In one, researchers from Radboud University Medical Centre in Nijmegen, the Netherlands, compared the performance of experienced radiologists to that of Transpara in detecting breast cancer on mammograms. More than 1,400 mammograms from three different vendors were retrospectively reviewed by 24 radiologists, which included 336 exams with cancer, 430 with benign abnormalities, and 669 normal mammograms. Results showed no significant difference between automated reading with the Transpara software and reading by the radiologists. In some instances, the radiologists had a higher area under the curve (AUC) performance, while Transpara had a higher AUC in others.
In an RSNA presentation of the second study, researchers will discuss how recent developments in machine learning offer opportunities to develop fully automated systems for reading mammograms. For instance, decision support to improve recall decisions and prescreening of exams. In fact, researchers of the third study will discuss precisely that -- preselecting mammograms without abnormalities using deep learning.