Researchers at UC Berkeley and UCSF have developed and tested an AI model that can identify patients at higher risk for breast cancer, allowing them to receive faster interpretations of mammography results, diagnoses and biopsies.
Researchers used an open-source AI model, called Mirai, developed by Adam Yala, a co-author of the study and an assistant professor in the UCSF-UC Berkeley Joint Program in Computational Precision Health. The model was first tested on prior mammograms, where it identified a number of high-risk patients that wouldn’t completely overwhelm the clinic. In the study, 12.7% of screened patients were flagged by the AI model, close to the 10% target.
At the hospital where it was implemented, the Zuckerberg San Francisco General Hospital and Trauma Center, the average wait time for a diagnostic evaluation decreased from multiple weeks to about an hour.
Maggie Chung, a co-author of the study, is a core faculty member in the joint program. The program connects clinicians, data scientists and public health experts to develop computational tools that will improve healthcare with real-world applications.
The study asserts AI’s ability to both detect disease and change how physicians will approach care in the future. Chung said she hopes it will help clinicians prioritize patients who require more intervention or swifter care. If this approach is utilized across medicine, Chung described more personalized screening and surveillance based on individual risk.
However, according to Chung, there are some operational challenges to consider in implementing this AI-based workflow, especially for hospitals with fewer resources. Still, the percentage of highest-risk patients can change based on what each screening program can handle. Potentially, AI models can flag a smaller number of critical patients, while still identifying more who would benefit from fast-tracked care, Chung said.
“The larger goal is to make breast cancer screening more personalized so that patients receive the right intervention tailored to their risk and needs at the right time,” Chung said in an email.
Building on this study, the workflow will now be evaluated in larger and more diverse conditions; according to Chung, it will be studied to see if it improves cancer detection, follow-up and long-term outcomes for patients.
The AI model Mirai is also being used in an MRI screening study to assess its role in detecting patients who may benefit from supplemental MRI scans, expanding the scope of its use.
“AI should help improve care not only in well-resourced settings, but also for patients who face the greatest barriers to timely diagnosis and treatment,” Chung said.