A recent study published in The Lancet Digital Health highlights that AI can significantly enhance breast cancer screening by detecting signs that radiologists might miss, potentially leading to earlier diagnosis.

The research found that a majority of invasive breast cancers (68.9%) lacked retrospective abnormalities, yet those cases with higher AI scores were more likely to be missed by radiologists, demonstrating AI's ability to identify subtle indicators.

AI systems, such as Transpara v.1.7.0 evaluated in the study, showed promise in replacing the second reader in mammography screening, which could improve detection rates and reduce missed cancers.

The retrospective analysis involved 119 mammograms from women with high breast density, averaging 57.3 years old, and focused on invasive breast cancers diagnosed within 24 months after a normal mammogram.

Three radiologists reviewed the mammograms, classifying cancers based on retrospective abnormalities, while AI generated scores indicating the likelihood of current and future breast cancers.

Integrating AI with human reading increased the sensitivity of screening from 2.9% to 8.4%, leading to more cancers being detected earlier.

Most cancers identified by AI, including invasive and larger tumors, were detected at more advanced stages, but AI also identified additional invasive cancers and tumors larger than 20 mm that were missed by human readers.

AI was particularly effective at detecting future-detected and interval cancers, suggesting earlier intervention opportunities.

The study analyzed over 42,000 mammograms from more than 42,000 women, with 580 cases labeled positive, including screen-detected, interval, and future-detected cancers.

Experts emphasize the need for an arbitration process for cases flagged by both AI and radiologists and call for further research involving diverse AI systems and datasets.

Limitations of the study include its retrospective design and limited access to detailed diagnostic mammograms, which restricts causal inferences.

Funded by the Cancer League of Eastern Switzerland, the study was led by Dr. Ritse Mann from Radboud University Medical Center, evaluating AI's performance in breast cancer screening.