AI helps spot early warning signs of breast cancer sooner

By Stephen Beech

Early warning signs of breast cancer could be spotted up to six years in advance using AI, suggests a new study.

The latest artificial intelligence technology can provide an "early alert" for the potentially deadly disease years before a diagnosis, say scientists.

Three commercially available radiology AI systems have shown the potential to flag early signs of breast cancer up to six years before a diagnosis, according to the research.

Swedish researchers tested three AI-based computer-assisted detection (AI-CAD) systems on mammogram data from a large screening population.

The findings, published in the journal Radiology, showed that cancer prediction scores issued by AI-CAD were elevated, on average, for people who were eventually diagnosed with breast cancer, while scores were low for those who remained cancer-free.

Senior co-author Fredrik Strand, of Karolinska University Hospital, Stockholm, said: "Approximately 20% of breast cancer cases demonstrate mammographic signs that are already visible to AI around six years before diagnosis.

"Our study confirms the potential of AI to, in some cases, find signs of cancer in the mammograms much earlier than when radiologists detected it."

AI-based systems have shown promise for predicting five-year risk of breast cancer and identifying women at risk of "interval" cancers between regular screening mammograms.

Strand's team investigated AI's potential to flag mammographic signs that were present up to 10 years in retrospect.

More from this section

+6

‘Intelligent tattoo’ detects deadliest skin cancer before it appears

World Cup creates perfect conditions for infectious diseases to spread – here are the biggest threats health experts are watching for

+3

Glass of wine, beer daily enough to raise cancer, heart disease risk: study

The research team included a total of 88,963 mammograms performed on 31,394 patients across a period of 10 years for the study.

Data were taken from the Validation of Artificial Intelligence for Breast Imaging (VAI-B) database, which collects breast imaging data from volunteers across four regions of Sweden.

The Swedish national breast screening program invites women between the ages of 40 and 74 to participate in screening examinations every two years, and each mammogram has traditionally been read by two radiologists.

Strand's team applied three commercially available AI-CAD systems to mammograms taken between January 2008 and April 2019.

Across that period, 12,072 of the participants (38.5%) were diagnosed with cancer by radiologist readers.

The AI-CAD systems successfully identified many of those cancers at earlier screening points.

It achieved 90% "specificity" — distinguishing between a true positive and a true negative result — in up to 19.7% of participants six years before their recorded diagnosis, up to 25.2% of individuals four years before diagnosis and up to 39.3% two years before diagnosis.

The researchers say AI-CAD scores could help radiologists spot early mammographic signs of potential future cancers.

And a personalized approach to interpreting individual scores could help identify patients who might benefit from closer vigilance.

Strand said: "This study aims to add to the growing literature regarding the application of AI in breast cancer screening and how it can help play a role in earlier detection of breast cancer."

He added: "Analyzing the AI scores of screened individuals over time could provide insight into how early detectable changes arise, potentially allowing for earlier intervention."