Medical Imaging Interaction Toolkit  2025.08.99-f7084adb
Medical Imaging Interaction Toolkit
MitkConvertToMultiLabelSegmentation User Guide

Overview

This command-line tool converts various medical data types (images, surfaces, and contours) into a multi-label segmentation using the Medical Imaging Interaction Toolkit (MITK).

Features

  • Converts images, surface meshes, and contours into a single multi-label segmentation output.
  • Supports grouping inputs into distinct label groups.
  • Automatically resolves conflicting label values.
  • Ensures geometry consistency and validates spatial alignment.
  • Accepts DICOM, NIfTI, and other MITK-compatible image formats.

Basic Usage

convertToMultiLabelSegmentation -i <input1> <input2> ... -o <output.nrrd> [options]

Required Arguments

Argument Description
-i, --inputs One or more input files (images, surfaces, or contours).
-o, --output Output file path for the resulting multi-label segmentation.

Optional Arguments

Argument Description
-r, --reference Reference image for defining the output geometry (required if no image inputs are used).
-g, --groups Enable grouping: each input is placed into a separate label group. Default: all inputs are in one group.
-h, --help Show help and usage information.

๐Ÿงช Examples

Example 1: Convert a single binary image mask

convertToMultiLabelSegmentation -i tumor.nii.gz -o tumor.nrrd

What happens:
A single binary image is converted to a multi-label segmentation with one label.

Label & Group Naming:

  • Group: Unnamed default group
  • Label: tumor Value 1

Resulting Pixel Values:

  • Label tumor Value 1: value 1

Example 2: Merge multiple binary masks into one group

convertToMultiLabelSegmentation -i mask1.nii.gz mask2.nii.gz -o merged.nrrd

What happens:
Two binary image masks are merged into a single label group.

Label & Group Naming:

  • Group: Unnamed default group
  • Labels:
    • mask1 Value 1
    • mask2 Value 1

Resulting Pixel Values:

  • Label mask1 Value 1: value 1
  • Label mask2 Value 1: value 2

Example 3: Group surface inputs into separate label groups (with reference image)

convertToMultiLabelSegmentation -i liver.stl spleen.stl kidney.stl -r reference.nii.gz -o organs.nrrd -g

What happens:
Each surface is rasterized using the reference image and placed in its own group.

Label & Group Naming:

  • Group liver: Label liver
  • Group spleen: Label spleen
  • Group kidney: Label kidney

Resulting Pixel Values:

  • Group liver, Label liver: value 1
  • Group spleen, Label spleen: value 2
  • Group kidney, Label kidney: value 3

Example 4: Convert a contour set using reference image

convertToMultiLabelSegmentation -i roi.contour -r referenceCT.nii.gz -o contour.nrrd

What happens:
Contours are rasterized into a binary mask using the reference image geometry.

Label & Group Naming:

  • Group: Unnamed default group
  • Label: roi

Resulting Pixel Values:

  • Label roi: value 1

Example 5: Input image with multiple pixel values (multi-label image)

convertToMultiLabelSegmentation -i MultiLabelSeg.nii.gz -o resegmented.nrrd

What happens:
An image containing multiple pixel values (e.g., 1 = liver, 2 = spleen, 3 = kidney) is converted.

Label & Group Naming:

  • Group: Unnamed default group
  • Labels:
    • MultiLabelSeg Value 1
    • MultiLabelSeg Value 2
    • MultiLabelSeg Value 3

Resulting Pixel Values:

  • MultiLabelSeg Value 1: value 1
  • MultiLabelSeg Value 2: value 2
  • MultiLabelSeg Value 3: value 3

๐Ÿ“‚ Details

๐Ÿท Label Naming Strategy

  • Image inputs:
    • Each unique pixel value is treated as a separate label.
    • Label names follow the format: <filename> Value <pixel_value>.
  • Non-image inputs (surfaces, contours):
    • Treated as binary masks.
    • Label name is the filename without extension.

๐Ÿ“ Group Naming Strategy

  • When --groups is enabled:
    • Each input becomes a separate group, named from the input filename.
  • When --groups is not enabled (default):
    • All labels are placed in a single unnamed group.

๐ŸŽฏ Label Value Assignment Strategy

  • Image Inputs:
    • Unique pixel values are extracted.
    • If label values overlap across inputs, the tool automatically remaps values to avoid conflicts.
    • Label names still reflect original values (e.g., Value 2), even if actual pixel values are remapped.
  • Non-Image Inputs:
    • Converted into binary masks.
    • Default pixel value is 1.
    • If conflicts exist, remapping is applied.
  • Label values after conversion are unique over all groups.

Output Format

The resulting file is a standard multi-label segmentation in NRRD format:

  • Can be loaded into MITK or compatible toolkits.
  • Encodes labels and groups as distinct entities.
  • Stores label metadata (names, values, group names).