AI algorithms outperform standard clinical risk models to predict breast cancer risk.

In a large study of thousands of mammograms, artificial intelligence (AI) algorithms outperformed standard clinical risk models for predicting five-year breast cancer risk. The research results are RadiologyJournal of the Radiological Society of North America (RSNA).

A woman’s breast cancer risk is typically calculated using clinical models such as the Breast Cancer Surveillance Consortium (BCSC) risk model. The model uses self-reported and other information about the patient, such as age, family history, and breast cancer status. having given birth and whether she has dense breasts – to calculate the risk score.

Clinical risk models rely on the collection of information from a variety of sources, but information is not always available or collected. Recent advances in AI deep learning have made it possible to extract hundreds to thousands of additional mammography features. “

Vignesh A. Arras, M.D., Chief Scientist, Kaiser Permanente Northern California research scientist and practicing radiologist

In this retrospective study, Dr. Arasu used data related to a negative (not showing visible evidence of cancer) 2D mammography screening performed at Kaiser Permanente in Northern California in 2016. Of the 324,009 women who underwent screening in 2016 and met the eligibility criteria, a random subcohort of 13,628 women was selected for analysis. In addition, all 4,584 patients in the eligible pool who were diagnosed with cancer within 5 years of their first mammogram in 2016 were also examined. All women were followed up to 2021.

“We selected from the entire year of mammograms performed in 2016, so our study population is representative of the Northern California community,” Dr. Alas said.

The researchers divided the five-year study period into three periods. One is the interval cancer risk, or the accidental cancer risk diagnosed between years 0 and he 1 year. Future cancer risk, or incidence of cancer diagnosed within 1 to 5 years. and any cancer risk or incident cancer diagnosed between years 0 and 5.

Using the 2016 screening mammograms, 5-year breast cancer risk scores were generated by 5 AI algorithms, including 2 academic and 3 commercial algorithms used by researchers. Risk scores were then compared to each other and to the BCSC clinical risk score.

“All five AI algorithms performed better than the BCSC risk model when predicting breast cancer risk from years 0 to 5,” said Dr. Arasu. “This strong predictive performance over five years suggests that AI is identifying both missed cancers and breast tissue features that help predict future cancer development. Something in the AI ​​will allow us to track breast cancer risk: this is the “black box” of AI. “

Some of the AI ​​algorithms were good at predicting patients at high risk for interval cancer. Interval cancer is often progressive and may require his second mammogram reading, additional screening, or short-term follow-up imaging. When assessing her 10% of women at highest risk as an example, AI predicted up to 28% of cancers, compared with 21% predicted by BCSC.

Even AI algorithms trained for short periods of time (as short as 3 months) were able to predict future cancer risk for up to 5 years if mammography was clinically cancer-free. Using AI in combination with the BCSC risk model further improved cancer prediction.

“We are looking for an accurate, efficient and scalable tool for understanding breast cancer risk in women,” said Dr. Alas. “His mammography-based AI risk model offers practical advantages over traditional clinical risk models because it uses a single data source, the mammogram itself.”

Dr. Aras said some facilities are already using AI to help radiologists detect cancer on mammograms. Individual future risk scores, which AI takes seconds to generate, can be integrated into radiology reports shared with patients and physicians.

“AI for cancer risk prediction offers us the opportunity to individualize all women’s care that is not systematically available,” he said. “This is a tool that will help us deliver personalized precision medicine at the national level.”