When it comes to breast cancer, artificial Intelligence (AI) has already made headlines as a powerful tool for early and accurate breast cancer detection—an essential goal that improves outcomes and saves lives.
Deaths from breast cancer have fallen 43 percent over the last three decades thanks to advancements in screening and treatment. Regular screening—most commonly through mammography—is currently the most effective way to detect breast cancer early, when it’s more likely to be smaller and contained in the breast. Plus, when breast cancer is caught at its earliest stages, patients are more likely to have better outcomes and may not need as aggressive treatment plans.
Mammography has long been the gold standard for breast cancer screening and is highly effective in detecting abnormalities in breast tissue that could be cancer long before patients experience lumps and other symptoms. But mammograms aren’t perfect. Some women need additional imaging (such as ultrasound or MRI) because they have dense breasts or because they have certain risk factors (gene mutations, family history). At times, women are called back for what turn out to be unnecessary biopsies because of false positives.
Still, mammograms and other screening tools undoubtably save lives. That’s why improving screening technology and pathology are major areas of research—and ones where AI shows a lot of promise.
Here in part two of BCRF’s “StAI Tuned” series, we discuss how AI and machine learning (ML) are being integrated into screening and diagnostic procedures to find breast cancer in its earliest stages. Read part one on AI’s potential in breast cancer research and part three on precision medicine and predictive analytics.
Highly skilled radiologists interpret breast images. They are experts in both selecting the imaging technique(s) to use with patients and comprehensively assessing each patient’s anatomy, any abnormalities or pathologies detected, and their significance. When a woman gets a mammogram or other screen, her radiologist describes and details each finding, including its location, size, shape, density, or other relevant features.
AI algorithms may make radiologists’ workflow far more efficient, and they can provide quantitative analyses that are not subject to human bias, making data-driven calls for questionable mammograms that could be interpreted differently. AI-powered software can automate interpretation of breast mammograms, ultrasounds, and MRI scans to get patients their results faster.
AI techniques can help radiologists identify breast cancer that would have otherwise been undetectable in its early stages. Techniques like image enhancement and de-noising—decreasing the background shadows—can improve the quality of breast images and allow radiologists to view anatomical structures more clearly. As such, AI-powered tools can detect subtle abnormalities, decipher ambiguous features, and identify patterns and characteristics that may not immediately jump out to the human eye. AI tools can also estimate tumor size and shape. Together, these advantages position AI as a powerful partner while doctors aim to make screening more accurate and reduce false positive and negative results.
AI-powered breast imaging may also improve breast cancer care in low-resourced or rural areas, where women often lack easy access to specialists and experts. AI systems enable remote interpretation of imaging, facilitating timely diagnosis and treatment for patients regardless of where someone lives. Not only that, but given the fact that mammography is the most time- and cost-effective screening option available now, any improvements in its accuracy will reduce follow-up visits, biopsies, and more costly screening tests like MRI. The synergy between AI and radiomics holds immense potential for advancing breast cancer care and improving outcomes for patients.
BCRF investigators Drs. Constance Lehman and Regina Barzilay developed and tested a mammography-based deep learning model called MIRAI. To do this, they utilized a large set of diverse patient data, analyzed multiple mammogram images over time, and integrated risk factor information into the tool’s analysis. The team demonstrated that MIRAI could yield individualized, equitable, and cost-effective improvements in breast cancer risk prediction compared to traditional risk models. Importantly, MIRAI provides consistent results across mammography sites and machines.
BCRF investigator Dr. Adam Yala and Dr. Barzilay are now conducting a BCRF-supported prospective study testing MIRAI’s ability to predict which patients are high-risk and follow their progress through MRI screening. This study will validate results demonstrating that MIRAI can identify patients with a high-risk of developing breast cancer in the next five years. Ultimately, the results of the study could help redefine breast cancer screening guidelines and design novel protocols for managing high-risk patients.
In her current BCRF project, Dr. Lehman is leveraging MIRAI to move breast cancer screening protocols from an age-based paradigm to a risk-based one, which would particularly benefit high-risk young (under 40) women who aren’t generally being screened. She and her team created a breast cancer risk prediction model that combines imaging, biologic, and behavioral data to personalize screening regimens and improve early detection while lowering the overall cost of screening. Their AI tool also supports more accurate breast cancer risk assessment from mammograms across diverse races and ethnicities compared to traditional risk models. Dr. Lehman is now expanding the model to measure changing risk levels in individuals over time, which will ultimately lead to an even more accurate way to predict future breast cancer.
Women who have already been treated for breast cancer are at increased risk for recurrence and for second breast cancers. Having dense breasts is also associated with an increased risk of breast cancer, in part because dense breast tissue can obscure abnormalities on mammograms making screening challenging. BCRF investigator Dr. Wendie Berg is using AI to enhance ultrasound imaging of dense breast tissue and make it more accurate.
If screening reveals a malignancy, timely and accurate diagnosis by breast tissue biopsy is critical to get a patient into treatment. During a biopsy, a pathologist studies a sample under a microscope for cancer cells, their growth patterns, and characteristics in the cells and tissue that indicate the breast cancer subtype and grade. Integrating AI into digital pathology is proving to be a revolutionary tool to improve imaging sensitivity and specificity and help pathologists work faster, while diagnosing breast cancers accurately.
AI can identify subtle patterns and features in digitized pathology images that may be overlooked by human eyes. For example, AI-powered algorithms can detect malignant cells, assess a tumor’s features, and predict tumor aggressiveness accurately and with high sensitivity. This technology is currently being leveraged to find lymph node metastases that are difficult to detect.
That AI can analyze vast amounts of medical data with speed and precision is also invaluable to pathologists. These tools can analyze whole slide images of breast lesions and classify them into distinct categories, including invasive carcinomas, microinvasive carcinomas, ductal carcinomas in situ (DCIS), and benign breast lesions. Microinvasive carcinomas are most often seen in association with DCIS and identifying it can be very difficult and time-consuming.
Breast MRI can reveal characteristics of a tumor and its environment that are helpful for diagnosing high-risk patients. MRI images contain complex information and content, making them an abundant source for AI to extract data. AI is especially helpful here as it can learn descriptive features like tumor shape, borders, and texture, providing another set of “eyes.” AI has improved MRI detection and characterization of breast cancer and will improve the use of imaging biomarkers in clinical decision-making.
Molecular biomarkers of breast cancer are also seen in tissue biopsy samples under the microscope. Breast cancer management is informed by key biomarkers in and on breast cancer cells including estrogen receptors (ER), progesterone receptors (PR), human epidermal growth factor 2 (HER2), and Ki-67. These biomarkers indicate the subtype of breast cancer and how rapidly it is growing. AI tools are enabling more accurate detection of these biomarkers, including low and ultra-low levels of HER2, so patients can get the right diagnosis and treatment.
BCRF investigator Dr. Karen Taylor is digitizing about 22,000 diagnostic pathology slides from breast cancer patients that have had expert analysis and creating infrastructure to validate AI algorithms using the data. The AI systems will analyze these images to look for new patterns in both the cancer cells and the surrounding tissue cells to gain insight into how the cancer will respond to different treatments. Her work validating these tests will aid pathologists in routine reporting and allow clinicians to make better individualized treatment decisions.
By harnessing the power of AI-driven analytics and automation, healthcare providers can deliver more precise, efficient, and tailored breast cancer care. AI is poised to revolutionize screening, offering unprecedented opportunities to make it faster and more accurate—ultimately saving lives and improving outcomes in the process.
Despite its transformative potential, there are challenges for AI on the road to widespread adoption in imaging. Using these tools requires extensive research to validate their accuracy and ensure they apply to widespread populations. Data standardization, regulatory compliance, and ethical considerations all pose significant barriers to scalability and widespread implementation in the real world. But if clinicians can understand and feel confident in AI tools, they can provide their patients with a rationale behind the recommendations and gain their trust, creating a powerful and valuable synergy in imaging.
This is why AI-related research is essential in breast cancer. BCRF investigators and others are on the cutting edge of this field, doing the early work to make AI’s transformative potential a reality for patients in the future.
Ahn, J. S., Shin, S., Yang, S. A., Park, E. K., Kim, K. H., Cho, S. I., … & Kim, S. (2023). Artificial Intelligence in Breast Cancer Diagnosis and Personalized Medicine. Journal of Breast Cancer, 26(5), 405.
Demircioglu, A., Grueneisen, J., Ingenwerth, M., Hoffmann, O., Pinker-Domenig, K., Morris, E., … & Umutlu, L. (2020). A rapid volume of interest-based approach of radiomics analysis of breast MRI for tumor decoding and phenotyping of breast cancer. PloS one, 15(6), e0234871.
Galati, F., Rizzo, V., Trimboli, R. M., Kripa, E., Maroncelli, R., & Pediconi, F. (2022). MRI as a biomarker for breast cancer diagnosis and prognosis. BJR| Open, 4(1), 20220002.
Kim, I., Kang, K., Song, Y., & Kim, T. J. (2022). Application of artificial intelligence in pathology: trends and challenges. Diagnostics, 12(11), 2794.
Ng, A. Y., Oberije, C. J., Ambrózay, É., Szabó, E., Serfőző, O., Karpati, E., … & Kecskemethy, P. D. (2023). Prospective implementation of AI-assisted screen reading to improve early detection of breast cancer. Nature Medicine, 29(12), 3044-3049.
What Do Doctors Look for in Biopsy and Cytology Samples? (2023, August 1). Www.cancer.org. https://www.cancer.org/cancer/diagnosis-staging/tests/biopsy-and-cytology-tests/testing-biopsy-and-cytology-samples-for-cancer/what-doctors-look-for.html
Yala, A., Lehman, C., Schuster, T., Portnoi, T., & Barzilay, R. (2019). A deep learning mammography-based model for improved breast cancer risk prediction. Radiology, 292(1), 60-66.
Zheng, D., He, X., & Jing, J. (2023). Overview of artificial intelligence in breast cancer medical imaging. Journal of Clinical Medicine, 12(2), 419.
Please remember BCRF in your will planning. Learn More
