Cancer · Diagnosis and imaging
AI help for reading mammograms
In a large Swedish trial (MASAI), mammogram screening supported by AI found more cancers than standard reading by two radiologists, and cut radiologists' reading workload by about 44%. It's strong trial evidence, but this way of screening isn't yet routine in most screening programmes.
How far away is this?
This is a randomised trial in a real screening programme, which is strong evidence. We haven't rated it 'Approved & in use' here because our page is about this screening approach as tested, and the results come from one country's programme and workflow.
Where is it approved?
| Region | Regulator | Status | Notes |
|---|---|---|---|
| United States | FDA | Cleared | The AI used in MASAI is Transpara (ScreenPoint Medical). FDA 510(k) clearance K241831 for Transpara 2.1.0, 25 November 2024. MASAI used an earlier version.[5] |
| European Union | CE mark / EMA | CE marked | Manufacturer states Transpara is CE certified (notified body 0344). We haven’t seen the EUDAMED record.[6] |
| United Kingdom | MHRA | Unknown | |
| Australia | TGA | Unknown |
MASAI tested an AI-supported way of screening (using Transpara software) inside Sweden’s programme. The software itself is cleared in the US and CE marked; that doesn’t mean this screening method is routine. UK and Australia are Unknown. "Unknown" means we haven't verified it, not that it's unavailable. Approval in one country isn't approval in another.
What is it?
In many screening programmes, each mammogram is read by two radiologists. In the MASAI trial, AI scored each mammogram for risk; low-risk scans got one reader and high-risk scans got two, with the AI flagging suspicious spots.
What's the evidence?
MASAI randomly assigned 105,934 women in Sweden's screening programme (April 2021 to December 2022) to AI-supported screening or standard double reading.[1]
Evidence stage: In clinical trials · #c1
AI-supported screening detected 6.4 cancers per 1,000 women screened, versus 5.0 per 1,000 with standard reading. That's about 1.4 extra cancers found per 1,000 women.[1]
Evidence stage: In clinical trials · #c2
Recall and false-positive rates were not significantly higher. The number of screen readings radiologists did fell by 44.2%; that's a benefit for the screening service, not a patient outcome.[1]
Evidence stage: In clinical trials · #c3
The trial's main measure was cancers diagnosed between screens (interval cancers): 1.55 per 1,000 with AI versus 1.76 per 1,000 without. AI-supported screening was non-inferior (not worse), which is what the trial was designed to test; the small difference between the two figures should not be read as proof that AI was better. Sensitivity was 80.5% with AI versus 73.8%, and specificity was 98.5% in both groups.[2]
Evidence stage: In clinical trials · #c4
An accompanying editorial notes results depend on how the AI is built into the reading workflow, so other programmes may see different results.[3]
Evidence stage: In clinical trials · #c5
What does it mean for me?
If you're having screening mammograms, nothing changes for you today. Keep going to your screens as recommended. If programmes adopt AI, the aim is to catch more cancers early without more call-backs, and to free up radiologists' time. Finding more cancers is only helpful if they're ones that matter: some screen-detected cancers would never have caused harm (overdiagnosis).
What could I ask my doctor?
- Does the screening service I use use any AI to help read mammograms?
- If AI is used, does a radiologist still look at my scans?
- What would a call-back mean, and what happens next?
These are general starting points, not a recommendation to choose any particular test or treatment.
If you're worried about your health right now, contact your doctor or local health advice line. In an emergency, contact your local emergency number.
Change history
- : First published.
- : Regulator status checked against official records; see Where is it approved?
- : Updated after independent fact-check and clinical review
- : Review label corrected: this page was not clinically reviewed. Earlier labels saying 'independent fact-check' and 'clinical review' were wrong; checks so far were AI-assisted source checks only.
General information only, not medical advice. Talk to your doctor about your situation. AI-assisted source check. Not clinically reviewed.
References
- Hernström et al., The Lancet Digital Health (2025): MASAI screening performance. thelancet.com (2025). Peer-reviewed randomised trial. https://www.thelancet.com/journals/landig/article/PIIS2589-7500(24)00267-X/fulltext
- Gommers et al., The Lancet (2026): MASAI interval cancer results. thelancet.com (2026). Peer-reviewed randomised trial. https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(25)02464-X/abstract
- Editorial: AI for mammography, The Lancet Digital Health. thelancet.com. Editorial. https://www.thelancet.com/journals/landig/article/PIIS2589-7500(25)00004-4/fulltext
- ClinicalTrials.gov NCT04838756. clinicaltrials.gov. Trial registry. https://clinicaltrials.gov/study/NCT04838756
- FDA record or statement: United States. accessdata.fda.gov. Regulatory or manufacturer record. https://www.accessdata.fda.gov/cdrh_docs/pdf24/K241831.pdf
- CE mark / EMA record or statement: European Union. screenpoint-medical.com. Regulatory or manufacturer record. https://screenpoint-medical.com/insights/transpara-regulatory-clearances
Publisher shown as the website hosting each source. Dates are given where the source label includes them.