maggie chung
Faculty FellowUCSF

Maggie Chung

Radiology

Dr. Chung is a radiologist and Assistant Professor at UCSF. She earned her MD at the Warren Alpert Medical School, Brown University, followed by an internship at Scripps Mercy Hospital in San Diego. 

Spark Award Project

Joint project with Adam Yala

National breast cancer guidelines are deemed inefficient, resulting in unnecessary procedures and missed early detection opportunities. Existing risk assessment models, relying on factors such as demographics and family history, often inaccurately predict individual risk. This project proposes a new solution: Pillar, an AI tool that utilizes multi-modal breast imaging data for enhanced risk prediction. Their previous model, Mirai, outperformed existing ones, and by integrating additional imaging data over time, they aim to substantially enhance risk assessment and facilitate early detection and prevention efforts.

Maggie Chung’s Story

Beyond the Boundaries: Bakar Fellows Address Gaps in Breast Cancer Diagnostics

September 6, 2024

By: Emily Qi

Early diagnosis of breast cancer is crucial for effective treatment and a positive prognosis. However, current standards are outdated and not sufficiently individualized, which can hinder early detection.

Pillar aims to address these limitations.

Dr. Maggie Chung, a radiologist and assistant professor at UCSF, and her collaborator Adam Yala, are pioneering a method to enhance breast cancer risk prediction by utilizing multi-modal imaging data. Pillar will integrate mammograms taken over time with other screening modalities to provide a comprehensive and personalized assessment of a patient’s risk and subsequent steps.

Q: What’s it like being a physician and a researcher from a day-to-day perspective? 

A: In a typical week, I devote about two days to research and the rest of the time to clinical work. Having both roles is quite rewarding because it creates a synergy between my patient care and research activities. For example, as I encounter various patient issues in the clinic, these real-world challenges often spark ideas for research. I might think, “Could AI help us address this problem more effectively?” This interplay between clinical observations and research allows me to identify meaningful research questions and, conversely, apply new insights to improve my clinical practice. In essence, the two roles complement each other beautifully, enhancing both my research and my everyday work with patients.

Q: Given your extensive background in radiology, what prompted you to focus on changing how breast cancer screenings are done? 

A: As a fellowship-trained breast imager, my entire clinical practice is dedicated to breast imaging. My work significantly revolves around breast cancer screening, with the goal of detecting cancers as early as possible while managing the inevitable false positives, which sometimes lead to unnecessary biopsies of benign tissue.

My collaboration with Adam, who has a PhD in developing AI models for assessing breast cancer risk based on mammograms, has been a major inspiration. Before joining UCSF and Berkeley, Adam’s work demonstrated that his AI model was more effective than the current standard of care, which relies on questionnaires assessing factors like age, family history, and breast density. While these questionnaires are useful on a population level, they fall short in providing individualized risk assessments and predicting when a person might develop breast cancer.

Our project aims to advance this by building a model that incorporates multiple mammograms over time, along with other imaging modalities such as MRI and 3D mammograms. There is a wealth of information in existing imaging databases that we believe can be leveraged to offer more precise risk assessments. Our goal is to develop a personalized screening plan for each patient, providing a clearer understanding of their risk over one, three, and five years. This would enable patients to have more informed discussions with their healthcare providers about whether additional screenings are necessary and when they should occur. Ultimately, we hope to empower patients and enhance the detection of breast cancer at an earlier stage, particularly for those at higher risk.

Q: How do you plan on accomplishing this project? 

A: Our approach is to build the project incrementally. Adam’s existing model assesses risk using a single mammogram, and our first step is to expand on this by developing a longitudinal model that incorporates multiple mammograms over time. Next, we plan to create a separate model for 3D mammograms (tomosynthesis), including a longitudinal version of this model as well.

Currently, we are working on building these individual models using the UCSF dataset, and we are collaborating with partners across the country to validate our work. By taking these incremental steps, we aim to achieve a robust, longitudinal multimodal model that provides a more accurate risk assessment for breast cancer.

Q: How do you see Pillar changing the future of imaging? 

A: Adam and I believe that Pillar has the potential to significantly enhance risk assessment for breast cancer and transform our approach to supplemental MRI screening. This tool could facilitate more nuanced discussions between patients and their providers. For instance, if Pillar can demonstrate that low-risk patients have a very low likelihood of future cancer, it might be possible to safely extend the interval between mammograms for some patients. However, any changes in screening frequency would require robust evidence to ensure patient safety.

Ultimately, our goal with Pillar is to shift towards more personalized care, providing patients and providers with better information to make informed decisions. By moving away from a one-size-fits-all approach, we can tailor screening recommendations to individual risk profiles. Additionally, we see potential for integrating other types of patient data, such as pathology results from benign biopsies, to further refine our risk prediction models.

Pillar is just the beginning, and we envision a future where these models continuously evolve to incorporate diverse patient information, improving breast cancer risk prediction and patient care.