TL;DR
A Swedish trial of 100,000 women found that AI-supported breast cancer screening reduced later cancer diagnoses by 12% and detected 81% of cases at the screening stage compared to 74% with standard reading. The AI also identified 27% fewer aggressive cancers.
Landmark Trial Results
The largest trial to date examining AI in cancer screening has produced striking results. Published in The Lancet, the research involved 100,000 women in Sweden who were randomly assigned to either AI-supported screening or standard double-reading by radiologists between April 2021 and December 2022.
The AI system analysed mammograms and assigned low-risk cases to a single reading while flagging high-risk cases for double review. It also highlighted suspicious findings to support radiologists in their assessment.
The Numbers
The results show meaningful improvements across key metrics:
- 12% reduction in cancer diagnoses in subsequent years (1.55 vs 1.76 cancers per 1,000 women)
- 81% of cancers detected at screening stage with AI support, versus 74% with standard reading
- 27% fewer aggressive subtype cancers in the AI-supported group
Cautious Optimism
Dr Kristina Lång from Lund University, the study’s lead author, said AI-supported mammography could reduce radiologist workload while improving early detection. However, she stressed that implementation must be cautious, with continuous monitoring of how AI influences different screening programmes.
Cancer Research UK’s Dr Sowmiya Moorthie called the findings promising but noted they come from a single centre. NHS trials launched last year will help determine the safest approach for UK implementation.
The research does not support replacing healthcare professionals with AI—screening still requires at least one human radiologist working alongside the technology.