Multilevel Thresholding

The Multilevel Thresholding tool automatically generates multiple segmentation masks from the target volume objects. It identifies different regions or objects within the volume image and creates a set of segmentation masks to capture these distinct areas. You can choose between Otsu's method (maximizing between-class variance) or K-means clustering for threshold determination.

Accessing the Tool

1. Navigate to the Segmentation tab in the ribbon.
2. Click Multilevel Thresholding in the Create Mask section.

How It Works

Otsu's multilevel thresholding algorithm:

1. Analyzes the intensity histogram of the volume.
2. Evaluates all possible threshold combinations.
3. Selects thresholds that maximize the variance between resulting classes.
4. Each class represents a distinct intensity range.

This method is particularly effective when the histogram shows clear multi-modal distribution (multiple peaks).

Parameters

Target Object(s)

• Description: Select the target volume objects on which the operation will be applied
• Options: Active Volume, Selected Volumes, Visible Volumes, All Volumes

Options

Thresholding Method

Choose the algorithm for automatic threshold determination:

• Otsu: Otsu thresholding is a global thresholding technique that selects the optimal thresholds by maximizing the between-class variance of the resulting segmented regions.
• K-means: K-means thresholding is a clustering-based approach to image segmentation. It partitions the image pixels into K clusters based on their intensity values, and the cluster centroids are then used as the thresholds for segmentation.

Number of Masks

• Description: The number of desired masks to generate
• Range: 1 to 999
• Default: 2
• Effect: Specifies how many distinct regions/classes the algorithm will identify

Workflow

1. Open the Multilevel Thresholding tool from the Segmentation tab.
2. Select the Target object(s) (Active Volume, Selected Volumes, Visible Volumes, or All Volumes).
3. Choose the Thresholding method (Otsu or K-means).
4. Set the Number of masks based on the expected number of distinct tissue types or regions.
5. Click Apply to run the automatic threshold analysis.
6. Review the generated masks—the tool creates separate mask objects for each identified region.

Use Cases

• Multi-tissue Segmentation: Automatically separate bone, soft tissue, and air in CT images.
• Material Analysis: Distinguish different material densities in industrial CT scans.
• Initial Segmentation: Create a starting point for further refinement.

info

The algorithm works best when different structures have distinct intensity peaks in the histogram. If structures have overlapping intensities, consider using other segmentation methods.

Tips

• Otsu Method: Best for images with clear multi-modal intensity distribution (distinct peaks in histogram)
• K-means Method: More flexible for varying intensity patterns; uses clustering instead of variance maximization
• Number of Masks: Start with a small number (2–3) and increase if needed
• Target Selection: Use "Active Volume" for single volume processing, or "Selected Volumes" for batch processing
• Review the histogram to understand the intensity distribution before running
• Use the resulting masks as starting points for more refined segmentation
• Combine with filtering operations to clean up the results

Technical Background

Otsu's method finds thresholds that minimize within-class variance (or equivalently, maximize between-class variance). For multilevel thresholding with thresholds $t_1, t_2, ..., t_N$, the algorithm maximizes:

$$\sigma^2_B = \sum_{i=1}^{N+1} \omega_i (\mu_i - \mu_T)^2$$

Where:

• $\omega_i$ is the probability of class $i$
• $\mu_i$ is the mean intensity of class $i$
• $\mu_T$ is the total mean intensity

See Also

• Threshold — Manual threshold segmentation.
• Iso-Threshold — Single-value threshold.
• Multi-label Mask — Working with multi-label masks.
• Segmentation Tab Overview — Overview of all segmentation tools.