The QC-Aware Report

The problem — A normal report presents the answer: a profile, a slide-level score, a count matrix. It says nothing about whether the answer should be believed. So a result computed on an out-of-focus field, a batch-confounded plate, or a mis-registered tile looks identical to a good one — the failure is invisible at exactly the moment a decision is made. A QC-aware report inverts the default: it surfaces the evidence of trustworthiness alongside the result, so a reader can see why to trust it, or see immediately that they shouldn't.

What it is / how it works — Structure the report by the four nested quality levels from the quality model, because evidence at one level says nothing about the levels above it:

• Image-space — per-image physics: focus, SNR, illumination uniformity, flagged with reference-free no-reference IQA where no clean target exists.
• Run / sample — consistency across the acquisition: field-to-field drift, batch structure, z-depth decay, registration stability — the level where the provenance vs quality metadata distinction earns its keep.
• Readout — fidelity of the quantity the platform actually reports, judged with metrics that reflect the domain interest (Metrics Reloaded) — segmentation accuracy, percent-replicating, transcript assignment — the only level the science consumes.
• Decision — the operational verdict the report must emit: reportable, needs-review, or reject-and-reacquire.

A QC-aware report ends in that decision, traces it back through the three levels of evidence, and carries the lineage (attached to the output) so the verdict travels with the result.

Where it breaks — The general failure is reporting a single level and implying the rest — almost always image-space scores standing in for readout truth — and the gap is platform-specific. A Cell Painting report showing sharp fields but hiding plate-level batch structure passes a confounded assay. A digital pathology report with acceptable global focus but no per-tile QC ships slide scores from regions a model never should have embedded. A light-sheet report citing PSNR after deconvolution, with no readout-level segmentation check, certifies synthesized structure. A spatial omics report omitting registration error lets a corrupted count matrix look clean. The discipline: report the decision, show the evidence at every level beneath it, and never let a green light at one level imply green lights above — the operational face of verified outputs.

References

• Huisman et al. (2019). A Perspective on Microscopy Metadata: Data Provenance and Quality Control.
• Varoquaux & Cheplygina (2022). Machine Learning for Medical Imaging: Methodological Failures and Recommendations. npj Digital Medicine.
• Maier-Hein, Reinke et al. (2024). Metrics Reloaded: Recommendations for Image Analysis Validation. Nature Methods.