[Capstone Project Spotlight] Bridging the Radiology Gap: AI Solutions in Mammography
“[This solution can] ease the burden on radiologists, reduce burnout, and improve the overall efficiency, sustainability, and accessibility of breast cancer screening programs.”
Breast imaging is facing a growing challenge: a significant shortage of specialized breast radiologists at a time when the volume of screening mammograms continues to rise. This imbalance is unsustainable, placing mounting strain on radiologists, increasing turnaround times, fueling professional burnout, and risking delays in patient care and diagnostic accuracy. Breast imaging is just one example of a broader challenge across radiology. Imaging volumes are surging across all modalities — CT, MRI, ultrasound — while the number of radiologists has remained largely flat. The gap between supply and demand is widening, threatening the quality and timeliness of patient care. Researchers at UC Berkeley and UC San Francisco led by Professors Adam Yala and Maggie Chung are collectively working to change this reality through innovative solutions.
An AI-Powered Solution
To address the growing mismatch between rising mammography volumes and the limited availability of specialized breast radiologists, Dr. Yala and Dr. Chung’s team is developing AI-powered tools designed to selectively automate parts of the mammogram interpretation workflow. Specifically, their AI is trained to identify a substantial portion of screening exams that are cancer-free with near-perfect accuracy. These low-risk cases can then be safely ruled out without requiring radiologist review. By reducing the number of cases needing human interpretation, their approach significantly lowers the average time and cost per screening. Most importantly, it enables imaging centers to scale their capacity, serving more patients with the same clinical team, while allowing radiologists to focus their expertise on more complex and suspicious exams.
For the exams that still require human interpretation, their team is also developing large vision-language model (VLM)-assisted reporting tools to further reduce radiologist workload. These systems automatically draft reports by synthesizing imaging findings, enabling radiologists to work more efficiently while maintaining high standards of diagnostic quality. While these technologies are still in development within their research lab, the MTM Capstone project has played a critical role in shaping their commercial vision. By combining selective automation with intelligent assistance, their team aims to ease the burden on radiologists, reduce burnout, and improve the overall efficiency, sustainability, and accessibility of breast cancer screening programs.
How This Solution Stands Out
Unlike traditional AI tools that still require radiologists to review every mammogram, this new approach selectively automates low-risk, cancer-free cases with near-perfect accuracy. By ruling out these exams, they are working to sharply reduce the number of images needing human review, allowing radiologists to focus on suspicious cases and enabling imaging centers to handle more patients without expanding staff.
Even for exams that require radiologist interpretation, this new tool goes beyond traditional report templating systems. While conventional templates offer static checklists and dropdowns, the newly-developed vision-language model (VLM)–assisted tools can suggest likely findings directly from the image and complete radiologists’ sentences as they dictate. This creates a faster, more seamless reporting experience, helping radiologists manage growing workloads without sacrificing quality. Together, selective automation and VLM-assisted reporting offer a compelling direction to address the growing imaging volumes.
“They are working to sharply reduce the number of images needing human review, allowing radiologists to focus on suspicious cases and enabling imaging centers to handle more patients without expanding staff”
Collaboration with the Master of Translational Medicine (MTM) Program
MTM students play a critical role in advancing this project. They conduct clinical interviews with radiologists to better understand workflow challenges and refine the value proposition of the solution. They also lead market research and competitive analysis efforts to position these tools strategically within the broader landscape of AI in medical imaging. Currently, they are exploring market sizing, pricing models, and regulatory pathways to inform the team with the go-to-market strategy. Their analyses have been instrumental in shaping the team’s product vision and identifying the right commercial path for their emerging research.
The Vision Behind the Project
Drs. Yala and Chung are professors at UC Berkeley and UCSF, developing AI tools to improve healthcare. Bridging the gap between research and real-world impact is at the core of their mission. “We’re excited about the MTM Capstone Project because it has helped us sharpen our vision, define a path to market, and accelerate the commercial potential of our work,” they explain. In collaboration with MTM, the team works to bring innovative ideas to tangible solutions.
For those interested in learning more about their work, reach out to Drs. Adam Yala and Maggie Chung at yala@berkeley.edu and maggie.chung@ucsf.edu.
While their current focus is on breast imaging, they are developing tools to improve workflows in various fields across radiology.
By developing an AI tool with selective automation, their project offers an innovative solution to rising imaging volumes and radiologist burnout. With continued development and support, these tools have proved to have the potential to improve the efficiency, sustainability, and accessibility of not only breast cancer screening but also radiology more broadly.
