AI algorithms outperformed traditional clinical risk models in a large-scale study, predicting five-year breast cancer risk more accurately. These models use mammograms as the sole source of data, offering potential advantages in individualizing patient care and improving predictive accuracy.
In a large study of thousands of mammograms, artificial intelligence (AI) algorithms outperformed standard clinical risk models for predicting five-year risk for breast cancer. The results of the study were published in Radiologya journal of the Radiological Society of North America (RSNA).
A woman’s risk of breast cancer is usually calculated using clinical models such as the Breast Cancer Surveillance Consortium (BCSC) risk model, which uses self-reported and other patient information—including age, family medical history, whether she has given birth, and whether she has dense breasts—to calculate a risk score.
“Clinical risk models depend on gathering information from various sources, which are not always available or collected,” said lead researcher Vignesh A. Arasu, MD, Ph.D., a research scientist and practicing radiologist at Kaiser Permanente Northern California. “Recent advances in AI deep learning give us the ability to extract hundreds to thousands of additional mammographic features.”
In a retrospective study, Dr. We analyzed data associated with negative (showing no visible evidence of cancer) screening 2D mammograms performed at Kaiser Permanente Northern California in 2016. Of the 324,009 women screened in 2016 who met eligibility criteria -therefore, a random sub-cohort of 13,628 women was selected for analysis. Additionally, all 4,584 patients from the eligibility pool who were diagnosed with cancer within five years of the original 2016 mammogram were also studied. All women were followed until 2021.
“We selected from the entire year of screening mammograms performed in 2016, so our study population is representative of communities in Northern California,” said Dr. Arasu.
The researchers divided the five-year study period into three time periods: intermediate cancer risk, or incident cancers diagnosed between 0 and 1 year; future cancer risk, or incident cancers diagnosed from between one and five years; and all cancer risks, or incident cancers diagnosed between 0 and 5 years.
Using 2016 screening mammograms, five-year breast cancer risk scores were generated by five AI algorithms, including two academic algorithms used by researchers and three commercially available algorithms. Risk scores were compared to each other and to the BCSC clinical risk score.
“All five AI algorithms performed better than the BCSC risk model for predicting breast cancer risk at 0 to 5 years,” said Dr. Arasu. “This strong predictive performance over five years indicates that AI is identifying both missed cancers and breast tissue features that help predict future cancer development. One thing on mammograms allows us to monitor breast cancer risk. This is the ‘black box’ of AI.”
“[AI] is a tool that will help us provide personalized, precision medicine at the national level..” — Vignesh A. Arasu, MD, Ph.D.
Some of the AI algorithms are good at predicting patients at high risk of interval cancer, which is often aggressive and may require second readings of mammograms, supplemental screening, or short-interval follow-up imaging . When examining women with the highest 10% risk as an example, AI predicted up to 28% of cancers compared to 21% predicted by BCSC.
Even AI algorithms trained for a short time (as low as 3 months) were able to predict future cancer risk up to five years when no cancer was clinically detected by screening mammography. When used in combination, AI and BCSC risk models further improved cancer prediction.
“We are looking for an accurate, efficient and measurable way of understanding women’s breast cancer risk,” said Dr. Arasu. “Mammography-based AI risk models provide practical advantages over traditional clinical risk models because they use a single data source: the mammogram itself.”
said Dr. Arasu says some institutions are already using AI to help radiologists detect cancer in mammograms. A person’s future risk score, which takes seconds to generate by AI, can be included in a radiology report shared with the patient and their physician.
“AI for predicting cancer risk offers us the opportunity to personalize each woman’s care, which is not systematically available,” he said. “This is a tool that will help us provide personalized, precision medicine on a national scale.”
Reference: “Comparison of Mammography AI Algorithms with a Clinical Risk Model for 5-year Breast Cancer Risk Prediction: An Observational Study” by Vignesh A. Arasu, Laurel A. Habel, Ninah S. Achacoso, Diana SM Buist, Jason B. Cord , Laura J. Esserman, Nola M. Hylton, M. Maria Glymour, John Kornak, Lawrence H. Kushi, Donald A. Lewis, Vincent X. Liu, Caitlin M. Lydon, Diana L . Miglioretti, Daniel A. Navarro, Albert Pu, Li Shen, Weiva Sieh, Hyo-Chun Yoon and Catherine Lee, 6 June 2023, Radiology.
DOI: 10.1148/radiol.222733
Collaborated with Dr. Arasu are Laurel A. Habel, Ph.D., Ninah S. Achacoso, MS, Diana SM Buist, Ph.D., Jason B. Cord, MD, Laura J. Esserman, MD, Nola. M. Hylton, Ph.D., M. Maria Glymour, Sc.D., John Kornak, Ph.D., Lawrence H. Kushi, Sc.D., Don A. Lewis, MS, Vincent X. Liu, MD , Caitlin M. Lydon, MPH, Diana L. Miglioretti, Ph.D., Daniel A. Navarro, MD, Albert Pu, MS, Li Shen, Ph.D., Weiva Sieh, MD, Ph.D., Hyo- Chun Yoon, MD, Ph.D., and Catherine Lee, Ph.D.