GitHub

Edge Classification

Cellular SMLM data has a boundary: localizations belong to a cell interior, sit on its membrane, or fall outside it in the surrounding background. The edge classification step carves one or more cell masks directly from the localization cloud and labels every localization :interior / :membrane / :outside. It is backed by SMLMClustering and, like clustering, is selected entirely by the type of its config:

  • KdeValleyConfig — the dSTORM-tuned default. Finds the cell boundary from a KDE valley between the dense interior and the sparse background.
  • OuterPolygonConfig — carves the outer ring of the dominant cell as a polygon.

Both are <: AbstractEdgeClassifyConfig, re-exported from SMLMClustering as const aliases, so they are available unqualified after using SMLMAnalysis.

(smld, info) = analyze(smld, KdeValleyConfig())     # → (smld, StepInfo)
(smld, info) = analyze(smld, OuterPolygonConfig())

Non-destructive by design

The step is state-modifying but emitter-preserving: the returned SMLD wraps the same emitters — nothing is filtered out. The authoritative per-emitter class is not written back onto the emitters (a per-emitter side-list would desync the moment a downstream step subsets emitters); it lives in the step's EdgeClassifyInfo, reached at the classify point:

(smld, step_info) = analyze(smld, KdeValleyConfig())
info = step_info.info                 # ::EdgeClassifyInfo
info.class                            # per-emitter Vector{Symbol}
interior_mask(info)                   # BitVector of the :interior emitters
in_cell(info)                         # :interior ∪ :membrane
interior_fraction(info)               # scalar summary

Only the cell-mask geometry travels downstream, mirrored into metadata:

  • metadata["edge_cells"] — the MultiCellMask (Vector{CellPolygon}, largest-first)
  • metadata["edge_outer_polygon"] — the dominant cell's outer ring (back-compat)

A later spatial-statistics step reads that geometry as its observation region — e.g. HopkinsConfig uses edge_cells / edge_outer_polygon to restrict its clustering tendency estimate to inside the cell.

When to use

  • Run late in the pipeline, on a cleaned BasicSMLD, when downstream analysis must distinguish cell interior from membrane or background — e.g. restricting a Hopkins clustering-tendency estimate to the cell, or reporting an interior fraction.
  • Because it removes nothing, it is safe to run before clustering or rendering; the emitters and their ordering are unchanged.

Outputs

At verbose >= STANDARD with an outdir, the step writes config.toml, info.toml, and an edge report (diagnostics plus figures) under {step}_edge_classify/. The StepInfo summary records the gate :method, emitter counts, and the number of cells.

See the SMLMClustering documentation for the algorithm details and the full per-config parameter list.