RSNA: Poster Presentation
Jiye Kim, Jacqueline Holt, Mirelle Aujero, Bryan Haslam
Abstract
Purpose
With considerable variability in radiologist performance interpreting screening mammograms, one approach to reduce missed cancers is a safeguard review process where AI identifies exams at risk of potential missed cancers, triggering review by a second, experienced breast imager. Here, we sought to test whether cancers detected through this process include aggressive cancers that would particularly benefit from immediate intervention for better prognosis.
Materials and Methods
An AI-driven safeguard review process was implemented prospectively in a community practice from July 2021 – June 2022. A custom-built AI algorithm was used to flag the most suspicious screening DBT exams that had not been recalled by the initial interpreting radiologist. An expert breast imaging specialist performed a second, safeguard review of each of the flagged exams, and consulted with the initial interpreting radiologist for discordant interpretations. Cancers detected during this period were followed up for pathology, nodal status, cancer size, hormonal status, and cancer stage. Cancers were deemed aggressive if an invasive cancer had lymph node involvement, had triple-negative status, or was stage IIA or greater. Patient demographics (e.g., age and race/ethnicity) were also collected.
Results
Out of a total of 40,532 screening exams during this period, 2,296 were flagged by the AI for a. safeguard review. The safeguard reviewer identified 130 of these exams with potential misses, resulting in the detection of 41 cancers. These additional 41 cancers were found on top of the 219 cancers detected without the safeguard process. 43.9% of the additional cancers caught were from patients with dense breasts, 75.6% were invasive, 9.8% were triple-negative, and 12.2% were stage IIA or greater. Together, 22.0% of these cancers were deemed aggressive, compared to 19.2% of the cancers caught without the safeguard process. The cancers detected through the AI-driven safeguard process were from patients with similar demographics (age, race, and ethnicity) to those whose cancers detected without the safeguard process.
Conclusion
The AI-driven safeguard review process resulted in the detection of more breast cancers that would have otherwise been missed. These cancers consisted of mostly invasive cancers with a substantial proportion being aggressive cancers. This result suggests that a combination of modern deep learning AI plus targeted safeguard review is practical and provides significant improvement in detecting important cancers in a community practice.
Clinical Relevance
An AI-driven safeguard process can reduce missed cancers, a substantial proportion of which are aggressive cancers that require early detection and treatment for improved prognosis.
Disclaimers:
Breast Suite comprises multiple applications including ProFound Pro, Breast Density, Safeguard Review, Risk Assessment, BAC and DeepHealth Viewer. DeepHealth Viewer is manufactured by eRAD, Inc. and distributed by DeepHealth, Inc.
Any claims made about Breast Suite may reference claims associated with its individual components.
Not all products and functionalities are commercially available in all countries. For clearance and commercial availability in your geography of functionalities listed and compatibility with other systems, please contact your account manager.