Pipeline Steps: Overview

This section documents every analysis step — one page per analyze() step. Each step is selected by the type of its config (see The Pipeline Model), runs in the order you list it in AnalysisConfig.steps, and returns an updated SMLD plus a StepInfo. This page is the map: the catalog, the typical order, and where each method comes from in the literature.

Catalog

Step	Config	Backed by	What it does
Detection & Fitting	`DetectFitConfig`	SMLMBoxer + GaussMLE	Detect ROIs in raw frames and fit a PSF to each → the initial localizations
Quality Filter	`FilterConfig`	SMLMAnalysis	Threshold on photons, precision, p-value, PSF width, z
Intensity Filter	`IntensityFilterConfig`	SMLMAnalysis	Reject multi-emitter events via a Poisson upper-tail test
Frame Connection	`FrameConnectConfig`	SMLMFrameConnection	Link repeated blinks into single, higher-precision localizations
Drift Correction	`DriftConfig`	SMLMDriftCorrection	Estimate and remove sample drift (fiducial-free)
Density Filter	`DensityFilterConfig`	SMLMAnalysis	Drop isolated localizations by neighbor count
Rendering	`RenderConfig`	SMLMRender	Render localizations to a super-resolution image
Bayesian Grouping	`BaGoLConfig`	SMLMBaGoL	Group localizations into true emitters (RJMCMC)
Clustering	`DBSCANConfig`, `VoronoiConfig`, `HopkinsConfig`, …	SMLMClustering	Label clusters / compute spatial statistics
Edge Classification	`KdeValleyConfig`, `OuterPolygonConfig`	SMLMClustering	Label localizations interior / membrane / outside a cell mask
Composite Render	`CompositeRenderConfig`	SMLMAnalysis + SMLMRender	Multi-channel RGB overlay
Cross-Alignment	`CrossAlignConfig`	SMLMAnalysis + SMLMDriftCorrection	Register channels to a common frame
Cross-Correlation	`CrossCorrConfig`	SMLMAnalysis + NearestNeighbors	Pair cross-correlation g(r) between channels

The last three operate on multiple channels (Vector{BasicSMLD}) and are usually driven by MultiTargetConfig — see Multi-Channel.

A typical pipeline

A standard single-color dSTORM / DNA-PAINT analysis runs roughly in this order:

steps = [
    DetectFitConfig(boxer = BoxerConfig(boxsize = 9, psf_sigma = 0.130),
                    fitter = GaussMLEConfig(psf_model = GaussianXYNBS())),
    FilterConfig(photons = (500.0, Inf)),     # cut weak / spurious fits
    FrameConnectConfig(max_frame_gap = 5),     # link repeated blinks
    DriftConfig(degree = 2),                   # remove drift
    RenderConfig(zoom = 20, colormap = :inferno),
]

Common variations:

Add an Intensity Filter early to reject multi-emitter events.
Add a Density Filter after drift correction to clean isolated noise.
Replace/augment the final render with Bayesian Grouping for counting and maximum precision, then render the grouped emitters.
Add a Clustering step to label structures, then render colored by label.

Steps are composable (see The Pipeline Model): reorder, repeat, or omit them as the experiment requires.

Primary literature

Each method's primary reference, as published by the backing package. Full citations are on the References page; each step page repeats its own reference and links to the upstream documentation for algorithm details.

Step	Method	Primary reference
Detection	Gaussian-filter local-max + sCMOS-aware thresholding	Huang et al., Nat. Methods 10, 653 (2013)
Fitting	MLE Gaussian/astigmatic PSF; CRLB precision	Smith et al., Nat. Methods 7, 373 (2010)
Frame connection	Spatiotemporal LAP clustering of repeated blinks	Schodt & Lidke, Front. Bioinform. (2021)
Drift correction	Fiducial-free entropy minimization (Legendre basis)	Cnossen et al., Opt. Express 29, 27961 (2021); Wester et al., Sci. Rep. 11, 23672 (2021)
Bayesian grouping	Collapsed reversible-jump MCMC grouping	Fazel et al., Nat. Commun. 13, 7152 (2022)
Clustering	DBSCAN; Voronoi tessellation (SR-Tesseler)	Ester et al., KDD-96 (1996); Levet et al., Nat. Methods 12, 1065 (2015)

The native SMLMAnalysis steps — Quality Filter, Intensity Filter, Density Filter, and the multi-channel steps — are described in full on their own pages, including the method and any reference.