Every year, Breast Cancer Awareness Month is observed in October throughout the United States to highlight the importance of screening to detect breast cancer at an early and manageable stage. 

According to the Breast Cancer Research Foundation, an estimated 290,560 people will be diagnosed with breast cancer in the United States in 2022. 

We spoke with Manisha Bahl, MD, MPH, a physician-scientist in the Department of Radiology at Massachusetts General Hospital, who is investigating cutting-edge imaging techniques for breast cancer detection and diagnosis. 

1. What key challenges are you currently investigating within the field of breast imaging?

I am investigating the use of artificial intelligence (AI) and deep learning (DL) to improve the detection, diagnosis, and treatment of breast cancer.

My current research, which is funded by a five-year grant from the National Institutes of Health, is focused on the use of AI/DL to support more targeted and precise treatment plans for women with ductal carcinoma in situ (DCIS), also known as Stage 0 breast cancer.

Our long-term goals are to decrease the morbidity and costs associated with overtreatment and improve patient outcomes.

The incidence of DCIS has markedly increased due to mammographic screening, with an estimated 51,400 diagnoses in the United States in 2022. Current guidelines recommend that women with DCIS be treated with surgery, radiation, and endocrine therapy.

However, concerns regarding overtreatment have led to active surveillance trials for DCIS, in which patients do not undergo surgery or radiation and instead are surveilled with imaging.

Successful clinical implementation of such active surveillance programs relies on the careful selection of eligible patients.

In particular, the selection criteria must be designed to exclude patients with occult invasive cancer, which is found at surgery in one-quarter of patients with biopsy-proven DCIS.

The goals of my research are to develop and implement a robust tool using AI and DL that can pre-operatively predict the risk of concurrent invasive cancer in women with DCIS.

2. What is unique about the research approach you are taking? 

Our approach is unique in that it utilizes AI and DL for image analysis. While conventional prediction models rely on human-engineered features, DL-based models extract and learn predictive features directly from imaging data.

The DL model is thus able to identify features and patterns that may not be discernible to the human eye and preserves the rich information contained within the image rather than reducing it to a limited range of values.

3. What are the clinical implications of the research work you are doing? 

The biologic heterogeneity of DCIS and the ensuing challenge to predict patient outcomes have led to aggressive treatment with routine use of surgery and radiation for all patients.

Our research, which is focused on accurate pre-operative prediction of the upgrade risk of DCIS to invasive disease, would offer both short- and long-term benefits for patients.

Short-term impact: A robust model to predict upgrade risk could immediately impact surgical decision-making since women with pure DCIS at core needle biopsy but found to have invasive disease at surgery need to undergo a second surgery for sentinel lymph node sampling.

This second surgery could be avoided if the patient is known to be at high upgrade risk and lymph node sampling is performed during the initial surgery.

Long-term impact: A paradigm shift in the treatment of DCIS is currently underway. Active surveillance is being evaluated in several multi-center trials as an alternative management approach for women with indolent nonhazardous DCIS.

Critical to these active surveillance programs is appropriate selection of eligible patients, but there are currently no tools that can accurately identify the subset of patients who have low-risk disease and are therefore appropriate candidates for active surveillance.

If successful, use of our AI/DL model could lead to the identification of the subset of women with DCIS who are appropriate candidates for active surveillance.

Ultimately, it could decrease overtreatment of DCIS and its associated side effects and costs and empower women to make more informed choices with regard to their treatment options.

4. What resources (collaborators, patient access, electronic health records) at Mass General have helped to advance your work? 

The rich resources at Mass General have been critical to the success of our work.

Our collaborative relationships across MGB and Harvard and complementary expertise in breast imaging, AI, DL, and data science are essential for the successful development and implementation of AI algorithms into clinical practice.

Access to clinical and imaging data and an infrastructure to extract imaging studies from the PACS are also critical components needed to advance our work.

5. How different is the field of breast imaging research today from five years ago? 

The field of breast imaging has transitioned through a variety of technological advances, and we are now in the midst of another technological advance – that of artificial intelligence, machine learning, and deep learning.

Mammographic technology has markedly advanced, from direct-exposure film mammography to screen-film mammography to modern full-field digital mammography and tomosynthesis.

Tomosynthesis combined with 2D mammography, which was approved by the FDA more than 10 years ago, has led to higher cancer detection rates and lower false-positive rates compared to 2D mammography alone.

Other breast imaging modalities that have seen recent advancements include breast MRI and contrast-enhanced mammography.

Breast MRI is increasingly utilized for high-risk screening and for extent of disease evaluation in women with biopsy-proven breast cancer, and contrast-enhanced mammography is an emerging modality that combines the morphologic and anatomic information provided by mammography with the vascular physiologic information provided by breast MRI.

6. Can you talk briefly about the implications of artificial intelligence for breast cancer screening? What has your latest research discovered? 

AI applications for breast cancer screening have quickly advanced from feasibility and reader studies to clinical implementation.

Commercial AI applications for screening mammography are available for lesion detection and diagnosis and triage, and research has shown that these algorithms could help radiologists improve not only their accuracy but also their efficiency.

However, this research is largely based on retrospective reader or simulation studies, which are limited by patient selection bias and applicability concerns.

The breast imaging community looks forward to rigorous clinical evaluation of these tools aligned with their specific intended use to clearly understand their true impact on patient outcomes.

7. Where do you see breast imaging research headed over the next five years? 

The potential of AI for breast imaging has yet to be realized. Most research-to-date on AI algorithms is based on reader studies with cancer-enriched datasets and simulation studies.

We are in need of rigorous post-implementation clinical evaluation aligned with each AI tool’s specific intended use to help determine its true impact on patient outcomes.

I also believe that the contrast-enhanced modalities – MRI and mammography – will continue to be investigated and improved.

In particular, we need to further define the role of contrast-enhanced mammography and compare it to the armamentarium of imaging techniques that are currently available for detecting and diagnosing breast cancer.

About the Mass General Research Institute
Research at Massachusetts General Hospital is interwoven through more than 30 different departments, centers and institutes. Our research includes fundamental, lab-based science; clinical trials to test new drugs, devices and diagnostic tools; and community and population-based research to improve health outcomes across populations and eliminate disparities in care.
Support our Research