Segmentations

Overview

The Segmentations module manages segmentations. Each segmentation can contain multiple segments, which correspond to one structure or ROI. Each segment can contain multiple data representations for the same structure, and the module supports automatic conversion between these representations (the default ones are: planar contour, binary labelmap, closed surface model), as well as advanced display settings and import/export features.

  • Visualization of structures in 2D and 3D views

  • Define regions of interest as input to further analysis (volume measurements, masking for computing radiomics features, etc.)

  • Create surface meshes from images for 3D printing

  • Editing of 3D closed surfaces

Motivation, features, and details of the infrastructure are explained in paper Cs. Pinter, A. Lasso, G. Fichtinger, “Polymorph segmentation representation for medical image computing”, Computer Methods and Programs in Biomedicine, Volume 171, p19-26, 2019 (pdf, DOI) and in presentation slides (pdf, pptx).

Use cases

Edit segmentation

Segmentation can be edited using Segment Editor module.

Import an existing segmentation from volume file

3D volumes in NRRD (.nrrd or .nhdr) and Nifti (.nii or .nii.gz) file formats can be directly loaded as segmentation:

  • Drag-and-drop the volume file to the application window (or use menu: File / Add Data, then select the file)

  • In Description column choose Segmentation

  • Optional: if a color table (specifying name and color for each label value) is available then load that first into the application and then select it as Color node in the Options section. Specification of the color table file format is available here.

  • Click OK

Tip

To avoid the need to always manually select Segmentation, save the .nrrd file using the .seg.nrrd file extension. It makes Slicer load the image as a segmentation by default.

Other image file formats can be loaded as labelmap volume and then converted to segmentation:

  • Drag-and-drop the volume file to the application window (or use menu: File / Add Data, then select the file)

  • Click Show Options

  • In Options column check LabelMap checkbox (to indicate that the volume is a labelmap, not a grayscale image)

  • Click OK

  • Go to Data module, Subject hierarchy tab

  • Right-click on the name of the imported volume and choose Convert labelmap to segmentation node

Tip

To show the segmentation in 3D, go to Segmentations module and click Show 3D. Alternatively, go to Data module and drag-and-drop the segmentation into each view where you want to see them - if the segmentation is dragged into a 3D view then it will be shown there in 3D.

Import an existing segmentation from model (surface mesh) file

3D models in STL and OBJ formats can be directly loaded as segmentation:

  • Drag-and-drop the volume file to the application window (or use menu: File / Add Data, then select the file)

  • In Description column choose Segmentation

  • Click OK

If the model contains very thin and delicate structures then default resolution for binary labelmap representation may be not sufficient for editing. Default resolution is computed so that the labelmap contains a total of approximately 256x256x256 voxels. To make the resolution finer:

  • Go to Segmentations module

  • In Representations section, click Binary labelmap -> Create, then Update

  • In the displayed popup:

    • In Conversion path section, click Closed surface -> Binary labelmap

    • In Conversion parameters section, set oversampling factor to 2 (if this is not enough then you can try 2.5, 3, 4, …) - larger values increase more memory usage and computation time (oversampling factor of 2x increases memory usage by 2^3 = 8x).

    • Click Convert

Other mesh file formats can be loaded as model and then converted to segmentation node:

  • Drag-and-drop the volume file to the application window (or use menu: File / Add Data, then select the file)

  • Click OK

  • Go to Data module, Subject hierarchy tab

  • Right-click on the name of the imported volume and choose Convert model to segmentation node

Editing a segmentation imported from model (surface mesh) file

Selection of a source volume is required for editing a segmentation. The source volume specifies the geometry (origin, spacing, axis directions, and extents) of the voxel grid that is used during editing.

If no volume is available then it can be created by the following steps:

  • Go to Segment editor module

  • Click Specify geometry button (on the right side of Source volume node selector)

  • For Source geometry choose the segmentation (this specifies the extents, i.e., the bounding box so that the complete object is included)

  • Adjust Spacing values as needed. It is recommended to set the same value for all three axes. Using smaller values preserve more details but at the cost of increased memory usage and computation time.

  • Click OK

  • When an editing operation is started then the Segment Editor will ask if the source representation should be changed to binary labelmap. Answer Yes, because binary labelmap representation is required for editing.

Note

Certain editing operations are available directly on models, without converting to segmentation. For example, using Surface Toolbox and Dynamic Modeler modules.

Export segmentation to model (surface mesh) file

Segments can be exported to STL or OBJ files for 3D printing or visualization/processing in external software:

  • Open Export to files section in Segmentations module (or in Segment editor module: choose Export to files, in the drop-down menu of Segmentations button)

  • Choose destination folder, file format, etc.

  • Click Export

Other file formats:

  • Go to Data module, right-click on the segmentation node, and choose Export visible segments to models (alternatively, use Segmentations module’s Export/import models and labelmaps section)

  • In application menu, choose File / Save

  • Select File format

  • Click Save

Export segmentation to labelmap volume

If segments in a segmentation do not overlap each other then segmentation is saved as a 3D volume node by default when the scene is saved (application menu: File / Save). If the segmentation contains overlapping segments then it is saved as a 4D volume: each 3D volume containing a set of non-overlapping segments.

To force saving segmentation as a 3D volume, export it to a labelmap volume by right-clicking on the segmentation in Data module.

For advanced export options, Segmentations module’s Export/import models and labelmaps section can be used. If exported segmentation geometry (origin, spacing, axis directions, extents) must exactly match another volume’s then then choose that volume as Reference volume in Advanced section. Using a reference volume for labelmap export may result in the segmentation being cropped if some regions are outside of the new geometry. A confirmation popup will be displayed before the segmentation is cropped.

Export segmentation to labelmap volume file

If source representation of a a segmentation node is binary labelmap then the segmentation will be saved in standard NRRD file format. This is the recommended way of saving segmentation volumes, as it saves additional metadata (segment names, colors, DICOM terminology) in the image file in custom fields and allows saving of overlapping segments.

For exporting segmentation as NRRD or NIFTI file for external software that uses 3D labelmap volume file + color table file for segmentation storage:

  • Open Export to files section in Segmentations module (or in Segment editor module: choose Export to files, in the drop-down menu of Segmentations button)

  • In File format selector choose NRRD or NIFTI. NRRD format is recommended, as it is a simple, general-purpose file format. For neuroimaging, NIFTI file format may be a better choice, as it is the most commonly used research file format in that field.

  • Optional: choose Reference volume if you want your segmentation to match a selected volume’s geometry (origin, spacing, axis directions) instead of the current segmentation geometry

  • Optional: check Use color table values checkbox and select a color table to set voxel values in the exported files from values specified in the color table. The label value is index of the color table entry that has the same name as the segment name. If a color table is not specified then, voxel values are based on the order of segments in the segment list (voxels that are outside of all segments are set to 0, voxels of the first segment are set to 1, voxels of the second segment are set to 2, etc).

  • Set additional options (destination folder, compression, etc.) as needed

  • Click Export

Labelmap volumes can be created in any other formats by exporting segmentation to labelmap volume then in application menu, choose File / Save.

Using a reference volume for labelmap export may result in the segmentation being cropped if some regions are outside of the new geometry. A confirmation popup will be displayed before the segmentation is cropped.

Create new representation in segmentation (conversion)

The supported representations are listed in the Representations section. Existing representations are marked with a green tick, the source representation is marked with a gold star. The source representation is the editable (for example, in Segment Editor module) and it is the source of all conversions.

  • To create a representation, click on the Create button in the corresponding row. To specify a custom conversion path or parameters (reference geometry for labelmaps, smoothing for surfaces, etc.), click the down-arrow button in the “Create” button and choose “Advanced create…”, then choose a conversion path from the list at the top, and adjust the conversion parameters in the section at the bottom.

  • To update a representation (re-create from the source representation) using custom conversion path or parameters, click the “Update” button in the corresponding row.

  • To remove a representation, click the down-arrow button in the “Update” button then choose “Remove”.

Adjust how segments are displayed

  • By right-clicking the eye icon in the segments table the display options are shown and the different display modes can be turned on or off

  • Advanced display options are available in Segmentations module’s Display sections.

Managing segmentations using Python scripts

See Script repository’s Segmentations section for examples.

DICOM export

  • The source representation is used when exporting into DICOM, therefore you need to select a source volume, create binary labelmap representation and set it as master

  • DICOM Segmentation Object export if QuantitativeReporting extension is installed

  • Legacy DICOM RT structure set export is available if SlicerRT extension is installed. RT structure sets are not recommended for storing segmentations, as they cannot store arbitrarily complex 3D shapes.

  • Follow these instructions for exporting data in DICOM format.

Panels and their use

  • Segments table

    • Add/remove segments

    • Edit selected: takes user to Segment Editor module

    • Set visibility and per-segment display settings, opacity, color, segment name

  • Display

    • Segmentations-wide display settings (not per-segment!): visibility, opacity (will be multiplied with per-segment opacity for display)

    • Views: Individual views to show the active segmentation in

    • Slice intersection thickness

    • Representation in 3D/2D views: The representation to be shown in the 3D and 2D views. Useful if there are multiple representations available, for example if we want to show the closed surface in the 3D view but the labelmap in the slice views

  • Representations

    • List of supported representations and related operations

    • The already existing representations have a green tick, the source representation (that is the source of all conversions and the representation that can be edited) a gold star

    • The buttons in each row can be used to create, remove, update a representation

      • Advanced conversion is possible (to use the non-default path or conversion parameters) by long-pressing the Create or Update button

      • Existing representations can be made master by clicking Make source. The source representation is used as source for conversions, it is the one that can be edited, and saved to disk

  • Copy/move (import/export)

    • Left panel lists the segments in the active segmentation

    • Right panel shows the external data container

    • The arrow buttons van be used to copy (with plus sign) or move (no plus sign) segments between the segmentation and the external node

    • New labelmap or model can be created by clicking the appropriate button on the top of the right panel

    • Multiple segments can be exported into a labelmap. In case of overlapping segments, the subsequent segments will overwrite the previous ones!

Subject hierarchy

  • Segmentations are shown in subject hierarchy as any other node, with the exception that the contained segments are in a “virtual branch”.

    • The segments can be moved between segmentations, but drag&drop to anywhere other than under another segmentation is not allowed

  • Special subject hierarchy features for segmentations

    • Create representation: Create the chosen representation using the default path

  • Special subject hierarchy features for segments

    • Show only this segment: Useful if only one segment needs to be shown and there are many, so clicking the eye buttons would take a long time

    • Show all segments

Tutorials

Limitations

  • When segmentation is displayed in 2D views using Closed surface representation (either because this is the only available representation or because this representation is chosen to be shown in Segmentations module: Display / Advanced / Representation in 2D views -> Closed surface) then filling of the contours may appear incomplete and/or inverted. This is just a rendering error and does not affect the segmentation content. To avoid seeing such rendering artifacts, create Binary labelmap representation and choose Representation in 2D views -> Binary labelmap.

Information for developers

References

Contributors

Authors:

  • Csaba Pinter (PerkLab, Queen’s University)

  • Andras Lasso (PerkLab, Queen’s University)

  • Kyle Sunderland (PerkLab, Queen’s University)

Acknowledgements

This work is funded in part by An Applied Cancer Research Unit of Cancer Care Ontario with funds provided by the Ministry of Health and Long-Term Care and the Ontario Consortium for Adaptive Interventions in Radiation Oncology (OCAIRO) to provide free, open-source toolset for radiotherapy and related image-guided interventions.