Will AI replace radiologists? What the evidence shows
AI is widely deployed in radiology as a second reader and triage tool, and large trials show it can find more breast cancers while cutting reading workload by about 44%. In every major deployment radiologists still read and sign off, and demand for radiologists has grown since AI pioneers predicted their obsolescence in 2016.
The job today
Task by task: what AI can actually do
In real use means deployed at scale; shown in tests or pilots means demonstrated but not yet routine; not yet means we found no credible evidence it does this reliably.
| Task | Status | Evidence |
|---|---|---|
| Flagging suspicious findings on scans | In real use | 1,094 of the 1,430 AI medical devices the FDA authorized through Dec 2025, about 76.5%, are for radiology. (AuntMinnie, Jul 2026) |
| Breast cancer screening support | In real use | In Germany's screening program (460,000+ women, 12 sites), radiologists using AI detected 17.6% more cancers without raising recall rates. (Nature Medicine via Pharmacy Times, Jan 2025) |
| Triaging screens to cut reading workload | Shown in tests or pilots | Sweden's MASAI randomized trial (106,000 women) cut radiologists' screen-reading workload 44% and interval cancers 12%; radiologists still read every mammogram. (The Lancet via Lund University, Jan 2026) |
| Drafting reports | Shown in tests or pilots | At Northwestern Medicine, AI-drafted X-ray reports cut documentation time about 15.5%, with no accuracy difference on 800 peer-reviewed exams; radiologists review and finalise each draft (Jun 2025). (Radiology Business, Jun 2025) |
| Reporting normal chest X-rays with no radiologist | Shown in tests or pilots | Oxipit's ChestLink got a European CE mark in Mar 2022 to report likely-normal chest X-rays autonomously, after vendor-reported pilots of 500,000 images; the report did not mention US FDA clearance. (AuntMinnie, Mar 2022) |
| Final diagnosis, consultation and procedures | Not yet | Mayo Clinic's radiologist numbers grew 55% after Geoffrey Hinton's 2016 prediction, while it adopted more than 250 AI models (NYT, May 2025). (TechCrunch, May 2025) |
Jobs evidence so far
Labelled by who says AI was the reason: the company itself, reporters, or a study.
- Mayo Clinic radiologist headcount up 55% since 2016; AAMC projects a shortage of up to 42,000 radiologists and other specialists by 2033 (reported) (TechCrunch, May 2025)
- Neiman Health Policy Institute projects US imaging use to grow 17-27% by 2055 and warns against downplaying AI's challenges in imaging (study) (AuntMinnie, Oct 2026)
- BLS projects radiologist employment to grow 3% in 2025-35 (study) (BLS OOH, 2025-35 projections)
What to do
- Learn to work with AI triageScreening programs are moving to AI-assisted single reading; understanding how the tools fail is part of safe practice.
- Lean into what AI can't doInterventional procedures, multidisciplinary consults and talking to patients and referrers rely on judgment and presence.
- Get involved in AI governanceMayo built a 40-person AI team inside radiology; hospitals need radiologists who can evaluate and monitor models.
- Watch autonomous-reading approvalsThe signal to track is regulators clearing AI to sign off normal studies without a radiologist, as Europe did for chest X-rays in 2022.
Questions people ask
Will AI replace radiologists?
Not on current evidence. AI is widely used but alongside radiologists, demand for imaging is rising, and BLS projects 3% growth to 2035.
What does AI do in radiology today?
It flags findings, triages screening mammograms, measures structures and drafts reports, with a radiologist reviewing and signing off.
Does AI make breast screening better?
Large studies say yes: 17.6% higher detection in Germany's real-world rollout and 12% fewer interval cancers in Sweden's MASAI trial, with a 44% lower reading workload.
Can AI read scans with no radiologist?
Only in narrow cases: one European-certified tool (CE mark, 2022) reports likely-normal chest X-rays autonomously. We did not verify any US equivalent.
Was Geoffrey Hinton wrong about radiologists?
So far, yes: in 2016 he said it was 'completely obvious' AI would replace them, but radiologist numbers have grown since.
How we grade
Each task is graded 'deployed' only when a named organisation reports AI doing it in real work at scale, 'demonstrated' when it works in tests, benchmarks or pilots, and 'not yet' when we found no credible evidence AI does it reliably. For job losses we separate what a company said in its own words ('company says AI') from what reporters or analysts attributed to AI ('reported as AI') and from research ('study'); employment and outlook figures come from the BLS 2025-35 projections.
Sources
- BLS Occupational Outlook Handbook: Physicians and surgeons (radiologists)
- AuntMinnie: radiology dominates FDA AI approvals
- Lund University: MASAI final results
- Applied Radiology: MASAI interim results
- Pharmacy Times: PRAIM real-world study
- Radiology Business: Northwestern AI report drafting
- AuntMinnie: Oxipit CE mark
- TechCrunch: radiologists aren't going anywhere
- AuntMinnie: HPI on radiology workforce
Spot an error or something out of date? Tell us; corrections are made and logged. Written with AI assistance from public sources; every fact checked and approved by Colin Balfe. General information, not legal, medical or financial advice.