AI in Breast Cancer Screening: A New Era of Detection and Risk Prediction

A series of recent studies and new global guidelines indicate that artificial intelligence (AI) is being evaluated and recommended for use in breast cancer screening to improve detection rates, predict individual risk, and reduce workload pressures on radiologists.

Advancements in AI Detection and Screening

Multiple studies have assessed the integration of AI into breast cancer screening workflows, reporting improvements in cancer detection and reductions in radiologist workload.

Swedish Trial (The Lancet)

In a study involving 100,000 women in Sweden conducted between April 2021 and December 2022, AI-supported mammography was associated with a reduction in subsequent cancer diagnoses.

The AI-supported group had 1.55 cancers per 1,000 women, compared to 1.76 per 1,000 in the standard screening group.

Early detection rates were higher in the AI group (81% vs. 74%), and there was a nearly 27% reduction in aggressive sub-type cancers. The study did not advocate for replacing healthcare professionals, noting that human radiologist involvement remains necessary.

UK NHS Studies (Nature Cancer)

Two major UK studies have reported on AI's impact.

The GEMINI study evaluated 10,889 women in the NHS Grampian region, integrating an AI tool (Mia) into clinical workflows. Using AI as a second reader improved cancer detection by 10.4%, reduced workload by up to 31%, and cut notification times for results from approximately 14 days to just 3 days.

A larger collaborative study involving 175,000 women (Imperial College London, Google, and multiple NHS Trusts) found that AI acting as a second reader improved the cancer detection rate from 7.54 to 9.33 per 1,000 women. It also significantly reduced false positives and, for first-time screens, reduced recalls by 39.3%. The time required to read a scan decreased by 32.1%. In a separate arbitration study, AI performed comparably to human arbitrators in resolving discrepancies between readers.

US AI Triage Tool (Nature Digital Medicine)

Researchers from UC San Francisco and UC Berkeley tested an AI model called Mirai on over 4,100 mammograms at Zuckerberg San Francisco General Hospital. The model identified 12.7% of patients as high-risk.

This triage system reduced the wait for diagnostic evaluation from several weeks to about one hour and reduced the average wait for a biopsy from over two months to fewer than 10 days.

The tool is intended as a triage aid for physicians, not a replacement for radiologists.

AI for Risk Prediction: BRAIx and Updated Guidelines

Two separate developments highlight AI's role in predicting future breast cancer risk.

Australian BRAIx Tool (The Lancet Digital Health)

An Australian-developed AI tool, BRAIx, has been reported to identify women at high risk of breast cancer who were not detected by standard screening. Trained on nearly 500,000 mammograms, the tool provides a personalized risk score.

Research found that 1 in 10 women in the top 2% of risk scores developed breast cancer within four years, despite having clear mammograms.

The tool is designed to assist, not replace, radiologists.

Updated NCCN Guidelines

The National Comprehensive Cancer Network (NCCN), a US-based alliance of 33 cancer treatment centers, has issued new global guidelines recommending the integration of AI into breast cancer screening.

The guidelines state that women should be eligible for AI-powered mammogram risk assessments starting at age 35. The stated aim is to more accurately predict individual breast cancer risk and adjust screening protocols accordingly.

Commentary from Medical Professionals

Medical professionals have commented on the potential and limitations of AI in breast cancer screening.

On the UK and Swedish Studies

"AI-supported mammography could help detect cancers earlier but implementation requires caution, tested tools, and continuous monitoring."
— Dr. Kristina Lång, Lund University, Sweden

Dr. Sowmiya Moorthie (Cancer Research UK) described the findings as promising but advised caution, stating that further research from multiple centers is needed.

Simon Vincent (Breast Cancer Now) stated the trial underscores AI's potential to support radiologists, emphasizing that early detection improves treatment outcomes.

On the Australian BRAIx Tool

Dr. Helen Frazer (St Vincent's BreastScreen) characterized the tool's ability to identify early cancer signals beyond human perception as a breakthrough.

Dr. Wendy Ingman (Adelaide University) noted the tool's advanced capability in defining risk levels compared to other global AI algorithms.

On the Updated NCCN Guidelines

Dr. Raman Narang (MOC Cancer Care and Research Centre, New Delhi) described the new NCCN guidelines as a shift from a 'detection first' to a 'prediction first' approach.

He stated that AI-powered screening could facilitate identification of high-risk patients regardless of family history or genetic markers and could allow for screening without requiring universal testing.

Future Outlook

Researchers in the UK have noted that the findings from these studies are expected to inform the upcoming EDITH trial, an international study designed to evaluate different AI tools in mammography across multiple sites.

Australian researchers have also indicated plans for a prospective study of the BRAIx tool, with a projected rollout within five years.