The Slide Goes Digital: What AI-Assisted Pathology Can—and Cannot—Tell Us About UK Disease

Pathology has long depended on the glass slide. A thin slice of tissue, stained and examined under a microscope, can distinguish benign from malignant, identify an infection or reveal the subtle architecture of a disease. In 2026, that centuries-old workflow is being rebuilt around digital images and artificial intelligence. The phrase ‘digital autopsy’ is often used to describe the shift, but it is more rhetorical than literal: this is not primarily about examining the dead. It is about extracting more information from living tissue, archived samples and the vast records that accompany them.

From glass to pixels

Whole-slide imaging scanners turn glass slides into high-resolution digital files. Pathologists can review them on screens, share them with colleagues, discuss them in multidisciplinary meetings and store them for future research. The practical benefits are easy to see: a specialist in one city can advise a smaller laboratory elsewhere; a second opinion can be obtained without posting fragile slides; and digital archives can be searched and analysed at scale. NHS England has supported digital pathology through its pathology networks, and many trusts have been building the scanners, storage and interoperability needed to make the model work. But coverage is uneven. Digitisation is not simply a matter of buying hardware; it requires training, information governance, reliable networks and systems that can talk to each other.

What AI adds to the microscope

Computational pathology uses algorithms to detect cells, quantify biomarkers, grade tumours and flag suspicious areas. Some tools act as a triage system, ranking slides so that the most urgent cases are reviewed first. Others act as a second reader, highlighting regions that a pathologist might have missed. In breast cancer, prostate cancer and skin pathology, research groups have tested AI on tasks such as counting mitoses, measuring hormone receptors and classifying tissue patterns. The appeal is consistency. Human grading can vary between observers, and AI might reduce that variation—or at least make it more visible.

A UK advance in context

One of the more closely watched UK examples is AI-assisted grading of prostate biopsies. Researchers working with NHS trusts and universities have trained algorithms on thousands of digitised biopsy images to distinguish benign tissue from cancer and to assign Gleason grades. In several evaluations, the tools have shown high agreement with expert pathologists on selected tasks. The potential benefit is not simply speed. Prostate grading is important for treatment decisions, and observer variation can affect care. An AI tool that flags difficult cases or provides a consistent first pass could help pathologists focus attention where it is most needed.

Yet much of the evidence comes from retrospective studies on curated cases. Those studies are valuable, but they are not the same as a prospective trial in a busy NHS laboratory with imperfect slides, variable scanners and a patient population that may differ from the training data. Questions remain about generalisability, false positives, false negatives and the way AI performance changes when tissue preparation or scanning equipment differs. Regulators, including the MHRA, must assess whether a tool is safe and effective before it is used in care. Hospitals must consider cost, workflow, storage and the risk of automation bias—the tendency for humans to defer to a machine’s suggestion.

The archive as a research resource

The digital slide is also a research object. Once tissue is scanned, it can be linked to clinical records, genomic data and outcomes, subject to consent and governance. Large UK cohorts and biobanks are exploring how digitised pathology can deepen understanding of disease progression and treatment response. This is where the ‘autopsy’ metaphor has some force: researchers can return to historical samples and ask new questions of them. But archived tissue is not neutral. It reflects the populations that were sampled, the diagnostic practices of the time and the quality of preservation. More data does not automatically mean better evidence.

Where the promise is strongest

The most credible near-term benefits are practical rather than revolutionary. Digital pathology can extend specialist expertise to understaffed laboratories, reduce the friction of second opinions and support quality assurance. AI can help with repetitive quantification tasks and prioritise workloads. These are meaningful gains in a health service under pressure. The weaker claims are the grand ones: that AI will soon diagnose disease autonomously or replace the pathologist’s judgement. Pathology is not just pattern recognition. It involves context, uncertainty and communication with clinicians and patients.

Evidence gaps and hype

For patients and the public, the key questions are about evidence and governance. Has the tool been evaluated in a population like mine? Who is accountable when it errs? Is the data used to train it representative? Can its decisions be explained? Is there a route to complain or seek a second opinion? Digital pathology also raises practical concerns: cybersecurity, long-term storage costs, interoperability between vendors and the environmental footprint of data centres. None of these is a reason to reject the technology. They are reasons to adopt it carefully.

What patients should ask

If you are told that AI is involved in your pathology, it is reasonable to ask whether it is used for triage, diagnosis or quality control; whether a pathologist reviews the final report; whether the tool is approved for NHS use; and how your data is governed. Transparency should not be an afterthought. Nor should patient and public involvement: the people whose tissue and data make this research possible deserve a say in how it is used.

A measured revolution

Digitised pathology is likely to change diagnosis and research in the UK, but it will do so incrementally. The most useful stance is measured optimism: recognise the genuine potential, insist on prospective evidence and robust regulation, and keep human judgement at the centre. The slide may have gone digital, but the need for careful interpretation has not.