Seeing the Same Thing Twice: Colour Assurance Before AI Enters UK Digital Pathology

A digital slide is an image, but in diagnostic use it is evidence. When a pathologist signs out from a screen, the colour on that screen must be traceable to the stain on the glass, the scanner that captured it and the display that renders it. In 2026, as NHS pathology networks consolidate reporting across sites and whole-slide imaging moves closer to routine primary diagnosis, that traceability is becoming an operational requirement rather than a technical nicety.

Colour is not a cosmetic setting

In histopathology, colour carries signal. Haematoxylin and eosin staining depends on subtle differences in nuclear chromatin, cytoplasmic density, collagen and mucin. Special stains add further colour-coded meaning. A shift in white balance or display gamma may look minor to the eye, but it can alter the apparent contrast between structures that a pathologist uses to judge grade, margin or atypia.

For artificial intelligence, the issue is sharper still. Many image-analysis models learn colour and texture correlations alongside morphology. If a model is validated on images from one scanner and deployed on another, colour differences can change its outputs without any change in the underlying tissue. That is not a hypothetical concern; it is a validation gap that UK laboratories need to close before digital slides become the primary record.

Where variation enters the chain

Colour variation is rarely caused by one component. It accumulates across the pre-analytical, analytical and display chain. Fixation time, section thickness, stain batch, reagent ageing and coverslip quality all influence the final colour. The scanner adds its own light source spectrum, sensor response, white balance, focus and compression. Software may or may not respect embedded colour profiles, and viewers can render the same file differently. The display then introduces brightness, contrast and ambient-light effects.

In a single laboratory, these variables can be controlled. In a pathology network spanning several sites, they become a governance problem. Two scanners from the same vendor may be configured differently; two sites may use different stain protocols; two pathologists may view the same slide on monitors calibrated to different targets. Without an explicit colour-management plan, the network has no reliable way to distinguish biological variation from instrument bias.

The validation gap

ISO 15189 requires laboratories to validate methods, equipment and changes that affect examination quality. Digital pathology does not suspend that duty. A validation dossier that records scanner concordance but not colour handling is incomplete. The same applies to AI tools regulated as medical devices: evidence of performance on a single scanner or a single stain batch is not evidence of transferability across a UK network.

UKAS accreditation assessors and MHRA device reviewers are increasingly likely to ask how colour was controlled during validation and how it will be maintained after deployment. The answer should not be a vendor brochure. It should be a documented chain of custody for colour: reference slides, calibration targets, profile settings, display checks and change-control records.

A practical UK assurance checklist

A UK-wide colour standard need not invent new physics. It needs to make existing colour-management practice explicit, auditable and contractual. Laboratories, vendors and commissioners can begin with the following questions.

• Is the colour-controlled chain defined end to end, including scanner, viewer, display and ambient light?
• Are reference slides or calibration targets used at a defined frequency, and are results recorded?
• Are ICC or DICOM colour profiles, viewer settings and display calibration records held under version control?
• Are displays checked with a photometer against a documented target, and is ambient light measured in reporting rooms?
• Does AI validation include multi-scanner, multi-site and multi-stain-batch testing, with performance reported by subgroup?
• Do scanner firmware updates, viewer upgrades and display replacements trigger a documented revalidation decision?
• Are colour acceptance criteria written into procurement specifications and service contracts?
• Is cross-site concordance audited using a challenge set of stained slides rather than a single site's routine work?
• Are staff trained to recognise and report display or ambient-light problems?
• Are colour-related deviations captured in the laboratory's CAPA system?

What commissioners should specify

Commissioners should resist reducing digital pathology to throughput, storage and image quality scores. A network contract should require a colour-management plan, calibration evidence, validation across participating sites and clear arrangements for exporting images with their colour profiles intact. Vendor-neutral archives matter here: if images cannot be rendered consistently after a system change, the diagnostic record is weaker than it appears.

The practical test is simple. Can the same stained slide be scanned at two sites, viewed on two calibrated displays and reported with the same diagnostic confidence? If the answer is unclear, the network does not yet have a colour standard. It has a collection of screens.

Conclusion

Digital pathology will not fail in the UK because pathologists cannot read screens. It will fail quietly if colour is treated as a local preference rather than a shared assurance property. Before AI reads the slides at scale, UK laboratories need a colour standard that is visible in validation reports, procurement documents and accreditation evidence. The goal is not perfect colour. It is reproducible evidence.