Pipeline stage

Preprocess

Correct systematic acquisition artifacts — illumination non-uniformity, optical aberrations, stripe noise, cycle misregistration — so that downstream segmentation and feature extraction measure biology, not instrument signature.

The problem — Every fluorescence image carries a systematic shading field: the center of the field is brighter than the edges because the excitation illumination is not perfectly uniform. For Cell Painting, this translates directly into position-on-plate signal: a cell at the corner of a field looks dimmer in every intensity feature than a cell at the center, regardless of its biology. Uncorrected, this optics artifact leaks into morphological profiles as a measurable batch effect. The same phenomenon appears in different forms across modalities: stripe noise in light-sheet, stain variation across a whole-slide image, cycle-to-cycle registration error in spatial omics.

What it is / how it works — Preprocess estimates and removes the systematic component of the acquisition artifact, leaving residual that is biological signal plus shot noise. For fluorescence HCS, BaSiCPy and CIDRE estimate a per-channel flat-field correction function from the data itself — no clean reference needed. For light-sheet, stripe removal (destriping) and deconvolution (RedLionfish, clij2) address the complementary artifacts of the light-sheet geometry. For spatial omics, cross-cycle registration (ASHLAR, SimpleITK) realigns tiles that drifted between rounds of cyclic immunofluorescence. The output in every case is an image set where the systematic artifact has been removed to the point that it no longer dominates downstream measurements.

Where it breaks — Over-correction is the symmetric failure: BaSiCPy with too few images produces a noisy correction function that introduces variance rather than removing it; deconvolution that is iterated too many times synthesizes structure that was never there. The rule is to validate correction with a downstream readout gate — in Cell Painting, Illumination Correction Before You Trust a Profile failing a Percent Replicating check is a signal that the correction function, not the biology, is the problem. Preprocess is never self-validating; it validates only through what it enables downstream.

An illumination correction that looks smooth is not necessarily correct. Validate it by measuring profile reproducibility — not by inspecting the correction surface.

References

Appears in these notes

  • Cell Painting Pipelines — From Images to Phenotypic ProfilesA Cell Painting profile is the output of a long pipeline — illumination correction, segmentation, feature extraction, batch correction, then a reproducibility readout — and the number that matters (percent-replicating / mAP) lives only at the end. Every stage upstream can silently degrade it.
  • Light-Sheet Microscopy — The Pipeline Problem Behind the Beautiful VolumeThe rendered light-sheet volume is the easy part. Behind it is a petabyte-scale systems problem — ingest and chunking, deconvolution and destriping, stitching and registration, then tracking — and every stage is where the science silently degrades, not the final render.
  • How to Prepare Microscopy Data for AIMost of the work of applying AI to microscopy is data preparation — formatting, normalization, tiling, and label hygiene — and the single most consequential decision is a split that respects the non-independence of microscopy data. Get the split wrong and every downstream metric is leaked, not earned.
  • From Research Script to Production PipelineA notebook that worked once is not a pipeline, and a community model that scored well on its paper's data is not a validated component. Turning either into something deployable is the engineering work of pinning, wrapping, gating, and validating — most of which the original artifact deliberately skipped.
  • The Bioimage Analysis Ecosystem — Where Each Tool FitsBioimage analysis is not one tool but a landscape organized along two axes — interactive vs scriptable, and generalist vs specialist. Fiji, napari, CellProfiler, QuPath, ilastik, and the deep-learning segmenters each occupy a different cell of that grid, and choosing well means knowing which cell your problem lives in.
  • Illumination Correction Before You Trust a ProfileUneven illumination silently biases every downstream feature. Correct it first, or your batch effects are really optics.

← Back to the constellation