Last week, the American Journal of Roentgenology published an exciting new study on artificial intelligence (AI). It involved a deep-learning model that was put together with data from MR images of breasts that’s able to aid in predicting the five-year risk in women with a higher chance of developing breast cancer.
According to the team, which was led by MIT’s Tally Portnoi, what makes the study so riveting is that it provides another way to personalize breast cancer screening for women who are at a high-risk of developing the disease. While there are many risk-assessment models out there, effective ones for women who are high-risk are few and far between.
“Most of the existing [risk-assessment] models are well calibrated at the population level, but they are not sufficiently discriminative at the individual level,” Portnoi et al wrote in the paper. “This limitation becomes even more pronounced for high-risk cohorts, for whom commonly used risk factors are not sufficiently predictive and for whom it is less clear which additional patient features drive outcomes.”
In the study, the research team used 1,656 breast-MR images that had been performed on 1,183 high-risk women between January 2011 and June 2013. They compared their AI model’s performance with other risk-assessment models, including the famous Tyrer-Cuzick model and a logistic regression model they created themselves based on well-known breast cancer risk factors. The team’s novel deep-learning model showed an improved individual risk discrimination in comparison to the other two models. Furthermore, their model makes its predictions solely off of MR images, without knowledge of risk factors and the like.
The research team elaborated on the importance of their study: “The model learns useful features directly from the data. This design enables the model to capture subtle patterns that may not be discernible to the human eye.”
The team says that their study’s results give a promising glimpse into the potential of deep learning to help personalize cancer screening in the future.