Threshold Segmentation

Thresholding is the most fundamental segmentation technique, classifying voxels based on their intensity values. This tutorial teaches you to use threshold-based tools effectively for extracting structures from volumetric data.

Estimated time: 25 minutes

Prerequisites:

• Completed previous Fundamentals tutorials
• A volumetric dataset with distinct intensity regions (CT data works well)

Medical Disclaimer

Segmentation results should be reviewed and validated by qualified professionals. Automated tools assist the segmentation process but are not a substitute for expert judgment, particularly in clinical or diagnostic applications.

---

Understanding Thresholding

Thresholding segments voxels by comparing their intensity to defined boundaries:

• Voxels with intensity within the range → included in mask (foreground)
• Voxels with intensity outside the range → excluded (background)

Mathematically:

$$Mask(x,y,z) = \begin{cases} 1 & \text{if } T_{lower} \leq I(x,y,z) \leq T_{upper} \\ 0 & \text{otherwise} \end{cases}$$

This works well when the target structure has distinct intensity values from surrounding tissues—for example, bone in CT imaging has much higher Hounsfield Unit values than soft tissue.

---

Basic Threshold Segmentation

Step 1: Open the Threshold Tool

1. Load a volumetric dataset (CT recommended for this tutorial).
2. Navigate to the Segmentation tab in the ribbon.
3. Click Threshold in the Create Mask section.

The Threshold panel opens with:

• Presets dropdown for predefined ranges
• Histogram display with threshold controls
• Lower Threshold and Upper Threshold sliders
• Profile Line tool for intensity analysis
• Result Options for mask configuration

Step 2: Understand the Histogram

The histogram shows how voxel intensities are distributed across your data:

• X-axis — Intensity values (Hounsfield Units for CT)
• Y-axis — Number of voxels at each intensity
• Peaks — Represent distinct tissues or materials

Typical CT histogram peaks:

• Air: around -1000 HU
• Soft tissue: around 0-100 HU
• Bone: above 200-300 HU

Step 3: Set Threshold Values

The Threshold tool provides multiple ways to set your range:

Using Presets:

• Click the Preset dropdown.
• Select a predefined range:
  • Custom — User-defined range (default)
  • Bone (CT) — Typical bone tissue threshold
  • Soft Tissue (CT) — Soft tissue in CT images
  • Air — Air cavity threshold
• The sliders adjust automatically to the preset values.

Manual Adjustment:

• Drag the Lower Threshold slider/handle on the histogram.
• Drag the Upper Threshold slider/handle.
• Or enter exact numeric values in the text fields.

Observe the Preview:

• A colored overlay appears in slice views showing selected voxels.
• Adjust until the target structure is cleanly selected.

Finding Good Thresholds

Start with a preset, then narrow gradually by looking for intensity "valleys" on the histogram—these often represent optimal boundaries between tissues. The histogram peaks represent distinct tissue types.

Step 4: Optional - Use the Profile Line Tool

For precise threshold determination along anatomical boundaries:

1. In the Threshold panel, enable the Profile Line group box.
2. In a slice view, draw a line across the boundary between structures:
   • Click to place the start point
   • Click again to place the end point
3. The Profile chart displays intensity values along the line.
4. Examine the profile to identify:
   • Where intensity changes sharply (structural boundaries)
   • Peak intensity values for your target structure
5. Use this information to refine your lower and upper threshold settings.

Profile Line Technique

Draw profiles across multiple anatomical transitions to understand your data's intensity characteristics. This helps you select thresholds that capture your target structure cleanly.

Step 5: Configure Result Options

Before applying the threshold:

1. Target Mask:

   • Select an existing mask to update, or
   • Choose "New Mask" to create a new mask object

2. Result Mode:

   • Replace — The thresholded voxels replace the existing mask content (default)
   • Merge — The thresholded voxels are added to existing mask content

3. Click Apply to create or update the mask.

Step 6: Review the Result

1. The new mask appears in the Object Browser.
2. Slice views show the mask overlay on the volume.
3. Check multiple slices to verify the segmentation quality.
4. If adjustments are needed, modify the threshold values and re-apply.

---

Refining Threshold Results

Threshold segmentation often requires post-processing to address:

• False positives — Unwanted regions included in the mask
• False negatives — Regions that should be included but aren't
• Noisy boundaries — Jagged or speckled edges
• Small artifacts — Disconnected fragments to be removed

Filtering Small Regions

Remove small disconnected noise during thresholding:

1. In the Threshold panel, enable Filter regions.
2. Set the Minimum region size threshold (in voxels).
3. Optionally enable Keep largest only to retain only the biggest connected component.
4. Apply the threshold.

Small artifacts and noise are automatically removed, leaving only meaningful regions.

Post-Processing with Smoothing

Clean up mask boundaries after thresholding:

1. Navigate to Segmentation → Operations → Smooth Mask.
2. Select a smoothing filter:
   • Median — Removes small isolated voxels (good for noise)
   • Gaussian — Smooths boundaries while preserving shape
3. Set an appropriate Radius (typically 1-3 pixels).
4. Click Apply.

Filling Internal Cavities

To close internal voids or holes in the segmentation:

1. Navigate to Segmentation → Operations → Cavity Fill.
2. Set the Maximum cavity size (in voxels).
3. Click Apply.

Internal cavities smaller than the threshold are automatically filled.

---

Multilevel Thresholding

When you need to segment multiple tissue types with different intensity ranges simultaneously:

Using Multilevel Thresholding

1. Navigate to Segmentation → Create Mask → Multilevel Thresholding.
2. Configure the parameters:
   • Number of masks — How many separate intensity classes to create (e.g., 3 for bone, soft tissue, air)
   • Method — Otsu's method automatically calculates optimal threshold boundaries
3. Click Apply.
4. Volvicon creates separate masks for each intensity class.

This approach is useful for:

• Segmenting multiple tissue types in one operation (bone, soft tissue, air)
• Creating initial segmentations when structure intensities don't overlap
• Analyzing the distribution of intensity classes in your data

---

Working with Multi-Label Masks

For complex segmentations with multiple regions:

Creating a Multi-Label Mask

1. Perform individual segmentations for each structure.
2. Navigate to Segmentation → Operations → Multi-label Mask.
3. Click Merge into Multi-label.
4. Select the masks to combine.
5. Select the target/result mask.
6. Click Apply.

Extracting Labels

To separate a multi-label mask:

1. Select the multi-label mask.
2. Navigate to Segmentation → Operations → Multi-label Mask.
3. Click Extract Labels as Separate Masks.
4. Individual masks are created for each label.

---

Threshold Best Practices

Choosing Thresholds

Approach	When to Use
Presets	Standard materials with known values (bone, air)
Histogram peaks	Data with clear intensity separation
Profile line	Precise boundary identification
Iterative refinement	When boundaries are gradual

Common Threshold Values (CT)

Structure	Approximate HU Range
Air	-1000 to -900
Lung tissue	-900 to -400
Fat	-150 to -50
Water/Fluid	-20 to +20
Soft tissue	+20 to +100
Muscle	+30 to +60
Contrast-enhanced vessels	+100 to +400
Trabecular bone	+200 to +500
Cortical bone	+500 to +1900
Metal	> +2000

HU Variation

These values are approximate and vary based on scanner, protocol, and patient factors. Always verify with your specific data using histogram analysis.

Quality Considerations

1. Partial Volume Effects — Voxels at boundaries contain mixed tissues, creating intermediate intensities. This causes jagged edges in threshold segmentations.

2. Noise — Random intensity variations can cause speckled segmentation results. Apply smoothing before thresholding for noisy data.

3. Inhomogeneity — The same tissue may have different intensities in different regions due to imaging artifacts. Consider region-based segmentation for such cases.

---

Practical Exercise: Segmenting Bone from CT

Follow this workflow to segment bone structures from a CT dataset:

Part 1: Initial Thresholding

1. Load a CT dataset (head, chest, or extremity).
2. Navigate to Segmentation → Create Mask → Threshold.
3. In the Threshold panel:
   • Click the Preset dropdown and select Bone (CT).
   • Observe the preview in the slice views showing selected bone.
   • If needed, fine-tune the lower threshold (typically 200-300 HU for cortical bone, 100-200 for trabecular).
4. Enable Filter regions:
   • Set Minimum region size to 100 voxels (removes noise).
   • Optionally enable Keep largest only if you want only the main skeleton.
5. Configure the result:
   • Target Mask: Select "New Mask" and name it "Bone".
   • Mode: Keep Replace (default).
6. Click Apply to create the mask.

Part 2: Review and Refine

1. Scroll through axial, sagittal, and coronal slices to verify segmentation.
2. Check for:
   • Are all bone regions captured?
   • Are there unwanted non-bone structures included?
   • Are boundaries smooth or speckled?
3. If adjustments are needed:
   • To add missing regions: Adjust threshold values lower and re-apply (or use Region Grow).
   • To remove extra regions: Adjust threshold values higher and re-apply.
   • To smooth noisy boundaries: Use Smooth Mask with Median filter and radius 1-2.

Part 3: Generate 3D Visualization

1. With the bone mask selected, navigate to Segmentation → Convert → Mask To Surface.
2. Use Optimal quality preset for final output.
3. Click Apply to generate the 3D mesh.
4. In the 3D view, rotate and inspect the model from multiple angles.
5. The surface mesh is now ready for:
   • 3D visualization
   • Export for 3D printing
   • CAD applications
   • Further geometric analysis

---

Troubleshooting

Too many voxels selected

• Lower threshold may be too low
• Upper threshold may be too high
• Try narrowing the range incrementally

Missing parts of the target structure

• Threshold range may be too narrow
• The structure may have varying intensity
• Consider using Region Grow to capture connected regions

Speckled or noisy result

• Data may be noisy—apply smoothing to the volume first
• Use morphological opening (erode then dilate) on the mask
• Enable region filtering to remove small fragments

Separate objects merged together

• Objects may have similar intensities
• Use Region Grow to isolate individual objects
• Use Watershed segmentation to separate touching objects

---

Scripting Threshold Segmentation

Automate thresholding with Python:

import ScriptingApi as api

app = api.Application()
mask_operations = app.get_mask_operations()

# Configure threshold parameters
threshold_params = api.ThresholdParams()
threshold_params.lower_threshold = 200
threshold_params.upper_threshold = 3000
threshold_params.filter_regions = True
threshold_params.min_region_size = 100
threshold_params.keep_largest = False

# Apply threshold to the active volume
result = mask_operations.threshold("CT_Volume", threshold_params)
print(f"Created mask: {result}")

See the MaskOperations API Reference for additional options.

---

Next Steps

With threshold segmentation mastered, continue to:

• Region-Based Segmentation — Learn connectivity-based techniques
• Manual Mask Editing — Refine segmentation results
• Creating 3D Surface Models — Generate meshes from masks

---

See Also

• Segmentation Tab Overview — Complete tool reference
• Threshold Tool Reference — Detailed parameter documentation
• Understanding Masks — Foundational concepts