Skip to content

Membrane Refinement

Membrane refinement is a post-processing step that improves the quality and consistency of organelle and membrane segmentations. This tutorial covers how to use SABER's GPU-optimized membrane refinement pipeline to clean up and enhance your segmentation results.

Membrane Refinement Example Membrane refinement eliminates false positive membrane detections and creates topologically consistent organelle-membrane pairs. This workflow takes raw binary segmentation with spurious membrane fragments and produces masks with clean boundaries and unique instance labels matching each organelle.

🔬 Understanding Membrane Refinement

Why Refine Membranes?

Raw segmentation outputs often contain artifacts and inconsistencies that can affect downstream analysis:

  • Boundary artifacts: Segmentation errors at image edges
  • Small noise objects: Tiny false positive detections
  • Topological inconsistencies: Organelles extending beyond membrane boundaries
  • Surface membrane confusion: Internal membrane fragments that should be removed

The Combined Mask Approach

SABER's membrane refinement uses a "combined mask" strategy:

How the combined mask approach works
  1. Membrane subtraction: Membrane pixels are subtracted from organelle pixels
  2. Interior creation: This creates a clean interior mask for each organelle
  3. Morphological cleaning: Opening operations remove small artifacts
  4. Constraint application: The cleaned interior constrains both organelle and membrane
  5. Topological consistency: Ensures organelles stay within membrane boundaries

This approach ensures that: - Organelles are properly contained within their membranes - Membranes are refined to match their corresponding organelles - Both segmentations maintain topological consistency - Small artifacts and noise are effectively removed

📋 Generating Initial Segmentations

For quick results, we recommend using MemBrain-seg to generate initial membrane segmentations. MemBrain-seg is specifically designed for membrane segmentation in cryo-electron tomography and provides high-quality starting points for refinement.

🚀 Running Membrane Refinement

Basic Command

Refine organelle and membrane segmentations using the CLI:

saber analysis refine-membranes \
    --config config.json \
    --org-info "organelles,saber,1" \
    --mem-info "membranes,membrane-seg,1" \
    --voxel-size 10 \
    --save-session-id "1"
Expected Output

The refinement process will create new segmentations in your copick project with the same object names but under the specified session ID.

For example, if you run the command above:

  • Input: organelles segmentation in session 1 with user saber
  • Input: membranes segmentation in session 1 with user membrane-seg
  • Output: organelles segmentation in session 1 with user saber-refined
  • Output: membranes segmentation in session 1 with user membrane-seg-refined

The refined segmentations will have the same voxel size and coordinate system as your input data, but with improved quality through morphological filtering and topological consistency.

Input Query Specification

SABER provides two flexible methods for specifying your input segmentations through the --org-info and --mem-info parameters:

Option 1: Simple Name Query (Uses Defaults)

When you only specify the segmentation name, SABER will use default values for user ID and session ID:

# Simple format - uses default user and session
--org-info "organelles"
--mem-info "membranes"

The default behavior will use the first available userID and sessionID found in the project. This is ideal for copick projects with single segmentations for the given segmentation names.

Option 2: Full Specification (Explicit Control)

For precise control over which segmentations to use, provide the complete query string:

# Full format: "name,userID,sessionID"
--org-info "mitochondria,saber,3"
--mem-info "membranes,membrain-seg,2"
Advantages of full specification
  • Reproducibility: Ensures you always use the same segmentations
  • Multi-user projects: Specify exactly which user's segmentations to use
  • Version control: Target specific session versions of your segmentations
  • Mixed sources: Use organelle and membrane segmentations from different users/sessions

Parameter Explanation

Parameter Description Default Example
--config Path to copick config file config.json my_project.json
--org-info Organelle segmentation info organelles,saber,1 mitochondria,user1,session2
--mem-info Membrane segmentation info membranes,membrane-seg,1 mito_membranes,user1,session2
--voxel-size Voxel size in Angstroms 10 5.2
--save-session-id Session ID for refined results 1 refined-v2

🔧 API Call

Using the Analysis Module Directly

For custom refinement workflows, you can use the analysis module directly:

from saber.analysis.refine_membranes import OrganelleMembraneFilter, FilteringConfig

# Create custom configuration
config = FilteringConfig(
    ball_size=5,              # Morphological operation kernel size
    min_membrane_area=10000,  # Minimum membrane component size
    edge_trim_z=5,           # Z-edge trimming pixels
    edge_trim_xy=3,          # XY-edge trimming pixels
    batch_size=8,            # GPU batch processing size
    keep_surface_membranes=False  # Remove internal membranes
)

# Initialize filter
filter_obj = OrganelleMembraneFilter(config, gpu_id=0)

# Run refinement
results = filter_obj.run(organelle_seg, membrane_seg)

# Access results
refined_organelles = results['organelles']
refined_membranes = results['membranes']

Configuration Parameters

Detailed parameter descriptions
Parameter Type Default Description
ball_size int 3 Radius of morphological operation kernel
min_membrane_area int 10000 Minimum area for membrane components
edge_trim_z int 5 Pixels to trim from Z edges
edge_trim_xy int 3 Pixels to trim from XY edges
min_roi_relative_size float 0.15 Minimum ROI size relative to organelle
batch_size int 8 GPU batch processing size
keep_surface_membranes bool False Keep only surface membranes

Parameter tuning guidelines:

  • ball_size: Larger values create smoother boundaries but may remove fine details
  • min_membrane_area: Adjust based on your expected membrane component sizes
  • edge_trim_*: Increase if you have boundary artifacts
  • keep_surface_membranes: Enable to remove internal membrane fragments

🚀 What's Next?

After membrane refinement, you can:

  • Analyze refined results: Use the cleaned segmentations for quantitative analysis
  • Visualize results: Create galleries and visualizations of refined segmentations
  • Export for other tools: Use refined segmentations in external analysis pipelines
  • Iterate and improve: Adjust parameters based on results and re-run if needed

Integration with other SABER tools: Refined segmentations work seamlessly with SABER's analysis and visualization modules.