Deep Learning Model to Assess Cancer Risk on the Basis of a Breast MR Image Alone

Leesburg, VA, June 24, 2019According to an Original Research article in the American Journal of Roentgenology, from January 2011 to June 2013, MIT’s Tally Portnoi and her colleagues in the electrical engineering and computer science departments collected 1656 consecutive breast MR images from breast cancer screening examinations performed on 1183 high-risk women at a single institution. 

Excluding women who lacked a five-year screening follow-up, as well as those who had developed cancer other than primary breast after examination, the team developed two models: a logistic regression model based on traditional risk factors and a deep learning model based on the MR image alone, which was trained to predict if a patient would develop breast cancer within five years of the screening. 

Examinations occurring within six months that yielded a cancer diagnosis were excluded from the testing sets in each fold of cross-validation. 

Comparing both newer models against the pre-existing Tyrer-Cusick breast cancer risk evaluation tool using mean (± SD) AUC values and observed-to-expected (OE) ratios across tenfold cross-validation, the risk factor logistic regression model and the deep learning model achieved mean AUC values of 0.558 ± 0.108 and 0.638 ± 0.094, respectively. 

By contrast, the Tyrer-Cusick model achieved an AUC value of 0.493 ± 0.092. 

Moreover, compared with the mean OE ratio of 1.091 ± 0.255 obtained using Tyrer-Cusick, the risk factor logistic regression models and the deep learning model achieved mean OE ratios of 0.828 ± 0.181 and 0.993 ± 0.658, respectively. 

Observing that the AUC value of Portnoi’s model increased and variability decreased as the size of the training dataset increased, Radiology Business noted that “focusing on the full spectrum of 3D data instead of 2D images shows potential to further improve risk prediction.”

Founded in 1900, the American Roentgen Ray Society (ARRS) is the first and oldest radiology society in the North America, dedicated to the advancement of medicine through the profession of radiology and its allied sciences. An international forum for progress since the discovery of the x-ray, ARRS maintains its mission of improving health through a community committed to advancing knowledge and skills with an annual scientific meeting, monthly publication of the peer-reviewed American Journal of Roentgenology (AJR), quarterly issues of InPractice magazine, topical symposia and AJR Live Webinars, print and online educational materials, as well as awarding scholarships via The Roentgen Fund®.