3D Slicer Architecture and Implementation You’ve just seen the why and the what, now we’ll discuss the how… Steve Pieper, PhD

Goals NA-MIC Kit: Software and Methodologies for Medical Image Computing Facilitate Research Promote Interoperability Stable, Cross-Platform Run Time Environment Full Set of Core Features Avoid Duplicated Effort Flexible Module Architecture Plug-ins should be As Simple As Possible

Overview MRML, Logic, GUI Core Libraries and Dependencies Some Details of MRML Modules

Slicer3 “Observer MVC” Pattern MRML (Model) For Scene Description and Application State MRML Nodes are Persistent and Undoable Scene and Nodes are Observable Logic Encapsulate VTK and ITK Pipelines (Controller) Observe MRML to Configure Pipelines Help Create/Manage Nodes No UI Components (no Widgets, Actors, Mappers, Renderers or RenderWindows) GUI (View) Observe and Edit MRML Interact with User and Display Hardware Modules Should Follow Same Conventions GUI Renderers Widgets Edit Observe Logic Observe Edit Observe Edit MRML Nodes Motivation for this architecture: modularize! Make the parts resuable. MVC is model/view/controller – often used as an idealized design pattern, but almost always gets customized for any give application “Observe” means generic event mechanisms are used to pass information. “Edit” means code can directly call methods. Example: GUI can call methods in Logic classes, but Logic cannot call GUI methods. MRML cannot call Logic or GUI methods. There can be many observers for any event.

Example: EM in Slicer3 MRML Logic GUI Manage MRML nodes API for access to parameters Manage hierarchy Segmentation algorithm Window to parameter set Wizard Interaction with images Global Parameters Hierarchical parameters Image data Segmentation output

Example: EM Batch Tool MRML Logic CLI Manage MRML nodes API for access to parameters Manage hierarchy Segmentation algorithm Command Line Parsing Load Parameter Description MRML Scenes Global Parameters Hierarchical parameters Image data Segmentation output

Library Dependencies ITK VTK Tcl/Tk Teem KWWidgets vtkTeem vtkITK MRML SlicerBaseLogic SlicerBaseGUI Slicer3

vtkITK, vtkTeem Slicer3 Teem: Multidimensional Raster Image Library (Gordon Kindlmann) Includes NRRD File Format ExtensiveTensor Manipulation Utilities vtkTeem, vtkITK: Create VTK Compatible Filters using ITK and Teem Code for Implementation Includes vtkITKArchetype* Readers and Writers Implemented with ITK I/O Factories SlicerBaseGUI SlicerBaseLogic MRML KWWidgets vtkTeem vtkITK Tcl/Tk VTK Teem ITK

MRML MRML: Medical Reality Markup Language Slicer3 MRML: Medical Reality Markup Language Library Provides Central Data Representation for Slicer3 Application State is Explicitly Described in MRML GUI and Modules: Observe MRML to Learn of Changes to State Manipulate State to Reflect User Interaction and Calculation Results SlicerBaseGUI SlicerBaseLogic MRML KWWidgets vtkTeem vtkITK Tcl/Tk VTK Teem ITK

SlicerBaseLogic Slicer3 Utility Code for Implementing Base Application Functionality Input/Output Volume Reslicing Image Filters No Rendering, and No GUI Code SlicerBaseGUI SlicerBaseLogic MRML KWWidgets vtkTeem vtkITK Tcl/Tk VTK Teem ITK

SlicerBaseGUI KWWidgets Subclasses vtkSlicerApplication Custom KWWidgets Rendering Utility Code VTK 3D Widget Subclasses SlicerBaseGUI SlicerBaseLogic MRML KWWidgets vtkTeem vtkITK Tcl/Tk VTK Teem ITK

Scene Description Provided by D. Gering

MRML Concepts Node Unit of Organization for MRML Group of Variables Representing an Object or Concept State Scene Slicer Application has Single Scene Instance that has Collection of Nodes and API for Access/Create/Delete Undo/Redo & Scene Snapshots Scene Swaps Nodes in to / out of Current State Logic/GUI Auto-Update through Observers Serialization / Deserialization Nodes Responsible for Read/Write of XML Version of State Scene Read/Write Analogous to Snapshots

MRML Concepts (cont) Observers / Events Scene and Nodes Invoke Custom Observable Events for Specific Actions (e.g. NodeAddedEvent) or ModifiedEvents for General Updates Node API Allows Disabling ModifiedEvent Invocation to Allow Groups of Operations with Single ModifiedEvent Node API Includes MRMLObserverManager Class to Simplify Addition/Removal of Multiple Event Observers Scene Provides EventBroker to Support Observer Introspection, Asynchronous Invocation, Logging, and Event Compression

MRML Concepts (cont) IDs & References Each Node is Given a Unique ID for Retrieval in Scene Nodes Can Refer to Other Nodes by ID E.g. TransformableNodes can Refer to a TransformNode by ID E.g. ModelNode can Refer to ModelDisplayNode by ID Scene Maintains ReferencedIDs and ReferencingNodes Lists to Maintain One-to-One Relationship During Scene Import

Node Types Displayable/Storable/Transformable Display/Storage/Transform Hierarchy Parameter Selection Slice SliceComposite Color Fiducial Snapshot Command Line Module Other Module-Specific Nodes

“Observer MVC” Example Event: User Picks Add Volume in Volumes GUI (KWWidget) Volumes GUI calls AddArchetypeVolume in VolumesLogic Volumes Logic Creates VolumeNode and VolumeStorageNode, Reads Data, and Adds to Scene Scene Invokes NodeAddedEvent NodeSelector Widgets Update Menus Volumes GUI Sets Active Volume on SelectionNode Slice Viewers Updated to New Volume GUI Renderers Widgets Edit Observe Logic Observe Edit Observe Edit MRML Nodes Motivation for this architecture: modularize! Make the parts resuable. MVC is model/view/controller – often used as an idealized design pattern, but almost always gets customized for any give application Notes: Event invocation happens synchronously in Slicer 3.2, meaning volume is read and updated before AddArchetypeVolume method returns. Asynchronous event processing may become the default in the future. GUI Currently propagates selection to each slice composite node in the scene. In the future the slice GUIs should be observing the scene and optionally display newly loaded volumes.

Coordinate Systems IJK (Index Coordinates of Volume) vtkMRMLVolumeNode uses vtkImageData to store voxels, with Origin 0,0,0 and Spacing 1,1,1 RAS (Right-Anterior-Superior) vtkMRMLVolumeNode::GetIJKToRASMatrix provides vtkMatrix4x4 that includes origin, spacing, and direction vectors to map to patient space Origin is at the center of the voxel Vectors Point in Specified Direction (i.e. R goes from Left to Right) LPS (Left-Posterior-Superior) Used by ITK code and DICOM Slicer Automatically Converts RAS->LPS

Transforms TransformToParent vtkMRMLTransformableNode Instances Contain TransformNodeID Reference to a vtkMRMLTransformNode Allows Many Nodes Inside Each Transform and Transform Hierarchies TranfrormNode Defines how to Go from Child to Parent Space (e.g. a ModelNode points to a TransfromNode that Scales it to Global RAS Coordinates) Transforms Represented by vtkGeneralTransforms (linear/nonlinear) (TransformFromParent – not yet implemented) Nonlinear Registration Results Often Better Represented as Resampling Transforms, Opposite of Current Modeling Transforms

Modularity Goals Keep the base package “lean and mean” Modules have individual identity Per-module web site, svn, downloads, mailing lists, wiki… Allow users to assemble their own set of tools Customized ‘Bundles’ by task or application Easy to download compatible extensions Analogous to Firefox extensions Integrate extension builds into developer/nightly/release processs NITRC Supplement to NA-MIC helping to pay for needed infrastructure (Neuroimaging Informatics Tools and Resources Clearinghouse) NITRC can host neuroimaging projects (gforge implementation)

Base Features vs. Modules Visualization MRML: Models, Volumes, Volume Rendering, Fiducials, Continuum Meshes, Labeled Data Can create custom MRML Nodes and behavior Filtering None Implemented as Modules using ITK or other Libraries Registration Transform Display and Edit, Save/Restore Calculate Transforms, Resample Data Segmentation Label Maps, Parcellated Surfaces Segmentation Algorithms in ITK or other Libraries Quantification Label, Image, Volume Statistics; Numpy access to MRML Applications in Python or MATLAB Real-time Integration VTK Rendering, KWWidgets framework, Tracker Support (as Transforms) Direct Manipulation of the MRML Scene; 2D/3D Widgets; Device Interfaces, OpenIGTLink Module Diffusion Imaging DWI, DTI, Fiber Bundles Tractography, Clustering, Atlases Applications “Bundles” of Modules in Distribution: Registration, Editor, some Filters, Customized Extensions, Domain-specific code, Optimized Interfaces…

Module Terminology Built-In Modules Few Core Features Linked to Slicer3.cxx Non-Core Modules becoming Loadable Loadable Modules Dynamically Discovered and Loaded Can Access Application’s MRML Scene Instance and Call SlicerBaseGUI and Related APIs (vtkRendering, KWWidgets…) Scripted Modules Like Loadable, but Written in Tcl or Python Command Line Modules Can Run as Stand Alone Executables Provide Standard Command Line Parsing, which Allows Slicer to Generate GUI Automatically Can be Compiled as Shared Library for Dynamic Link to Slicer

Loadable Modules Create vtkSlicerModuleGUI Subclass http://www.slicer.org/slicerWiki/index.php/Slicer3:How_to_implement_an_Interactive_Module_GUI http://www.slicer.org/slicerWiki/index.php/Slicer3:Loadable_Modules:HOWTO Create vtkSlicerModuleGUI Subclass Override Virtual Methods for CreateGUI, ProcessMRMLEvents, etc. Provide Custom vtkMRMLNode Subclass(es) to Represent, Save and Restore state Create Logic Classes to Implement Module Functionality Create CMake Configuration Files to Build and Install Shared Libraries for Runtime Discovery and Packaging

Scripted Modules A Generic ScriptedModule Module Provides Mapping of C++ Method Entry Points to a Corresponding Naming Convention for Scripts Scripts are Dynamically Discovered at Runtime A Generic ScriptedModuleNode Allows Save/Restore of Keyword / Value Pairs to Represent Module State Full API of Slicer, MRML, KWWidgets, VTK, available (except where not wrappable; these APIs have been designed for scriptability) Editor is extensive Tcl-based Module Several Python examples are included in current svn trunk Python modules can also access Numpy routines for Numerical processing (Volumes and other DataArrays are directly mapped to Numpy arrays for easy manipulation)

Command Line Modules Use the GenerateCLP command (provided with Slicer) to convert XML description of program arguments into a C++ .h file Main program of module uses PARSE_ARGS macro provided by .h file to instance variables corresponding to parameters defined in XML file PARSE_ARGS also defines a --xml argument that returns original XML string on stdout Slicer parses the XML to create GUI for each module Are Covered Extensively in the HelloWorld Tutorial Note: Command Line Modules can be written in any language. C++ utilities are provided, but can be emulated / recreated as needed.