Random Walker Segmentation

Random Walker Segmentation

The Random Walker Segmentation algorithm is a graph-based method for image segmentation. It assigns labels to pixels(or voxels in 3D) based on probabilities derived from random walks starting at seed points. These seed points are defined by the user as belonging to specific regions, such as "foreground" or "background." The algorithm then calculates the likelihood of each unassigned pixel being reached by a random walker starting from each seed. This approach is particularly effective for handling noisy data and achieving smooth segmentations while respecting the boundaries in the input image.

This use case demonstrates the application of the random walker algorithm on the walnut.vvd dataset using a predefinedsample workspace. The focus is on Voreen's Application Mode rather than pipeline construction.

To access the segmentation-randomwalker.vws workspace in Voreen, follow these steps:
File → Open Workspaces → Workspaces → segmentation-randomwalker.vws → Open

A canvas element and a large network will now appear. Since the network does not require manipulation in this case, switch the view from Network Mode to Application Mode for easier handling. This can be done via the toolbar Toolbar.

After these steps, your screen should resemble the following:

As shown, this workspace takes 3D data and displays it in a 2x2 grid, featuring a 3D view alongside three slices representing different planes: $XY, \ XZ, \ YZ$ (or, in medical terms, the axial, sagittal, and coronal planes).

Additionally, predefined properties (Input; 2D Rendering, RandomWalker, Seeds, and Output) enable users to interact with the visualization.

Operations in the Properties Space:

• Input:
  Use Clear Volume to remove the current input file and Load Volume to load new data.
• 2D Rendering:
  Adjust the transfer function for visualization or explore different data slices.
• Random Walker:
  Select from various noise models in the Noise Model dropdown for the segmentation.
• Seeds:
  Clear all existing seeds for segmentation.
• Output:
  Save either the probabilities file or the segmentation file as output.

Operations in the Canvas Space:

• Adjust the slice number for each plane using the scroll wheel.
• Add or delete seeds in each plane with the following operations:
  • Add a foreground seed: Ctrl + Left Click
  • Delete a foreground seed: Ctrl + Right Click
  • Add a background seed: Shift + Left Click
  • Delete a background seed: Shift + Right Click