Applying Artificial Intelligence to Mammography Images
Computerized clinical toolkit could one day help radiologists identify benign cases earlier and reduce unnecessary medical procedures.
Mammography is the standard screening exam for breast cancer, but high recall rates are a major concern. Many women who are asked to undergo additional work-up later receive benign results. These extra medical procedures can cause unnecessary stress and increase health care costs. One potential solution is to use deep learning methods to automatically identify clinically relevant image features that may not be detected through visual assessment.
A study published Oct. 11 in Clinical Cancer Research shows that automatic deep learning methods can identify subtle imaging features from mammograms to distinguish among malignant, negative and recalled-benign cases, in which patients were assessed as not having cancer in a follow-up. The researchers developed and evaluated deep learning classifiers using two independent mammography datasets, which included a total of 3,715 patients and 14,860 images. They constructed a neural network model to classify mammography images into malignant, negative and recalled-benign categories. Approximately 95 percent of the images were used for model training, and the remaining data were used for testing.
The model achieved the best performance when distinguishing between negative and recalled-benign cases. The second-best performance was achieved for distinguishing between negative and malignant images, suggesting that deep learning could also assist in breast cancer diagnosis. According to the authors, incorporating deep learning-based artificial intelligence into breast cancer screening may help improve the ability of radiologists to interpret mammograms and reduce unnecessary repeat screenings.