Segmentation Recipes

Recipes capture complete segmentation workflows that can be replayed, tested, and optimized. They record click locations, parameters, and expected results for reproducible segmentation.

What is a Recipe?

A recipe is a saved sequence of segmentation actions:

# Example recipe structure
{
    "name": "Brain Tumor Segmentation",
    "sample_data": "MRBrainTumor1",
    "gold_standard": "MRBrainTumor1_tumor",
    "actions": [
        {
            "type": "adaptive_brush",
            "ras": [-5.31, 34.77, 20.83],
            "params": {
                "algorithm": "watershed",
                "brush_radius_mm": 25.0,
                "edge_sensitivity": 65
            }
        },
        # ... more actions
    ]
}

Using Recipes

Loading a Recipe

1. Open the SegmentEditorAdaptiveBrushReviewer module

2. Click Load Recipe

3. Select a .json recipe file

Replaying a Recipe

The Reviewer module provides step-by-step replay:

Button	Action
Play	Run all steps automatically
Step	Execute one action at a time
Reset	Return to initial state

Comparing to Gold Standard

When a recipe has an associated gold standard:

1. The gold standard loads automatically (shown in green)

2. Your segmentation shows in red

3. Overlap shows in yellow

4. Dice coefficient displays in real-time

Creating Recipes

From the Recorder

1. Open SegmentEditorAdaptiveBrushReviewer

2. Click Start Recording

3. Perform your segmentation in Segment Editor

4. Click Stop Recording

5. Save the recipe

From Python

from SegmentEditorAdaptiveBrushTesterLib import Recipe, Action

recipe = Recipe(
    name="My Segmentation",
    sample_data="MRBrainTumor1",
    gold_standard="MRBrainTumor1_tumor"
)

# Add actions
recipe.add_action(Action(
    action_type="adaptive_brush",
    ras=[-5.31, 34.77, 20.83],
    params={
        "algorithm": "watershed",
        "brush_radius_mm": 25.0,
    }
))

# Save
recipe.save("my_recipe.json")

From Optimization Results

After running optimization, the best trial can be converted to a recipe:

# In Slicer Python console
from SegmentEditorAdaptiveBrushTesterLib import Recipe
import json

# Load optimization results
with open("optimization_results/2026-01-26_.../results.json") as f:
    results = json.load(f)

best = results["best_trial"]

# Create recipe from best trial
recipe = Recipe(
    name="Optimized Brain Tumor",
    sample_data="MRBrainTumor1",
    gold_standard="MRBrainTumor1_tumor"
)

for click in best["user_attrs"]["click_locations"]:
    recipe.add_action(Action(
        action_type="adaptive_brush",
        ras=click["ras"],
        params=click["params"]
    ))

recipe.save("optimized_recipe.json")

Recipe File Format

Recipes are stored as JSON:

{
    "version": "1.0",
    "name": "Brain Tumor Segmentation",
    "description": "5-click tumor segmentation using watershed",
    "sample_data": "MRBrainTumor1",
    "gold_standard": "MRBrainTumor1_tumor",
    "created": "2026-01-26T06:20:00",
    "actions": [
        {
            "type": "adaptive_brush",
            "ras": [-5.31, 34.77, 20.83],
            "params": {
                "algorithm": "watershed",
                "brush_radius_mm": 25.0,
                "edge_sensitivity": 65,
                "threshold_zone": 60
            }
        }
    ],
    "metadata": {
        "author": "optimization",
        "dice": 0.9991,
        "notes": "Best result from 50-trial optimization"
    }
}

Gold Standards

What is a Gold Standard?

A gold standard is a reference segmentation used to:

• Measure recipe quality (Dice coefficient)

• Test for regressions

• Optimize parameters

Gold Standard Structure

GoldStandards/
└── MRBrainTumor1_tumor/
    ├── gold.seg.nrrd    # The segmentation
    ├── metadata.json    # Parameters, clicks, etc.
    └── reference_screenshots/

Creating a Gold Standard

From the Reviewer module:

1. Create a high-quality segmentation

2. Click Save as Gold Standard

3. Enter a name and description

4. The gold standard is saved with metadata

From Python:

from SegmentEditorAdaptiveBrushTesterLib import GoldStandardManager

manager = GoldStandardManager()
manager.save_as_gold(
    segmentation_node=seg_node,
    volume_node=vol_node,
    segment_id="Segment_1",
    name="MRBrainTumor1_tumor",
    click_locations=clicks,
    description="Reference tumor segmentation",
    algorithm="watershed",
    parameters={"brush_radius_mm": 25.0}
)

Testing Recipes

Running Regression Tests

# From command line
Slicer --python-script scripts/run_regression.py recipes/brain_tumor_1.py

In Python

from SegmentEditorAdaptiveBrushTesterLib import RecipeTestRunner

runner = RecipeTestRunner()
result = runner.run_recipe("recipes/brain_tumor_1.py")

print(f"Dice: {result.dice:.4f}")
print(f"Passed: {result.passed}")

Best Practices

Recipe Design

1. Use consistent click locations that capture the structure well

2. Start from the center of the structure

3. Add boundary clicks for edges that need attention

4. Document the purpose in the recipe description

Gold Standard Quality

1. Manual refinement - Correct any errors before saving

2. Include edge cases - Test challenging boundaries

3. Document parameters - Record what worked

Version Control

• Store recipes in version control

• Gold standards can be regenerated from recipes

• Track parameter changes over time