Researchers from Massachusetts General Hospital in Boston have found that their deep-learning algorithm's categorization of dense breast tissue correlated to radiologist breast density assessments and also refined them, according to a study published online February 20 in Academic Radiology.
"Our deep-learning model had a high clinical acceptance rate among both academic and community radiologists and reduced the proportion of mammograms assessed as dense," wrote a team led by Dr. Brian Dontchos.
For the study, Dontchos and colleagues implemented a deep-learning model at a dedicated breast imaging practice in October 2018. The practice staff included academic and community breast imaging radiologists.
The algorithm evaluated 2,174 screening mammograms for density. Reading radiologists were then presented with this data at the time of their interpretation. The researchers compared radiologist agreement by type (academic or community) with the algorithm's assessment.
Academic radiologists had a 94.9% acceptance rate of the algorithm's density categorization, and community radiologists had a 90.7% acceptance rate, the team found. The algorithm helped radiologists refine their density categories as well, decreasing the proportion of mammograms defined as dense from 47% to 41%.
"[Our study] is an important step to validating our deep learning model prior to potential widespread implementation," Dontchos and colleagues concluded.
