A new AI model can accurately determine a patient’s five-year cancer risk based on a single breast MR image, outperforming state-of-the art risk assessment models.
The general screening population has access to accurate risk assessment models, but those platforms are not applicable to higher-risk patients and do not take into account individual risk factors, according to Tally Portnoi, department of electrical engineering and computer science at Massachusetts Institute of Technology in Cambridge, and colleagues.
To address this, the team created two AI models: an image-based deep learning model trained to predict if a patient would develop breast cancer within five years of screening and a risk factor logistic regression (RF-LR) model based on traditional risk factors.
Portnoi and colleagues included 1,656 breast MR images from screening exams performed at a single institution for 1,183 high-risk women from January 2011 to June 2013.
Comparing the two models to an established breast cancer risk evaluation platform—Tyrer-Cusick (TC) model—the DL model scored a mean area under the ROC curve (AUC) value of 0.638 ± 0.094, while the RF-LR model notched an AUC of 0.558 ± 0.108.
“Our DL model can assess the 5-year cancer risk on the basis of a breast MR image alone, and it showed improved individual risk discrimination when compared with a state-of-the-art risk assessment model,” Portnoi and authors wrote.
Admittedly, the authors wrote, the study is limited by its inclusion of a relatively small dataset taken from a single institution. But, Portnoi et al. believe such models may be invaluable for breast cancer detection.
“These models have the potential to provide more precise information to support personalized screening and preventive strategies for women,” the authors concluded.
The study was published online April 1 in the American Journal of Roentgenology.