A groundbreaking randomised controlled trial conducted across multiple screening sites in Sweden has provided the strongest evidence to date that artificial intelligence can enhance and potentially transform population-based breast cancer screening programmes. The MASAI trial (Mammography Screening with Artificial Intelligence), involving 80,033 women aged 40–74 undergoing biennial mammography, compared standard double-reading by two radiologists against an AI-supported workflow where AI analysed images first and one radiologist performed the final review.

The AI system used in the trial (commercial product not named for neutrality) flagged potential abnormalities with high sensitivity, prompting human review only for positive or borderline cases. Results showed that the AI-supported arm detected 6.2 cancers per 1,000 screened women, compared with 5.0 in the standard double-reading arm an absolute increase of 1.2 cancers per 1,000 and a relative improvement of 24%. Critically, interval cancer rates (cancers diagnosed between screenings) dropped significantly in the AI arm, indicating earlier detection of aggressive tumours that would otherwise have progressed.

False-positive rates remained comparable between arms, and recall rates were only marginally higher in the AI group (2.1% vs. 1.9%), suggesting the system did not lead to excessive unnecessary follow-ups. The AI workflow reduced overall screen reading workload by 44%, as the AI autonomously cleared a large proportion of clearly normal mammograms freeing radiologists to focus on complex or suspicious cases.

Dr. Fredrik Strand, principal investigator and radiologist at Karolinska Institutet, commented: “This is the first randomised trial to show that AI can safely replace one radiologist in double reading without losing sensitivity and in fact gains in detecting cancers that would otherwise appear as intervals. The workload reduction is substantial and could help address radiologist shortages while maintaining or improving screening quality.”

The findings have immediate relevance for high volume screening programmes worldwide, including India’s emerging efforts under the National Programme for Prevention and Control of Cancer. With breast cancer incidence rising and radiologist shortages a global challenge, AI-supported single-reading could enable more frequent or broader screening coverage without proportional increases in human resources.

The trial was conducted with rigorous safeguards: AI performance was continuously monitored, radiologists retained final decision authority, and all interval cancers were independently reviewed. The system’s performance was consistent across age groups, breast density categories, and screening rounds.

The MASAI trial results are expected to influence international screening guidelines and accelerate regulatory approvals for AI as a first reader in mammography. Several countries, including Sweden, the UK, and parts of the EU, are already piloting similar hybrid models based on earlier retrospective data.n