Skip to content
Ruling 36 of 61

Work and careers · The ruling

Can AI replace radiologists?

Verdict

No

The verdict

No. In the UK, AI helps radiologists with narrow jobs like flagging fractures, and a human expert still reads the scan and makes the call.

No means: no, or not in a way you should rely on. How we rule

An engraved illustration of a blank sheet of paper with a round terracotta seal, a fountain pen, an envelope and reading glasses on a wooden desk.
Plate A sealed sheet of paper, because every question page ends in a ruling.

Short answer: wherever UK bodies describe AI in radiology, it is helping with one defined job, and a person still reads the image and makes the decision. That holds for the fracture tools NICE has assessed and for EDITH, the government’s AI breast screening trial.

Start with who does the job. A consultant radiologist is a doctor. To be appointed to an NHS consultant post, a doctor has to be on the GMC’s Specialist Register. Foundation trusts are an exception in law, but the register is still the usual route. An AI tool isn’t on that register, and nothing in the sources gives it a route onto it.

The clearest official verdict on a specific tool is NICE’s early value assessment of AI for spotting fractures on X-rays in urgent care, published in January 2025. NICE said four tools (Rayvolve, TechCare Alert, BoneView and RBfracture) can be used in the NHS as options to help healthcare professionals detect fractures, with conditions attached. Evidence has to be gathered while they are used. They need the right regulatory approval, including NHS England’s Digital Technology Assessment Criteria. NICE will look at the evidence again after two years. A fifth tool, qMSK, was limited to research. NICE called the clinical risk low because the tools are used alongside standard care, where healthcare professionals make the treatment decisions, and because they do not replace the definitive radiology review. That is the official description: a second check, with the radiologist’s report still to follow.

Breast screening is the other big test. In February 2025 the government announced the EDITH trial, funded with £11 million through the NIHR and involving nearly 700,000 women. The trial registry entry, registered later in 2025, plans for 660,000 women at about 30 screening centres. Today two NHS experts read each mammogram. The trial is trying two ways of adding AI: one where the AI looks after the human reader for cancers they may have missed, and one where the AI first sorts images into higher and lower risk. In every arm, a human expert reads every image, and humans make the final decisions. The trial is due to run until July 2029. The government’s stated hope is that one specialist could do what currently needs two. The trial was set up to find out whether that is safe, so it hasn’t shown it yet.

The pattern holds across both. Approval covers a narrow job, such as flagging a likely fracture or giving a mammogram a second look, on a particular kind of image. A radiologist’s report covers the whole image, the clinical question and what should happen next.

How to do it well

If AI is used on your scan, you’re entitled to understand how. These questions are reasonable to ask the team looking after you:

  1. Will a radiologist or another qualified professional still review and report my scan? In the setups NICE and EDITH describe, the answer is yes.
  2. What is the AI doing? Flagging possible findings, putting urgent scans first or acting as a second reader are different jobs. The answer tells you where the human check comes in.
  3. What happens if the AI and the clinician disagree? NICE says hospitals need clear local protocols for exactly this, so staff know what to do.
  4. Am I in a trial? For EDITH, women are told about the study in their screening invitation letter. The registry says AI may flag more areas than human readers, so more women may be called back for extra tests that often find nothing wrong.

If you want to understand a report you’ve already had, ask the clinician who requested the scan to go through it with you. A general chatbot can explain what a medical term means in general. That is a different thing from reading your image, and it doesn’t replace a clinician’s report.

Rules and risks

Regulation. The MHRA says many software and AI products in health and care are regulated as medical devices. Its guidance on standalone software lists diagnosing a disease or injury as a medical purpose. For people using an app themselves, its advice is that an app with a medical purpose should be UKCA or CE marked. If something goes wrong with medical software, you can report it to the MHRA through the Yellow Card scheme.

Accountability. In the setups NICE and EDITH describe, healthcare professionals make the decisions and AI supports them. Who is legally liable when a tool contributes to a missed finding is a different question. None of these sources answers it, and the answer depends on the facts of the case. If you’re dealing with a real case, read using AI for legal advice and then speak to a solicitor.

Your data. UK GDPR treats health data as special category data and gives it extra protection. A scan and its report count. The ICO has guidance on how data protection applies to AI, written for organisations. Before you upload a scan or report to any chatbot, check what the service keeps and who can see it.

Workforce. The Royal College of Radiologists4’ 2025 census reports a 32% shortfall of clinical radiology consultants across the UK, up from 29% in 2024. That is more than 2,300 doctors. It forecasts 40% by 2030 if nothing changes. The same report says 75% of radiology departments now use AI. Over half use it to help read images, but nearly three-quarters of those say it has had no effect on their workload. The RCR wants AI use in radiology to be led by radiologists and backed by clear evidence. In its view, tools that aren’t proven to improve care should not be used.

What the sources do not settle

None of these sources gives a reliable figure for how many radiologists the NHS will need in the long run, or a date when AI might take over the job. Predictions like that come from opinion pieces, not from these sources. NICE lists open questions about the fracture tools. They include how accurate professionals are with and without AI, how often the software fails, whether it helps or hinders spotting problems other than fractures, and how well it works in different groups of patients. EDITH hasn’t reported yet. For now, the honest position is that AI helps with parts of the job, and the evidence on how much it helps is still coming in.

This is general information, not medical or legal advice.

Sources

9
  1. nice.org.uk1NICE HTG739: AI technologies to help detect fractures on X-rays in urgent care (early value assessment), Recommendations
  2. gov.uk2GOV.UK (DHSC): World-leading AI trial to tackle breast cancer launched
  3. isrctn.com3ISRCTN81384017: EDITH, AI to help healthcare professionals detect cancer in the UK breast screening programme
  4. rcr.ac.uk4Royal College of Radiologists: 2025 Clinical radiology workforce census report (PDF)
  5. gov.uk5MHRA: Software and artificial intelligence (AI) as a medical device
  6. assets.publishing.service.gov.uk6MHRA: Medical device stand-alone software including apps (PDF)
  7. gmc-uk.org7GMC: The Specialist Register
  8. ico.org.uk8ICO: What is special category data?
  9. ico.org.uk9ICO: Guidance on AI and data protection

FAQ

Questions people ask

Can AI read my scan instead of a radiologist in the NHS?

Not on its own, going by what NICE and the EDITH breast screening trial describe. NICE says the fracture tools it assessed are used in addition to standard care and do not replace the definitive radiology review. In EDITH, every image is read by a human expert. If you want to know how your own scan is handled, ask the team caring for you.

Is AI better than radiologists at spotting cancer or fractures?

The UK sources don't say so. For fractures, NICE says the evidence suggests AI may improve detection in urgent care, but it lets the tools be used only while more evidence is gathered, including on accuracy with and without AI. For breast screening, the EDITH trial is comparing AI-assisted reading with two human readers, and the trial record shows no results yet.

Who is responsible if an AI tool misses something on a scan?

The sources show professionals making the decisions: NICE describes healthcare professionals making treatment decisions, and EDITH leaves final decisions to human experts. Who is legally liable in a particular case is something they don't settle, and it depends on the facts. For a real case, get advice from a solicitor.

Is the UK short of radiologists, and will AI fill the gap?

The Royal College of Radiologists' 2025 census reports a 32% shortfall of clinical radiology consultants, more than 2,300 doctors, and forecasts 40% by 2030 if nothing changes. It also reports that most departments using AI to help read images say it has had no effect on their workload. Whether AI will close the gap is still an open question.

Can I upload my own scan to ChatGPT to get a second opinion?

It is not a substitute for a clinician's report, and this page does not suggest doing it. A scan is health data, which UK GDPR gives extra protection, and the MHRA says an app used for a medical purpose should be UKCA or CE marked. If something about your scan worries you, ask the clinician who requested it.

General information for UK readers, not legal, financial or medical advice. How verdicts are decided.