Working with in situ data

This guide outlines strategies for analyzing in situ membrane segmentations using Mosaic, demonstrated on a Giardia lamblia dataset.

Segmentation

Load your segmentation via File > Load Session. If you don’t have one, follow the membrane segmentation guide.

Tip

Load Session sets default values that optimize downstream processing and export.

For large datasets, adjust point rendering in Preferences > Appearance (Linux: High preset; macOS: Ultra or Adaptive).

Raw membrane segmentation

Connected Components

Separate the dataset into disjoint membrane partitions:

1. Select the segmentation in the Object Browser

2. In the Segmentation tab, configure Cluster:

   • Method: Connected Components

   • Use Points: Check

   • Drop Noise: Check

   • Distance: -1.0

3. Click Apply

Color by entity (View > Coloring > By Entity) to distinguish components.

Separated membrane components

Tip

Distance -1 uses single-voxel connectivity. Increase to merge components separated by multiple voxels.

Refinement

Remove small erroneous clusters using size-based filtering:

1. Click Select in the Segmentation tab

2. Adjust cutoffs to identify suitable size ranges (here, <25,000 voxels)

3. Click Remove

Size-based cluster filtering

Tip

Manual editing is available via Actions > Pick Objects and Actions > Point Selection (or keyboard shortcuts). Use Trim for lamella editing.

Clustering

Some membrane systems (e.g. double membranes) remain merged after connected components. Graph-based clustering can separate them.

Envelope Extraction

Optionally thin membranes to their envelope first, reducing computation and improving separation:

1. Select the target cluster

2. Configure Cluster: Method: Envelope, Use Points: Check, Distance: -1.0, click Apply

Note

Check Drop Noise to add the inner membrane part as a second cluster.

Slice through initial cluster

Identified envelope points

Leiden Clustering

Leiden clustering uses graph connectivity to separate membrane systems. The resolution parameter controls fineness — start at -7.3 and increase in steps of 1.0.

Here, resolution -7.3 yielded two clusters. Repeating at -6.3 for each produces the results below. Merge the resulting clusters into distinct membrane systems by selection.

Leiden clustering result

Merged membrane segmentation

Repeat for the remainder of the dataset.

Clustering applied to the entire dataset.

Tip

When connectivity alone is insufficient, use distance-based methods like K-Means. DBSCAN and Birch can also work but are harder to tune.

Meshing

Fit triangular meshes to analyze geometric properties:

1. Select membrane clusters in the Object Browser

2. In Parametrization, configure Mesh:

   • Method: Alpha Shape

   • Smoothness: 1.0, Curvature Weight: 1.0

   • Pressure: 0.0

   • Boundary Ring: 1, Alpha: 1.0

   Changed in version v1.2.1: Elastic Weight renamed to Smoothness (rescaled). Volume Weight renamed to Pressure. Neighbors, Scaling Factor, and Distance were removed. Curvature Weight: 10.0 is sensible pre 1.2.1.

3. Click Apply

Membrane segmentations

Membrane meshes

Alpha shapes work well for convex membrane morphologies. For non-convex membranes, use Ball Pivoting (e.g. with core-thinning via Segmentation > Skeletonize at radius 40, then Ball Pivoting at radius 50). Poisson reconstruction also produces complete meshes using a different completion strategy.

Changed in version v1.1.0: Thin was renamed to Skeletonize.

Analyze geometric properties via Segmentation > Properties:

Analyzing mesh properties.

Mesh area

Mesh volume

Mesh mean curvature (radius 10)