Segmentation

The segmentation (Seg) plugin contains a number of algorithms and functionalities for image segmentation, i.e. detecting and delineating objects in 2D images or 3D volumes.

Note

This plugin does not offer application-specific automated algorithms (e.g. lung segmentation from CT scans) but rather a set of common useful algorithms to either define segmentations manually or process them.

Segmentations can be represented either as label maps or annotations (meshes for volumes, splines for 2D images). The plugin contains algorithms (for instance Extract Mesh) to go from one representation to another, or perform further processing.

A label map is an image, typically with the same size/resolution of the original image, containing in each pixel a discrete value that represents the object that this pixel belongs to. Label maps are almost always encoded as unsigned byte images, and have a Modality set to LABEL. Some algorithms actually rely on this Modality flag and may not be available otherwise; the algorithm Set Modality can be used to indicate this information (e.g. on a label map that is loaded from a file and does not have this property yet). In those cases, the label map is visualized according to a special mode with color assignments. A general assumption is to have the value zero mapped to a background class shown by default as transparent, while all other values are attributed an opaque color.

You can use the Structures View to open a high-level GUI dedicated to browsing and configuring the display of label maps/segmentations.

We generally recommend to use our dedicated software ImFusion Labels to generate segmentations. When both applications are installed, the Annotation Plugin for the ImFusion Suite allows to create label maps with all annotations tools. (right-click on the image, then select Segmentation > Segmentation Toolset).