Population-Based Evaluation of Artificial Intelligence (AI) for Mammography
This multi-site study led by the University of Washington, the VBCSS and collaborators from the Breast Cancer Surveillance Consortium (BCSC), has three aims. The first is to evaluate the performance of five commercially available AI technologies for automated breast cancer screening in a diverse and generalizable screening cohort. To accomplish this, the VBCSS will use previously collected data as well as breast images provided to the us from the University of Vermont Medical Center Breast Imaging Department, along with six other geographically diverse breast imaging registries' data and breast images to externally validate the commercially available AI technologies. Key investigators on the UVM team include Dr. Brian Sprague and Dr. Hannah Perry.
The second aim of this study is to use multi-level analyses to identify targeted approaches for both improving AI performance and incorporating AI into radiologists' clinical workflow. We will evaluate the performance of five commercially available AI technologies based on woman-, exam-, tumor, and radiologist-level characteristics to inform future targeted algorithm training and refinement efforts in a continuous feedback loop. Then we will develop and explore targeted approaches for improving clinical workflow efficiency, including use of AI technologies as standalone tools to safely triage negative exams.
The third and final aim of this project is to determine the long-term benefits, harms, and costs of widespread AI-driven breast cancer screening. Using an established CISNET simulation model, we will compare population-level, long-term benefits, harms, and costs associated with widespread translation of the most up-to-date AI technologies for screening both as a standalone tool and as a second reader to radiologists in the U.S. screening population.
Please click here to access publications associated with this study.
This project is funded by grant R01 CA262023 from the National Institutes of Health (NIH).