In the U.S., only 25% of breast biopsies among women ages 40 to 79 who are undergoing screening mammography wind up being positive for cancer. As a result, it's an important goal to reduce the number of unnecessary biopsies while still maintaining sensitivity for breast cancer, according to lead author Karen Drukker, PhD, of the University of Chicago.
The group applied its convolutional neural network-based deep-learning method to the mammography exams of more than 100 patients who had lesions visible only as microcalcifications. An expert radiologist also provided a subjective probability of malignancy. The researchers then compared the performance of the deep-learning method and the expert radiologists for distinguishing between benign and malignant microcalcification clusters.
They found that the deep-learning method would have avoided nearly three times as many benign breast biopsies -- without losing sensitivity -- than would have been obviated based on the subjective radiologist probability of malignancy. The difference was statistically significant.
What else did the team find? Stop by this talk on Monday afternoon for all the details.
