AI improves radiologists’ cancer detection rates when reading mammograms
Artificial intelligence-based computer-aided detection (AI-CAD) software can help radiologists detect more cancers when interpreting mammograms, according to a new study published in the Journal of Digital Imaging. The authors also observed improvements in reader accuracy.
“About 50% of mammographically detected breast cancers are visible retrospectively on prior studies,” wrote lead author Alyssa T. Watanabe, MD, Keck School of Medicine at the University of Southern California in Los Angeles, and colleagues. “Many of these cancers are obscured by dense breast tissue, subtle on mammography, or missed through human error. In addition, there exists a high percentage of false-positive mammography results and unnecessary biopsies. For these reasons, the need exists for methods and techniques that can improve sensitivity and specificity in mammography interpretation.”
Watanabe et al. tested how cmAssist, an AI-CAD algorithm developed by CureMetrix, impacted the performance of seven radiologists with various levels of experience. The study included 2D mammogram data from 122 patients, and they were all originally evaluated as negative. All mammograms were performed from February 2008 to January 2016. Readers participating in this study did not have access to prior imaging results for each patient.
Radiologists viewed each mammogram with and without marking provided by the AI-CAD algorithm, and all readers saw improvements in their cancer detection rates (CDRs). Overall, the mean CDR for the seven radiologists was 51% without the AI-CAD algorithm and 62% with the algorithm. The mean percentage increase after applying the algorithm was 27%.
The authors also found that one small cancer in a fatty breast was missed by six of seven radiologists without the AI-CAD algorithm. Using the algorithm, that same cancer was missed by two radiologists.
It was not possible to determine a single area under the ROC curve (AUC) for each individual reader, but the authors observed that the radiologists as a group saw their AUC rise from 0.760 to 0.815, a 7.2% overall increase.
“This study shows how AI-based software can provide clinical benefit to radiologists in interpretation of screening mammograms,” the authors concluded. “The use of AI in clinical practice may potentially expedite workflow, enhance earlier detection of cancer, and reduce false-negative mammograms. The impact of AI on medical imaging in the future is likely to be profound. To the authors’ knowledge, this is the first peer-reviewed scientific study that shows significant benefit of AI to radiologists in clinical image interpretation.”
This research for this peer-reviewed study was funded by CureMetrix, the company behind cmAssist.
Michael has more than 18 years of experience as a professional writer and editor. He has written at length about cardiology, radiology, artificial intelligence and other key healthcare topics.