AI, mammograms and EHRs can predict early breast cancer

An algorithm that integrates machine and deep learning with a linked set of digital mammography images and electronic health records detected early breast cancer as well as radiologists.

The finding from a study of the capability opens the door to the use of the tool as a second reader in clinical settings.

Breast cancer is the second leading cause of cancer-related deaths, and the most commonly diagnosed cancer in women throughout the world. Digital mammography is the primary imaging modality of breast cancer screening in asymptomatic women.

However, analyzing the images is challenging because of the subtle difference between lesions and background tissue, different lesion types, the non-rigid nature of the breast and the relatively small proportion of cancers in the screening population.

The study researchers, from Israel, collected 52,936 images in 13,234 women who had at least one mammogram between 2013 and 2017 and who had health records for at least one year before undergoing a mammogram. They created a machine learning and a deep learning model that combined a set of algorithms and linked it with the images and clinical data in the EHRs. They then trained and tested the model.

mammogram_wide-CROP.jpg

The resulting algorithm predicted breast malignancy detected within 12 months from the index exam, and improved risk prediction over the Gail model, one of the most common breast cancer risk assessment algorithms currently used. The algorithm identified white blood cell profiles and thyroid function tests as associated with breast cancer even though these factors are not currently integrated in published risk scores.

Perhaps most notably, the algorithm identified breast cancer in 48 percent of women in whom the initial radiologist’s interpretation was negative for cancer but in whom the cancer was detected within a year, which the researchers found to be of “immediate clinical relevance.”

“ML technology emphasizes the need for linking data sets from multiple modalities to improve the accuracy of breast cancer detection and save experts’ valuable time on high-probability healthy individuals,” the authors stated.

The study was published online June 18 in Radiology.

The study authors pointed out that the algorithm did not perform better than the radiologists; it performed differently because it used different tools.

“The algorithm, which combined machine-learning and deep-learning approaches, can be applied to assess breast cancer at a level comparable to radiologists and has the potential to substantially reduce missed diagnoses of breast cancer,” the researchers noted.

For reprint and licensing requests for this article, click here.