vtkMRMLGridTransformNode: Deprecated. The transform does exactly the same as vtkMRMLTransformNode but has a different class name, which are still used for showing only certain transform types in node selectors. In the future this class will be removed.
How to programmatically apply a transform to a transformable node:
vtkNew<vtkMRMLTransformNode> transformNode; scene->AddNode(transformNode.GetPointer()); ... vtkNew<vtkMatrix4x4> matrix; ... transform->SetMatrixTransformToParent( matrix.GetPointer() ); ... vtkMRMLVolumeNode* transformableNode = ...; // or vtkMRMLModelNode*... transformableNode->SetAndObserveTransformNodeID( transformNode->GetID() );
How to set a transformation matrix from a numpy array:
# Create a 4x4 transformation matrix as numpy array transformNode = ... transformMatrixNP = np.array( [[0.92979,-0.26946,-0.25075,52.64097], [0.03835, 0.74845, -0.66209, -46.12696], [0.36608, 0.60599, 0.70623, -0.48185], [0, 0, 0, 1]]) # Update matrix in transform node transformNode.SetAndObserveMatrixTransformToParent(slicer.util.vtkMatrixFromArray(transformMatrixNP))
Example of moving a volume along a trajectory using a transform
# Load sample volume import SampleData sampleDataLogic = SampleData.SampleDataLogic() mrHead = sampleDataLogic.downloadMRHead() # Create transform and apply to sample volume transformNode = slicer.vtkMRMLTransformNode() slicer.mrmlScene.AddNode(transformNode) mrHead.SetAndObserveTransformNodeID(transformNode.GetID()) # How to move a volume along a trajectory using a transform: import time import math transformMatrix = vtk.vtkMatrix4x4() for xPos in range(-30,30): transformMatrix.SetElement(0,3, xPos) transformMatrix.SetElement(1,3, math.sin(xPos)*10) transformNode.SetMatrixTransformToParent(transformMatrix) slicer.app.processEvents() time.sleep(0.02) # Note: for longer animations use qt.QTimer.singleShot(100, callbackFunction) # instead of a for loop.
Because a transform node is also a transformable node, it is possible to concatenate transforms with each others:
vtkMRMLTransformNode* transformNode = ...; vtkMRMLTransformNode* transformNode2 = ...; transformNode2->SetAndObserveTransformNodeID( transformNode->GetID() ); ... transformable->SetAndObserveTransformNodeID( transformNode2->GetID() );
How to convert the transform to a grid transform (also known as displacement field transform)?
transformNode=slicer.util.getNode('LinearTransform_3') referenceVolumeNode=slicer.util.getNode('MRHead') slicer.modules.transforms.logic().ConvertToGridTransform(transformNode, referenceVolumeNode)
- Conversion to grid transform is useful because some software cannot use inverse transforms or can only use grid transforms.
- Displacement field transforms are saved to file differently than displacement field volumes: displacement vectors in transforms are converted to LPS coordinate system on saving, displacement vectors in volumes are saved to file unchanged.
How to export the displacement magnitude of the transform as a volume?
transformNode=slicer.util.getNode('LinearTransform_3') referenceVolumeNode=slicer.util.getNode('MRHead') slicer.modules.transforms.logic().CreateDisplacementVolumeFromTransform(transformNode, referenceVolumeNode, False)
How to visualize the displacement magnitude as a color volume?
transformNode=slicer.util.getNode('LinearTransform_3') referenceVolumeNode=slicer.util.getNode('MRHead') slicer.modules.transforms.logic().CreateDisplacementVolumeFromTransform(transformNode, referenceVolumeNode, True)
- Slicer stores transforms in VTK classes in memory, but uses ITK transform IO classes to read/write transforms to files. ITK's convention is to use LPS coordinate system as opposed to RAS coordinate system in Slicer (see Coordinate systems page for details). Conversion between VTK and ITK transform classes are implemented in vtkITKTransformConverter.
- ITK stores the transform in resampling (a.k.a., image processing) convention, i.e., that transforms points from fixed to moving coordinate system. This transform is usable as is for resampling a moving image in the coordinate system of a fixed image. For transforming points and surface models to the fixed coordinate system, one needs the transform in the modeling (a.k.a. computer graphics) convention, i.e., transform from moving to fixed coordinate system (which is the inverse of the "image processing" convention).
- Transform nodes in Slicer can store transforms in both modeling (when ToParent transform is set) and resampling way (when FromParent transform is set). When writing transform to ITK files, linear transforms are inverted as needed and written as an AffineTransform. Non-linear transforms cannot be inverted without losing information (in general), therefore if a non-linear transform is defined in resampling convention in Slicer then it is written to ITK file using special "Inverse" transform types (e.g., InverseDisplacementFieldTransform instead of DisplacementFieldTransform). Definition of the inverse classes are available in vtkITKTransformInverse. The inverse classes are only usable for file IO, because currently ITK does not provide a generic inverse transform computation method. Options to manage inverse transforms in applications:
- Create VTK transforms from ITK transforms: VTK transforms can compute their inverse, transform can be changed dynamically, the inverse will be always updated automatically in real-time (this approach is used by Slicer)
- Invert transform in ITK statically: by converting to displacement field and inverting the displacement field; whenever the forward transform changes, the complete inverse transform has to be computed again (which is typically very time consuming)
- Avoid inverse non-linear transforms: make sure that non-linear transforms are only set as FromParent
- Transforms module in Slicer shows linear transform matrix values in RAS coordinate system, according to resampling convention. Therefore to retrieve the same values from an ITK transforms as shown in Slicer GUI, one has switch between RAS/LPS and modeling/resampling:
// Convert from LPS (ITK) to RAS (Slicer) // input: transformVtk_LPS matrix in vtkMatrix4x4 in resampling convention in LPS // output: transformVtk_RAS matrix in vtkMatri4x4 in modeling convention in RAS // Tras = lps2ras * Tlps * ras2lps vtkSmartPointer<vtkMatrix4x4> lps2ras = vtkSmartPointer<vtkMatrix4x4>::New(); lps2ras->SetElement(0,0,-1); lps2ras->SetElement(1,1,-1); vtkMatrix4x4* ras2lps = lps2ras; // lps2ras is diagonal therefore the inverse is identical vtkMatrix4x4::Multiply4x4(lps2ras, transformVtk_LPS, transformVtk_LPS); vtkMatrix4x4::Multiply4x4(transformVtk_LPS, ras2lps, transformVtk_RAS); // Convert the sense of the transform (from ITK resampling to Slicer modeling transform) vtkMatrix4x4::Invert(transformVtk_RAS);
# Copy the content between the following triple-quotes to a file called 'LinearTransform.tfm', and load into Slicer tfm_file = """#Insight Transform File V1.0 #Transform 0 Transform: AffineTransform_double_3_3 Parameters: 0.929794207512361 0.03834792453582355 -0.3660767246906854 -0.2694570325150706 0.7484457003494506 -0.6059884002657121 0.2507501531497781 0.6620864522947292 0.7062335947709847 -46.99999999999999 49 17.00000000000002 FixedParameters: 0 0 0""" import numpy as np # get the upper 3x4 transform matrix m = np.array( tfm_file.splitlines().split()[1:], dtype=np.float64 ) # pad to a 4x4 matrix m2 = np.vstack((m.reshape(4,3).T, [0,0,0,1])) def itktfm_to_slicer(tfm): ras2lps = np.diag([-1, -1, 1, 1]) mt = ras2lps @ m2 @ ras2lps mt[:3,3] = mt[:3,:3] @ mt[:3,3] return mt print( itktfm_to_slicer(m2) ) # Running the code above in Python should print the following output. # This output should match the display the loaded .tfm file in the Transforms module: # [[ 0.92979 -0.26946 -0.25075 52.64097] # [ 0.03835 0.74845 -0.66209 -46.12696] # [ 0.36608 0.60599 0.70623 -0.48185] # [ 0. 0. 0. 1. ]]
When a transform node is observed by a transformable node, vtkMRMLTransformableNode::TransformModifiedEvent is fired on the transformable node at observation time. Anytime a transform is modified, vtkCommand::ModifiedEvent is fired on the transform node and vtkMRMLTransformableNode::TransformModifiedEvent is fired on the transformable node.