Segmentation Methods

Segmentation is the process of identifying and isolating anatomical structures in medical images. In LeaRes, segmentation and 3D reconstruction live in a single workflow opened from the Segment and Reconstruct panel.

Segmentation Overview

Why Segment?

Segmentation enables:

• 3D Visualization: Create surface models of organs
• Functional Analysis: Run virtual spirometry and surgical planning on lung anatomy
• Simulation: Turn an anatomy into a CFD project (the bridge to the engineering workflow)

Segmentation Methods

The panel offers three methods, selected from the method selector at the top:

1. Auto segmentation (default): server-side AI on GPU. Organ groups come from pre-trained SegResNet checkpoints exported from MONAI Auto3DSeg (one each for cardiac, lungs and whole body). Two optional add-on models run as background jobs: TfeNet for airways and nnU-Net for coronary arteries.
2. Region growing: seed-based growth controlled by an intensity threshold
3. Thresholding: select voxels within an intensity range

Note: Voxel segmentation is driven by the three methods above. Two interactive tools sit outside this panel: Scissors (toolbar, 3D image carving) draws a spline contour on a slice and cuts the volume inside or outside it, with undo/redo; and Edit mask (Geometry editing group) offers a sphere/cube brush for adding or removing regions. Edit mask is not yet wired to the backend in this build.

Opening the Segmentation Panel

1. Load a DICOM acquisition in the viewer (see Uploading DICOM Data)
2. In the vertical toolbar, click Segment and Reconstruct
3. The panel opens at the top of the left Data sidebar, above the tree

The panel contains:

• Method selector: Auto segmentation, Region growing, or Thresholding
• Method controls: parameters for the selected method
• Select segments: the list of produced segments
• Surface post-processing lives in the Geometry editing group on the vertical toolbar (covered in 3D Reconstruction)
• Store selection as project: save the anatomy as a project/device

Method 1: Auto Segmentation

Auto segmentation uses AI models running on the server (GPU) to identify anatomy automatically.

Running Auto Segmentation

1. Make sure Auto segmentation is selected (it is the default)
2. Choose the anatomy to segment from the Target dropdown (default cardiac)
3. For the lungs, optionally enable the Segment airways (TfeNet) checkbox to also extract the trachea and bronchial tree
4. Press the round Start button

The job runs server-side on GPU. When it finishes, the produced segments appear in the Select segments list and an initial 3D surface is generated automatically.

Tip: Enable Segment airways (TfeNet) whenever you plan to run virtual spirometry — airway geometry improves the downstream lung analysis.

Method 2: Region Growing

Region growing segments a connected region that shares similar intensity values, starting from a seed you place on the volume.

Using Region Growing

1. Select Region growing in the method selector
2. Click Add Seed, then Alt+Click inside the target structure in any 2D view (axial, sagittal or coronal). Press Esc to remove the seed.
3. Adjust the Threshold range — it opens centred on the seed’s HU value ±100, within a ±300 track
4. Set Smoothness (1–10, default 4) to control how much the resulting surface is smoothed
5. Press the run button, or Enter, to launch region growing
6. Press Backspace to discard the resulting Region growing mask and start over

Tip: Place the seed away from organ boundaries and large vessels to avoid the region leaking into neighbouring tissue.

Method 3: Thresholding

Thresholding selects every voxel whose intensity falls inside a chosen range, regardless of connectivity.

Using Thresholding

1. Select Thresholding in the method selector
2. Set the Threshold range — it defaults to −500 … +500 HU on a −1000 … +1000 track — and the Smoothness level (1–10, default 4)
3. Press the generate button to launch the segmentation; the voxels inside the range form the segment

Approximate CT reference ranges (Hounsfield Units):

Tissue	HU range
Air / lung	-1000 to -700
Fat	-100 to -50
Water	~0
Soft tissue	20 to 100
Bone	200+