Difference between revisions of "Documentation/Nightly/Developers/Tutorials/MigrationGuide/Slicer"

From Slicer Wiki
Jump to: navigation, search
Tag: 2017 source edit
Tag: 2017 source edit
Line 119: Line 119:
  
 
Model hierarchies no longer exist in Slicer5, but instead various kinds of hierarchies are now replaced by "subject hierarchy", which can accommodate any node types in a single hierarchy. Accordingly, `ExportSegmentsToModelHierarchy`, `ExportAllSegmentsToModelHierarchy`, etc. are replaced by `ExportSegmentsToModels`, `ExportAllSegmentsToModels`, which take a subject hierarchy folder item ID as input.
 
Model hierarchies no longer exist in Slicer5, but instead various kinds of hierarchies are now replaced by "subject hierarchy", which can accommodate any node types in a single hierarchy. Accordingly, `ExportSegmentsToModelHierarchy`, `ExportAllSegmentsToModelHierarchy`, etc. are replaced by `ExportSegmentsToModels`, `ExportAllSegmentsToModels`, which take a subject hierarchy folder item ID as input.
 
+
Documentation/Nightly
<pre>
+
See code example in [[Documentation/{{documentation/version}}/ScriptRepository#Export_model_nodes_from_segmentation_node|script repository]].
segmentationNode = getNode("Segmentation")
 
shNode = slicer.mrmlScene.GetSubjectHierarchyNode()
 
exportFolderItemId = shNode.CreateFolderItem(shNode.GetSceneItemID(), "Segments")
 
slicer.modules.segmentations.logic().ExportAllSegmentsToModels(segmentationNode, exportFolderItemId)
 
</pre>
 
  
 
=== Slicer 4.11: teem python module renamed to vtkTeem, explicit import required ===
 
=== Slicer 4.11: teem python module renamed to vtkTeem, explicit import required ===

Revision as of 13:39, 13 February 2020

Home < Documentation < Nightly < Developers < Tutorials < MigrationGuide < Slicer

Contents

Slicer backward incompatible changes

Slicer 5.0: API changes since 4.10

  • Removed protected method vtkMRMLModelDisplayableManager::FindPickedDisplayNodeFromMesh

Python 2 to Python 3

Slicer core has been updated to only support Python 3.

C++ classes and python scripts have been updated to use idioms and constructs only available in Python 3.

Update to python scripts have been done leveraging the CLI provided by https://python-future.org by (1) iteratively applying each one of the associates "fixes", (2) reviewing associated changes and (3) updating as needed.

Updates specific to extensions are discussed in Documentation/Nightly/Developers/Tutorials/MigrationGuide#Slicer_5.0:_Python2_to_Python3

Interactor styles

Limitations of VTK widgets (editable points, lines, curves, etc.) prevented Slicer from having sophisticated user interaction in slice and 3D views. In Slicer5, we replaced VTK widgets with MRML widgets. These widgets are still VTK-based and somewhat similar to VTK widgets, but they operate directly on MRML nodes, they use direct method calls between widgets and their representation, and they use a more efficient and flexible event processing. Instead of hardcoding how viewers behave in response to interaction (mouse move, button click, keyboard, ...) events in an interactor style, all these events are translated to actions and performed in a MRML widget. Most modules are not expected to observe interactor events or styles directly, but if they did, then they may need to be updated accordingly.

  • vtkSliceViewInteractorStyle renamed to vtkMRMLSliceDViewInteractorStyle to reflect that it uses MRML classes directly.
  • vtkThreeDViewInteractorStyle renamed to vtkMRMLThreeDViewInteractorStyle to reflect that it uses MRML classes directly.

slicer.util functions

  • slicer.util.loadVolume (and other node load functions) now return the loaded node instead of a True/False flag. In case of an error, a RuntimeError exception is thrown.
    • Old way of loading a node and get it in a variable: volumeNode = slicer.util.loadVolume('path/to/volume.nrrd', returnNode=True)[1]
    • New way of loading a node and get it in a variable: volumeNode = slicer.util.loadVolume('path/to/volume.nrrd')

Markups

  • vtkCommand::Modified events are no longer invoked when control points are added/removed/modified to improve performance. Modules that need to know If a point position is modified need to add observers to vtkMRMLMarkupsNode::PointAddedEvent, vtkMRMLMarkupsNode::PointRemovedEvent, vtkMRMLMarkupsNode::PointModifiedEvent events. See example in script repository.
  • vtkMRMLMarkupsNode::MarkupAddedEvent is renamed to vtkMRMLMarkupsNode::PointAddedEvent. This is called even when preview point is created. Use vtkMRMLMarkupsNode::PointPositionDefinedEvent to get notification only when position of a point becomes defined (and not during preview).
  • vtkMRMLMarkupsNode::MarkupRemovedEvent is renamed to vtkMRMLMarkupsNode::PointRemovedEvent. This is called even when preview point is removed. Use vtkMRMLMarkupsNode::PointPositionUndefinedEvent to get notification only when position of a point becomes undefined (and not during preview).
  • vtkMRMLMarkupsNode::NthMarkupModifiedEvent is replaced by vtkMRMLMarkupsNode::PointModifiedEvent
  • During placement of markups, a preview markup point is created. If number of already placed markup points needs to be determined then GetNumberOfDefinedControlPoints() method can be used.
  • GetDefaultMarkups...() and SetDefaultMarkups...() methods are removed. Instead default display node can be accessed by GetDefaultMarkupsDisplayNode() method and default values can be get/set in that class.
  • vtkMRMLMarkupsNode::GetNthMarkupSelected() is replaced by GetNthControlPointSelected()
  • vtkMRMLMarkupsNode::PointPositionDefinedEvent event is added. This event is invoked whenever position is defined for a new point.
  • vtkMRMLMarkupsNode::PointPositionUndefinedEvent event is added. This event is invoked whenever point with defined position is removed (point is deleted or its position gets undefined).
  • For more details, see vtkMRMLMarkupsNode

Segmentations

Binary labelmap segmentations can now be represented as shared labelmaps. The previous implementation of binary labelmaps was performance intensive as each labelmap was represented using a separate vtkDataObject. Visualizing and editing segmentations that contained a large number of segments could cause performance issues, due to the large number of vtkActors required, as well as calculating masks and overwriting other segments when editing.

By default, newly created segments will now be contained on the same layer. Segments will only be separated into multiple layers if the user creates an overlapping segment when editing.

Segments are now saved as a 4D volume with shared 3D layers. For a segmentation that only uses one layer, the resulting image is a 3D volume. Before saving, the labelmaps will be collapsed into as few layers as possible.

  • seg.nrrd files now contain two additional attributes for each segment: SegmentX_LabelValue and SegmentX_Layer
  • The label value of a segment can be found using vtkSegment::GetLabelValue()
  • Whether or not a segment is shared can be found using vtkSegmentation::IsSharedBinaryLabelmap()
  • The other segments sharing the same labelmap can be found using vtkSegmentation::GetSegmentIDsSharingBinaryLabelmapRepresentation()
  • Segment editor effects should generally use modifySelectedSegmentByLabelmap rather than SetBinaryLabelmapToSegment to manage layer separation
  • Conversion rules now call PreConvert() and PostConvert() before and after conversion to perform pre and post processing steps on the segmentation as a whole
  • The function signature for vtkSegmentationConverterRule::Convert now accepts a vtkSegment rather than two vtkDataObjects
  • slicer.util.arrayFromSegment has been deprecated. slicer.util.arrayFromSegmentBinaryLabelmap and slicer.util.arrayFromSegmentInternalBinaryLabelmap can be used instead
Erase the contents of a single segment
segmentation = segmentationNode.GetSegmentation()
segmentation.ClearSegment(segmentId)


Move a segment from a shared labelmap to a separate layer
segmentation = segmentationNode.GetSegmentation()
segmentation.SeparateSegmentLabelmap(segmentId)
Combine all binary labelmaps to as few layers as possible
segmentation = segmentationNode.GetSegmentation()
segmentation.CollapseBinaryLabelmaps(forceToSingleLayer=false)

Get a read-only labelmap for a single segment:

labelmap = slicer.vtkOrientedImageData()
segmentationNode.GetBinaryLabelmapRepresentation(segmentId, labelmap)

(similarly, use GetClosedSurfaceRepresentation with an additional parameter to get a read-only surface mesh)

or

labelmapNumpyArray = slicer.util.arrayFromSegmentBinaryLabelmap(segmentationNode, segmentId)
Get a modifiable shared labelmap
labelmap = slicer.vtkOrientedImageData()
segmentationNode.GetBinaryLabelmapInternalRepresentation(segmentId, labelmap)

(similarly, use GetClosedSurfaceInternalRepresentation to get a modifiable surface mesh)

or

labelmapNumpyArray = slicer.util.arrayFromSegmentInternalBinaryLabelmap(segmentationNode, segmentId)
Export segments to models

Model hierarchies no longer exist in Slicer5, but instead various kinds of hierarchies are now replaced by "subject hierarchy", which can accommodate any node types in a single hierarchy. Accordingly, `ExportSegmentsToModelHierarchy`, `ExportAllSegmentsToModelHierarchy`, etc. are replaced by `ExportSegmentsToModels`, `ExportAllSegmentsToModels`, which take a subject hierarchy folder item ID as input. Documentation/Nightly See code example in script repository.

Slicer 4.11: teem python module renamed to vtkTeem, explicit import required

  • Since the module provides VTK classes interfacing with "teem", the name is now representative of the class it contains.
  • vtkTeem classes are expected to be used by explicitly importing the module.

Replace code like this:

import teem

class CalculateTensorScalars(object):
  def __init__(self):
    self.dti_math = teem.vtkDiffusionTensorMathematics()

By this:

import vtkTeem

class CalculateTensorScalars(object):
  def __init__(self):
    self.dti_math = vtkTeem.vtkDiffusionTensorMathematics()

Slicer 4.11: Display window/level (brightness/contrast) adjustment

  • A new "Window/level" mouse interaction mode was introduced. Volume display window/level can only be changed if this mode is activated by clicking the corresponding button in the toolbar. The new mouse mode prevents accidental modification of volume window/level (when for example the user accidentally clicked too far from a markup) and it also allows more sophisticated window/level adjustments.
  • New region-based auto window/level feature added: activate "Window/level" mouse mode and use Ctrl + left-click-and-drag to highlight a region and optimize window/level for that (pressing Escape or right-click cancels the operation).
  • Auto window/level reset: activate "Window/level" mouse mode and double-click the left mouse button.
  • Improved auto window/level algorithm to prevent too bright display of images. Window/level is set to display values between 0.1th and 99.9th percentile of gray levels. See details here: https://discourse.slicer.org/t/feedback-requested-how-to-improve-mouse-interaction-in-views/6420.
  • Removed class vtkImageBimodalAnalysis

Slicer 4.10: Registration of runTest function done in ScriptedLoadableModule base class

Following r27617:

  • the ScriptedLoadableModule class takes care of registering the runTest function.
  • the runTest function expects msec keyword argument.

Error message similar to:

Traceback (most recent call last):
  File "/path/to/Slicer-SuperBuild/Slicer-build/bin/Python/slicer/ScriptedLoadableModule.py", line 205, in onReloadAndTest
    test(msec=int(slicer.app.userSettings().value("Developer/SelfTestDisplayMessageDelay")), **kwargs)
TypeError: runTest() got an unexpected keyword argument 'msec'
Reload and Test: Exception!

runTest() got an unexpected keyword argument 'msec'

Replace code like this:

class sceneImport2428(ScriptedLoadableModule):
  [...]
  def __init__(self, parent):
    ScriptedLoadableModule.__init__(self, parent)
    parent.title = "..."
    [...]
    parent.acknowledgementText = "..."
    self.parent = parent 	 
	  	 
    # Add this test to the SelfTest module's list for discovery when the module 	 
    # is created.  Since this module may be discovered before SelfTests itself, 	 
    # create the list if it doesn't already exist. 	 
    try: 	 
      slicer.selfTests 	 
    except AttributeError: 	 
      slicer.selfTests = {} 	 
    slicer.selfTests['sceneImport2428'] = self.runTest 	 
 
  def runTest(self): 	 
    tester = sceneImport2428Test() 	 
    tester.runTest()

  [...]

By this:

class sceneImport2428(ScriptedLoadableModule):
  [...]
  def __init__(self, parent):
    ScriptedLoadableModule.__init__(self, parent)
    parent.title = "..."
    [...]
    parent.acknowledgementText = "..."

  [...]

Slicer 4.9: Update of VTK version from 9.0 to 8.2

Following kitware/VTK@b703d78be, VTK has updated to use version number 8.2 instead of 9.0. This was discussed in on the VTK mailing list in http://vtk.1045678.n5.nabble.com/Discussion-OK-to-change-VTK-s-version-number-from-9-0-to-8-2-tt5748702.html

At first, this VTK commit and its companion kitware/VTK@8a00b357e were both reverted from the Slicer/VTK fork. Then, since having the corresponding changes reverted in VTK was not possible, it was decided to also update Slicer. This was done in the following commits:

  • r27472: COMP: Update c++ classes to support building against VTK >= 9 and VTK >= 8.2
  • r27473: COMP: Update VTK to include version change from 9.0 to 8.2. Fixes #4623

This means that code depending on VTK must also be updated to include similar fixes.

Replace this:

 #if VTK_MAJOR_VERSION >= 9

By this:

 #if VTK_MAJOR_VERSION >= 9 || (VTK_MAJOR_VERSION >= 8 && VTK_MINOR_VERSION >= 2)

and

Replace this:

 #if VTK_MAJOR_VERSION < 9

By this:

 #if VTK_MAJOR_VERSION <= 7 || (VTK_MAJOR_VERSION <= 8 && VTK_MINOR_VERSION <= 1)


Slicer 4.9: ITK_LEGACY_REMOVE is now OFF

In preparation to switch to ITK 5.0, we disable legacy functionality in ITK. This might affect some modules which rely on ITK. Take a look at ITK 4 migration guide before ITK 5 migration guide.

Slicer 4.9: vtkMRMLPlotDataNode renamed to vtkMRMLPlotSeriesNode

Plotting was improved in this commit

Replace this:

 vtkMRMLPlotDataNode

By this:

 vtkMRMLPlotSeriesNode

Slicer 4.9: CMake: Module MIDAS not available

The test infrastructure of your project should be updated to use ExternalData built-in CMake module instead of the specific MIDAS module.

See EMSegment commit r17150 for an example of transition.

This means that instead of using midas_add_test with the MIDAS{path/to/file.ext.md5} syntax for addressing the test data, the function ExternalData_add_target is used by specifying both DATA{path/to/file.ext} and a download target name.

Replace this:

 midas_add_test(NAME test1 COMMAND ...)
 midas_add_test(NAME test2 COMMAND ...)

By this:

 ExternalData_add_test(EMSegmentData NAME test1 COMMAND ...)
 ExternalData_add_test(EMSegmentData NAME test2 COMMAND ...)
 
 [...]
 
 ExternalData_add_target(EMSegmentData)


A key difference with the former approaches is that instead of adding two tests (one named <testName>_fetchData to downoad the data and one running the test command), only one test is added but a common download target is added at the end using ExternalData_add_target function.

This means that test data can now be downloaded in parallel (and cached) at build time instead of testing time.

Slicer 4.9: CMake: Module SlicerMacroCheckExternalProjectDependency not available

Since the module was removed in r26992, consider updating your build system to use CMake module ExternalProjectDependency

Slicer 4.9: CMake: Module SlicerMacroEmptyExternalProject not available

Since the module was removed in r26991

Replace this:

include(SlicerMacroEmptyExternalProject)

[...]

SlicerMacroEmptyExternalProject("${proj}" "${${proj}_DEPENDENCIES}")

By this:

include(ExternalProjectDependency)

[...]

ExternalProject_Add_Empty(${proj} DEPENDS ${${proj}_DEPENDENCIES})

Slicer 4.9: CMake: Module SlicerBlockSetCMakeOSXVariables not available

Since it was renamed to SlicerInitializeOSXVariables in r26982

Replace this:

include(SlicerBlockSetCMakeOSXVariables)

By this:

include(SlicerInitializeOSXVariables)

Slicer 4.9: Application: isRelease() function not available

See #Slicer_4.8:_Application:_isRelease.28.29_function_not_available_or_deprecated

Slicer 4.9: slicer.util.getNode() raises exception if node not found

If slicer.util.getNode() is called and the node is not found then instead of just returning None (Slicer 4.8 behavior), the method now raises a MRMLNodeNotFoundException. This makes code debugging easier (the error is reported when it happens), and in general more consistent with Python conventions.

How to update existing code:

It is advisable to only use slicer.util.getNode in tests, or interactively in the Python console, as its behavior is somewhat unpredictable (it may either found a node by name or ID, and result of wildcard search is even less deterministic). In general, it is recommended to use the MRML scene's GetFirstNodeByName and GetNodeByID methods instead.

Replace this:

n = slicer.util.getNode(nodeNameOrID)

By one of these:

If node is to be found by name:

  n = slicer.mrmlScene.GetFirstNodeByName(nodeName)

If node is to be found by ID:

  n = slicer.mrmlScene.GetNodeByID(nodeID)

If node is to be found by name or ID (slower, less predictable, recommended for testing only):

try:
  n = slicer.util.getNode(nodeNameOrID)
except slicer.util.MRMLNodeNotFoundException:
  n = None

More information: https://github.com/Slicer/Slicer/commit/b63484af1b1b413f35396f8f7efb73e870448bd4

Slicer 4.8: Application: isRelease() function not available or deprecated

Error message similar to:

   Missing/deprecated qSlicerCoreApplication::isRelease()

or

   Missing/deprecated slicer.app.isRelease()


Solution:

Use qSlicerCoreApplication::releaseType() == "Stable"

Summary:

Prior to r26420, the variable Slicer_VERSION_TWEAK was used to check if a "stable release" was built. The variable value was set by updating the sources and defining the variable to an integer greater or equal to 0. In other word, if the variable evaluated to an empty string, a nighty or experimental build was being done, if it evaluated to an integer, a stable release build was being done.

The approach had few issues:

  • the name of the variable was confusing
  • identifying a "stable release" only from a source tree revision was not enough. Indeed the environment defining a "release" is the one found on the build machines used to generate the installer.
  • nightly build are also considered as release

To address this, the CMake variable Slicer_RELEASE_TYPE was introduced. As of 2017-10-04, it can be set to Experimental, Nightly or Stable with Experimental being the value hard-coded in the source.

Identifying a build as "stable" is now explicitly done by setting Slicer_RELEASE_TYPE to Stable at configure time.

Also, since the concept of release types was introduced, the function isRelease() has been removed in favor of releaseType().

References:

https://github.com/Slicer/Slicer/pull/354

Slicer Python Module: modulewidget and others removed.

Summary Python classes formerly in "slicer.moduledm", "slicer.modulelogic", "slicer.modulemrml" and "slicer.modulewidget" are now directly available in the slicer module.

See example of change here.

Rational:

See comments in commit messages referenced blow.

References:

https://github.com/Slicer/Slicer/commit/628f83fe7a6f4e0710e306bcaf7c04b9e3e5e6bd

https://github.com/Slicer/Slicer/commit/9cb5668fde1abc8f0430a91ca37fc29277ceeb4e

MRML: Slicer 4.6: Moved up vtkMRMLStorableNode in the MRML node hierarchy.

Rational:

vtkMRMLStorableNode is not a children of vtkMRMLTransformable node anymore, but directly a children of vtkMRMLNode.

This allows making a node storable without requiring it to be also transformable. It is important for several node types (color maps, tables, etc), which require separate storage node but are not transformable.

References:

Error message similar to:

   /tmp/LongitudinalPETCT/MRML/vtkMRMLLongitudinalPETCTStudyNode.cxx: In member function ‘void vtkMRMLLongitudinalPETCTStudyNode::ObserveRegistrationTransform(bool)’:
   /tmp/LongitudinalPETCT/MRML/vtkMRMLLongitudinalPETCTStudyNode.cxx:478:28: error: ‘class vtkMRMLVolumePropertyNode’ has no member named ‘GetParentTransformNode’
                  && propNode->GetParentTransformNode()
                               ^
   /tmp/LongitudinalPETCT/MRML/vtkMRMLLongitudinalPETCTStudyNode.cxx:480:23: error: ‘class vtkMRMLVolumePropertyNode’ has no member named ‘SetAndObserveTransformNodeID’
                propNode->SetAndObserveTransformNodeID(
                          ^
   /tmp/LongitudinalPETCT/MRML/vtkMRMLLongitudinalPETCTStudyNode.cxx:503:23: error: ‘class vtkMRMLVolumePropertyNode’ has no member named ‘SetAndObserveTransformNodeID’
                propNode->SetAndObserveTransformNodeID(NULL);
                          ^

Solution:

Removes lines and/or refactor code


MRML: Slicer 4.5: Introduction of vtkMRMLLabelMapVolumeNode

Rational:

Before vtkMRMLScalarVolumeNode was used for both scalar and label map volumes and the LabelMap custom MRML node attribute was used for distinguishing between them (0=scalar; 1=label map volume).

This made conversion between labelmap/scalar volumes very easy but made it difficult to customize behavior, display, processing of segmentation information.

Now a new vtkMRMLLabelMapVolumeNode class is used for storing segmentation information (still using vtkMRMLScalarVolume used as base class for backward compatibility; but in the future the base class may be changed to reflect that segmentation can be represented in various ways, not just as volumes).

Error message similar to:

 error: ‘class vtkMRMLScalarVolumeNode’ has no member named ‘SetLabelMap’
    outputVolumeNode->SetLabelMap(1);
                      ^

Solution (part1: down cast to vtkMRMLLabelMapVolumeNode, remove call to SetLabelMap)

Replace lines like:

    vtkMRMLNode* outputNode = d->OutputLabelVolumeMRMLNodeComboBox->currentNode();
    vtkMRMLScalarVolumeNode* outputVolumeNode = vtkMRMLScalarVolumeNode::SafeDownCast(outputNode);
    [...]
    outputVolumeNode->SetLabelMap(1);


with:

    vtkMRMLLabelMapVolumeNode* outputVolumeNode =
      vtkMRMLLabelMapVolumeNode::SafeDownCast(d->OutputLabelVolumeMRMLNodeComboBox->currentNode());
    [...]


Solution (part2: Update UI file):

Replace lines like:

 <widget class="qMRMLNodeComboBox" name="InputLabelVolumeMRMLNodeComboBox">
  <property name="nodeTypes">
   <stringlist>
    <string>vtkMRMLScalarVolumeNode</string>
   </stringlist>
  </property>
  [...]
 </widget>

with:

 <widget class="qMRMLNodeComboBox" name="InputLabelVolumeMRMLNodeComboBox">
  <property name="nodeTypes">
   <stringlist>
    <string>vtkMRMLLabelMapVolumeNode</string>      <------------- Update Here
   </stringlist>
  </property>
  [...]
 </widget>


Solution (part3: Update node selector configuration):

Replace lines like:

 nodeSelector.setNodeTypes(QStringList("vtkMRMLScalarVolumeNode"));
 nodeSelector.addAttribute("vtkMRMLScalarVolumeNode", "LabelMap", "1");

with:

 nodeSelector.setNodeTypes(QStringList("vtkMRMLLabelMapVolumeNode"));


References:


CLI: Slicer 4.3: Add ITKFactoryRegistration library centralizing ITK IO factory registration

Rational:

 Linking against ITKFactoryRegistration ensures that ITK IO factory are properly registered on all supported platforms.

Error message similar to:

 Undefined symbols for architecture x86_64:
 "itk::itkFactoryRegistration()", referenced from:
 _main in ImageMakerTest.cxx.o
 ld: symbol(s) not found for architecture x86_64

Solution:

Replace lines like:

 target_link_libraries(${CLP}Test ${CLP}Lib)

with:

 target_link_libraries(${CLP}Test ${CLP}Lib ${SlicerExecutionModel_EXTRA_EXECUTABLE_TARGET_LIBRARIES})

References: