The results of the largest clinical trial of AI-assisted breast cancer screening were published this week in The Lancet Oncology, and they are unambiguously positive.
The trial, conducted across 17 hospitals in Sweden, the Netherlands, Germany, and the UK, enrolled 104,000 women between 2023 and 2025. Half received standard double-reading mammography screening (where two radiologists independently review each scan). The other half received AI-assisted screening, where a deep learning system called BreastScreen-AI pre-analysed each scan and flagged areas of concern before a single radiologist reviewed the results.
The headline finding: the AI-assisted group had an 11.3% higher cancer detection rate than the double-reading group, with no statistically significant increase in false positives. In practical terms, this means the AI system found cancers that two trained radiologists, working independently, missed.
Critically, the cancers detected by the AI-assisted pathway were disproportionately small and early-stage - precisely the cancers where early detection most improves survival. Of the additional cancers found, 73% were Stage I or Stage II.
"This is what AI in medicine is supposed to look like," said Professor Maria Eriksson of the Karolinska Institute, the trial's lead investigator. "Not replacement. Augmentation. The AI sees patterns in the pixel data that the human eye cannot reliably detect at 3am after reviewing 200 scans."
The trial also measured radiologist workload. In the AI-assisted arm, radiologists spent 44% less time per scan on average, because the AI pre-sorted cases by likelihood of abnormality. Low-risk scans could be reviewed faster. High-risk scans received more attention.
BreastScreen-AI is not a black box. Its outputs include heat maps showing which areas of the mammogram triggered the flag, allowing the radiologist to evaluate the AI's reasoning rather than simply accepting or rejecting a binary recommendation.
Several European national health services are now evaluating the system for inclusion in population screening programmes.
What we know for certain
A 104,000-patient clinical trial published in The Lancet Oncology found AI-assisted mammography screening detected 11.3% more cancers than standard double-reading, with no increase in false positives. 73% of additional cancers found were early-stage.
What we are inferring
If these results replicate at national scale, AI-assisted screening could meaningfully reduce breast cancer mortality in population screening programmes. The model's interpretable outputs reduce the 'black box' risk seen in other medical AI deployments.
What we couldn't verify
Long-term survival outcomes for patients whose cancers were detected by the AI but missed by radiologists. Five-year follow-up data is not yet available.