AI-based model may help to better predict future breast cancer risk

It outperformed the Breast Cancer Surveillance Consortium, the typical risk assessment method.
Mrigakshi Dixit
Doctor looking at the patient's mammogram.
Doctor looking at the patient's mammogram.


An artificial intelligence model has been able to efficiently predict future breast cancer risks. In fact, it “outperformed” the conventional clinical risk model, which has been widely used to assess the likelihood of breast cancer.

This large-scale study was led by Vignesh A. Arasu, a research scientist and practicing radiologist at Kaiser Permanente, Northern California. 

The clinical risk model needs multiple inputs

A woman's risk of developing breast cancer is usually analyzed using clinical risk models, like the Breast Cancer Surveillance Consortium (BCSC). However, this model requires a variety of information, including a mammogram analysis, to calculate the risk score.

This model is based on patient self-reported data and other information related to age, family history of cancer, childbirth history, and breast density. 

“Clinical risk models depend on gathering information from different sources, which isn’t always available or collected. Recent advances in AI deep learning provide us with the ability to extract hundreds to thousands of additional mammographic features,” said Arasu in an official release.

AI risk models only use one source of data, i.e., a mammogram. The team gathered information from thousands of 2016 mammograms taken at Kaiser Permanente, Northern California.

AI-based risk models

In 2016, roughly 324,009 women were screened for breast cancer. Of this, 13,628 were selected to derive data for creating the AI-based risk model. The data of 4,584 women diagnosed with breast cancer within five years of the initial mammogram in 2016 was also taken into consideration. All of these study participants were followed up on until 2021.

The five-year study period was divided into three time periods: “interval cancer risk, or incident cancers diagnosed between 0 and 1 years; future cancer risk, or incident cancers diagnosed from between one and five years; and all cancer risk, or incident cancers diagnosed between 0 and 5 years.”

Based on this information, the team generated five AI model algorithms. The risk scores were then compared to the BCSC clinical risk score and with each AI model. 

“All five AI algorithms performed better than the BCSC risk model for predicting breast cancer risk at 0 to 5 years,” Dr. Arasu said. “This strong predictive performance over the five-year period suggests AI is identifying both missed cancers and breast tissue features that help predict future cancer development. Something in mammograms allows us to track breast cancer risk. This is the ‘black box’ of AI.”

Patients with a high risk of interval cancer were accurately predicted by some of the AI algorithms. AI predicted up to 28 percent of cancers when evaluating women at the highest risk, whereas the BCSC's scored 21 percent. 

The team highlights that when the AI and BCSC risk models were used in combination, the prediction results improved significantly. 

“AI for cancer risk prediction offers us the opportunity to individualize every woman’s care, which isn’t systematically available. It’s a tool that could help us provide personalized, precision medicine on a national level,” said Arasu.