ideasIdea
Set each woman's mammogram interval from her last mammogram, using AI risk
Instead of every woman every two or three years, an AI reading of the current mammogram would set who comes back in one year and who can safely wait four.
Image-based risk models (Mirai and commercial density-plus-AI scores) outperform Tyrer-Cuzick for five-year risk. MyPeBS and WISDOM test risk-based intervals using classical risk plus polygenic scores. Propose a national programme trial where the AI score on the index mammogram assigns one-, three-, or four-year recall, with interval cancer rate and advanced-cancer incidence as endpoints.
Hypothesis
AI-image-based intervals reduce stage II+ cancer incidence by at least 15% at equal or lower total mammogram volume compared with fixed intervals.
Rationale
Most interval cancers occur in a small high-risk stratum identifiable on the prior image; most low-risk women get no benefit from frequent recall.
What would test it
Cluster-randomised comparison within an existing programme (e.g. NHS Breast Screening Programme sites) of fixed versus AI-assigned intervals; five-year advanced-cancer endpoint.
Maturity
early clinical
Who has to act
clinic
Cost to try
Medium ($1M to $50M)
Years to first evidence
6
Bottlenecks it attacks
- Most lethal cancers are found late · Screening exists for only a few cancers. Pancreatic, ovarian, liver, oesophageal and most lung cancers are found when cure is unlikely.
- Overdiagnosis and false alarms · Finding more cancer is not the same as saving lives. Screening also finds cancers that would never have hurt anyone, and treats them.