Modality

Digital Pathology

Digital pathology processes gigapixel whole-slide images — H&E, IHC, and multiplexed tissue sections scanned at 20–40x — through a pyramid of tiling, stain normalization, nuclear segmentation, and foundation-model embeddings to produce graded, biomarker-resolved, annotated WSI reports.

What it is — Digital pathology converts a glass tissue section into a computational object. A scanner captures the stained slide at 20–40x magnification, writing a pyramidal file (SVS, NDPI, or MRXS) that encodes the full-resolution image alongside progressively downsampled overviews. A single slide is a gigapixel image — OpenSlide or tifffile reads the pyramid; no consumer imaging library does. Everything downstream operates on tiles extracted from this pyramid, making tiling strategy a pipeline decision, not an afterthought: tile size, stride, overlap, and foreground-only extraction all affect what the model sees and what gets silently discarded.

The preprocessing contract — Staining is done by hand, batch by batch, and the colour distribution of an H&E slide drifts with the technician, the reagent lot, and the scanner. Stain normalization (Macenko, Vahadane) maps each slide into a common stain space before any downstream model runs. Skipping it does not make the variability disappear; it buries it in the embedding. Upstream of normalization, slide-level QC must catch blurred tiles, folded tissue, ink artefacts, and insufficient tissue coverage — artefacts a segmentation model will confidently process without flagging. Why Microscopy Needs Verified Outputs applies here: a WSI QC report is a first-class pipeline deliverable, not optional metadata.

Segmentation and embeddings — Tissue structure is segmented at two scales: regions (tumour, stroma, gland boundaries) and cells (nuclei classification and detection). HoVer-Net, StarDist, and QuPath handle the nuclear layer with different trade-offs between throughput, instance separation, and classification granularity. Above segmentation, foundation-model embeddings are now standard at the features stage: UNI (ViT-L trained on 100M+ histology patches), H-optimus-0, and CONCH each produce tile-level vectors that aggregate to slide-level representations via attention-based MIL. Foundation Models Are Not Pipelines is the precise account of why the encoder is not the system: stain drift, tile QC, and domain shift do not resolve themselves inside a frozen encoder.

Clinical readout — The pipeline terminus is a grading or biomarker prediction tied to ROIs on the slide — grade regression, PD-L1 scoring, or spatial cell-type fractions — plus a heatmap overlay that makes the evidence auditable. Without that overlay and the QC chain that precedes it, a slide-level prediction is a number without a provenance trail.

The whole-slide file is not the input — a defensible, QC-gated, stain-normalised tile set is. Build the pipeline from the ingest contract forward, not from the model backward.

How we run it — the Digital Pathology pipeline, stage by stage — the nine-stage imaging backbone reshaped for this modality, with the tools, sub-steps and deliverable at each stage.

References

Appears in these notes

  • The QC-Aware ReportA QC-aware report doesn't just present a result — it surfaces the evidence that the result is trustworthy, structured by the four nested levels of quality and ending in an explicit decision. Its job is to make a number un-trustable on sight when it shouldn't be trusted.
  • Choosing a Bioimage Analysis Pipeline — Off-the-Shelf vs CustomMost microscopy questions are answered by composing existing, validated tools — and a custom pipeline is justified only when off-the-shelf options fail on a specific, identifiable axis. This note gives the decision criteria and the signals that you have genuinely crossed into custom territory.
  • Data Standards and Scalable StorageThe OME data model and OME-NGFF/OME-Zarr exist because a microscopy image without a shared data model and a chunked, cloud-native layout is neither interpretable nor analyzable at scale. Format is not plumbing — it decides whether a petabyte dataset can be opened at all.
  • Evaluating Segmentation Models in Real Microscopy WorkflowsA segmenter's benchmark Dice rarely predicts its behavior in a real workflow. Instance counting, merge/split errors, and cross-instrument generalization are the quantities that decide whether the downstream readout is correct — and standard semantic overlap scores are blind to all three.
  • Foundation Models Are Not PipelinesA foundation model is a powerful feature extractor, not a deployable measurement pipeline. It still needs ingestion, QC, preprocessing, confounder correction, validation, and provenance around it — and a frozen encoder makes the surrounding system more important, not less.
  • Ground Truth Is a Design ProblemGround truth in microscopy is constructed, not given — every label is the output of an annotation protocol, an annotator, and a fusion rule, each with its own bias and variance. Treating it as a fixed answer key is how inter-rater noise and label bias get baked into every score computed against it.
  • Image Quality Is Not One NumberFovea treats microscopy quality as four nested levels — image-space, run/sample, readout, and decision — because a metric at one level says nothing about the levels above it. Full-reference scalars like SSIM and PSNR measure only the first, yet teams report as if they measured the third.
  • Why Interoperability Matters in MicroscopyA microscopy pipeline is always several tools in sequence, so the seams between them — file formats, metadata, coordinate conventions — decide whether the pipeline composes or quietly corrupts. Open standards like OME-Zarr/NGFF turn hand-offs from lossy conversions into clean interfaces, and avoid lock-in to any one vendor or tool.
  • Why mAP, Dice, and SSIM Are Not EnoughDice, IoU, mAP, and SSIM measure overlap or similarity against a reference — not whether the result supports the decision the experiment exists to make. The discipline is problem-aware metric selection: derive the metric from the question, not from what is easy to compute.
  • Model Zoos for Bioimage Analysis — From a Zoo Model to a Validated WorkflowThe BioImage Model Zoo and community packaging tools make pre-trained models shareable, runnable, and reproducible across tools — solving distribution and provenance. They do not solve fitness: a downloaded model is a component, and the path from a zoo entry to a trustworthy readout still runs through validation.
  • Verification — A Pipeline You Can Re-RunVerification is the engineering discipline of making a microscopy pipeline deterministic, version-pinned, and re-executable from a provenance record. Without it, every result is a one-off, and "we changed nothing" is unprovable.
  • 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.
  • Every Output Should Know Where It Came FromProvenance is a property of the artifact, not the pipeline — every mask, profile, and count matrix should carry the lineage that produced it. Where verification asks whether a run reproduces, provenance asks whether a single output, found alone, can still explain itself.
  • The Challenge of Rare Events and MutantsWhen the phenotype you care about appears in 10 cells out of 300,000, accuracy is meaningless, validation has almost no positives to learn from, and false discovery dominates. Rare-event detection is a class-imbalance and ground-truth-scarcity problem — best answered by adaptive acquisition that goes looking for the positives.
  • How to Make Imaging Data Ready for ReanalysisReanalysis-ready means a third party with no contact with the original lab can re-run, re-segment, or re-interpret the data correctly — which requires FAIR principles, complete metadata, and deposition in a public archive. The bar is reuse by a stranger, not retrieval by the author.
  • 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.
  • Uncertainty, Drift, and Failure Modes in Bioimage AnalysisModels fail silently when the data drifts away from what they were trained on — and an overconfident, miscalibrated model gives no warning. Trustworthy deployment needs calibrated uncertainty, explicit distribution-shift detection, and a way to flag failure when there is no ground truth to check against.
  • Validation Without a Ground TruthValidation asks whether a microscopy pipeline's outputs are biologically true and fit for purpose — but biology rarely supplies a clean answer key. The discipline is choosing metrics that reflect the question, manufacturing ground truth honestly, and conditioning performance on the experiment.
  • Verification and Validation Are Two Different QuestionsVerification asks whether the pipeline was built right — deterministic, reproducible, correct to spec. Validation asks whether it is the right pipeline — outputs that are biologically true and fit for purpose. Microscopy makes both hard, and conflating them is how silently wrong results get shipped.

← Back to the constellation