Can the Machine Read that Mammogram Better than You the Physician? How Artificial Intelligence is Changing Breast Cancer Screening
By Manijeh “Mani” Berenji
Over the past 20 years, medical technology has been advancing in the breast imaging sciences. Better quality scans can help physicians identify masses that need further work-up, optimizing treatment and leading to better outcomes for the patient. This has led to a decline in the U.S. breast cancer death rate (dropping 40% from 1989 to 2017) (1). Even with the advancements in computer-aided detection software and medical assistive tools, they have not been able to systematically pinpoint those hard-to-detect masses that could become cancerous (2,3).
1 in 8 U.S. women will develop invasive breast cancer over the course of her lifetime. In 2019, an estimated 268,600 new cases of invasive breast cancer were diagnosed in women in the U.S., along with 62,930 new cases of non-invasive (in situ) breast cancer (4). While there has been progress in screening, the number of false positives and false negatives continue to be of-issue. Screening mammograms do not find about 1 in 5 breast cancers. Women with dense breasts are more likely to get false-negative results (5). And false-positive results are more common in women who are younger, have dense breasts, have had breast biopsies, have breast cancer in the family, or are taking estrogen (5). Such misdiagnoses can lead to more patient anxiety and more morbidity, costing the healthcare systems millions of dollars in unnecessary spending. While radiologists are still considered the subject matter experts in interpreting advanced breast imaging scans (including mammograms), given the sheer volume of images to be processed per patient, the increasing number of patients worldwide, and shortages of trained specialists, radiologists need new methodologies to process the images quickly and identify potential masses accurately.
Recent research from the UK (in conjunction with Google Health) evaluated the performance of a new artificial intelligence (AI) system for breast cancer prediction and compared those predictions to those to made by readers in routine clinical practice (3). The UK test set consisted of screening mammograms that were collected between 2012 and 2015 from 25,856 women at two screening centers in England (where women are screened every three years and diagnoses made by two radiologists). The US test set consisted of screening mammograms that were collected between 2001 and 2018 from 3,097 women at one academic medical center (where women are screened every one to two years and diagnoses made by a single radiologist). A deep learning model for identifying breast cancer in screening mammograms was developed and evaluated using these two datasets in three primary ways: (i) Comparing AI predictions with the historical decisions made in clinical practice; (ii) Retesting the UK data on the US data to evaluate the generalizability across populations; and (iii) Comparing the performance of the AI system with outside readers in an independent study using a subset of the US test set.
In the UK, the AI system showed specificity superior to that of the first reader. In the USA, the AI system exhibited specificity and sensitivity superior to that of radiologists practicing in an academic medical center. With the AI system, there was an absolute reduction of false positives (US 5.7% and UK 1.2%) and absolute reduction of false negatives (US 9.45 and UK 2.7%). While this technology is still in its infancy, the potential implications are significant. If AI, in conjunction with live radiologist, can provide accurate identification of a mass suspicious for malignancy, that can yield enormous benefit for the patient, healthcare system, and society-at-large.
1. DeSantis CE, Ma J, Gaudet MM, Newman LA, Miller KD, Goding Sauer A, Jemal A, Siegel RL. Breast cancer statistics, 2019. CA Cancer J Clin. 2019;69(6):438–451.
2. Kohli A, Jha S. Why CAD Failed in Mammography. J Am Coll Radiol. 2018 Mar;15(3 Pt B):535–537.
3. McKinney SM, Sieniek M, Godbole V, Godwin J, Antropova N, Ashrafian H, Back T,Chesus M, Corrado GC, Darzi A, Etemadi M, Garcia-Vicente F, Gilbert FJ,Halling-Brown M, Hassabis D, Jansen S, Karthikesalingam A, Kelly CJ, King D, Ledsam JR, Melnick D, Mostofi H, Peng L, Reicher JJ, Romera-Paredes B, Sidebottom R, Suleyman M, Tse D, Young KC, De Fauw J, Shetty S. International evaluation of an AI system for breast cancer screening. Nature. 2020;577(7788):89–94.
4. Breastcancer.org. U.S. Breast Cancer Statistics. Available at: https://www.breastcancer.org/symptoms/understand_bc/statistics. Accessed 5 January 2020.
5. American Cancer Society. Limitations of Mammograms. Available at: https://www.cancer.org/cancer/breast-cancer/screening-tests-and-early-detection/mammograms/limitations-of-mammograms.html. Accessed 5 January 2020.