Machine Learning May Improve Accuracy of Breast Cancer Diagnoses
Posted: Wednesday, November 13, 2019
Results of a diagnostic study published in JAMA Network Open suggest that machine-learning methods may be a helpful tool for distinguishing between preinvasive and invasive breast cancer lesions. Differentiating and diagnosing biopsy specimens, particularly atypia and ductal carcinoma in situ, can be challenging, even among expert pathologists. This automated new technology may assist pathologists in improving the diagnosis of the full spectrum of breast biopsy samples.
“Medical images of breast biopsies contain a great deal of complex data and interpreting them can be very subjective,” stated Joann G. Elmore, MD, MPH, of the University of California, Los Angeles (UCLA), in a UCLA press release. “Sometimes, doctors do not even agree with their previous diagnosis when they are shown the same case a year later.”
Using 240 breast biopsy images, the research team trained a computer to recognize patterns associated with different types of breast lesions. Cases of atypia and ductal carcinoma were oversampled for statistical power. The performance of the machine-learning analyses was compared with independent interpretations from 87 practicing pathologists.
The automated approach was less accurate than human doctors when differentiating cancer from noncancer cases. The identification by pathologists showed a sensitivity of 0.84, whereas the sensitivity of the machine learning was between 0.49 and 0.70. However, the automated approach more accurately determined the difference between ductal carcinoma and atypia. The automated system had a sensitivity between 0.88 and 0.89, whereas pathologists’ average sensitivity was 0.70.
“These results are very encouraging,” Dr. Elmore concluded. “The computer-based automated approach shows great promise.”
Disclosure: The study authors’ disclosure information may be found at jamanetwork.com.