Scientists have developed a new deep learning model capable of predicting breast cancer up to five years in advance, according to a study published in Radiology. The model, called AsymMirai, builds on previous AI research and provides a more interpretable and accurate way to assess risk from routine mammograms.
Breast cancer is a leading cause of cancer-related deaths among women, with 1 in 8 women in the U.S. developing invasive breast cancer in their lifetime. Early detection through regular mammograms significantly reduces mortality, but accurately predicting who will develop cancer remains a challenge.
Previously, an AI tool named Mirai showed promise in forecasting cancer risk. However, it acted like a “black box,” making predictions without revealing how it reached its conclusions — raising concerns of overreliance and misdiagnosis. 
To solve this, researchers at Duke University introduced AsymMirai, which simplifies the prediction process. Instead of relying on complex calculations, the model compares subtle differences between left and right breast tissue, a factor traditionally used only for detection, not prediction.
“We developed an interpretable AI method that allows us to predict breast cancer from mammograms 1 to 5 years in advance,” said lead author Jon Donnelly, a PhD student at Duke University. “AsymMirai is much simpler and much easier to understand than Mirai.”
This innovation could revolutionize breast cancer screening, enabling more personalized monitoring and earlier interventions — ultimately saving lives.
