Google’s AI system for detecting breast cancer in mammograms has demonstrated the ability to identify slightly more cancers than human radiologists, capturing cases that doctors initially overlooked. Introduced in 2020, the AI system’s evaluations showed it achieved higher sensitivity while maintaining statistically equivalent specificity compared to human experts. However, the studies highlighted that physician distrust in the AI’s outputs remains a significant barrier to its clinical adoption. By integrating this technology into established protocols, the system not only aimed to improve diagnostic accuracy but also to reduce radiologists’ workloads, performing faster than human counterparts in trials conducted across UK clinics.

iCAD: iCAD specializes in AI-powered solutions for cancer detection and therapy, particularly in breast imaging. It has integrated Google’s mammography AI into its ProFound Breast Health Suite under a long-term agreement aimed at enabling clinical deployment as an independent second reader. The collaboration focuses on supporting radiologists in double-reading protocols for breast cancer screening.
Google: Google develops advanced AI technologies for various applications, including healthcare diagnostics. Its mammography AI system was evaluated in recent multicenter studies for breast cancer detection in real-world UK clinical settings. The company maintains an ongoing partnership with iCAD to commercialize the system for potential use as an independent reader in screening workflows.
Marc Wilson: Marc Wilson is a researcher contributing to AI development for healthcare at Google. He co-authored the 2026 studies on the company’s mammography AI system, examining its performance in identifying cancers and reducing radiologist workload in simulated and live clinical scenarios. His involvement underscores efforts to bridge AI research with practical diagnostic integration.
Christopher J. Kelly: Christopher J. Kelly is a researcher at Google focused on AI applications in medical imaging. He co-led recent studies published in 2026 that tested Google’s breast cancer detection AI across retrospective and prospective evaluations in UK clinics. The work highlights the system’s ability to improve detection while addressing implementation challenges.
University of Surrey: University of Surrey is a UK academic institution with research strengths in health sciences and technology. It contributed to the Google-led studies on AI for mammogram analysis, supporting data analysis and clinical protocol evaluations. The university’s involvement aided in testing the system’s feasibility for broader adoption.
Imperial College London: Imperial College London is a prominent UK research university with expertise in medicine, AI, and data science. It participated in the collaborative studies evaluating Google’s breast cancer AI, providing clinical and methodological support for assessments of diagnostic accuracy and fairness. The institution helped facilitate real-world testing in NHS settings.
National Health Service Breast Screening Centres: National Health Service Breast Screening Centres are UK facilities dedicated to population-based breast cancer screening programs. They provided data and infrastructure for testing Google’s AI system in both historical and live scan evaluations. The centres enabled assessment of how the technology could fit into standard double-reading workflows.
Royal Surrey National Health Service Foundation Trust: Royal Surrey National Health Service Foundation Trust is a UK healthcare provider specializing in cancer services and diagnostics. It collaborated on the prospective and retrospective evaluations of Google’s AI mammography tool, supplying clinical data and expertise from NHS Breast Screening Centres. The trust helped assess real-world integration and performance metrics.

AI Performance: Google’s updated AI system for breast cancer detection demonstrated superior sensitivity compared to initial human readers while maintaining statistically equivalent specificity in multicenter evaluations.
Trust and Adoption: Researchers identified physician distrust in AI outputs as a primary barrier to clinical adoption, emphasizing the need for greater explainability and direct engagement with doctors to build confidence in the technology.
Clinical Integration: The AI tool performed faster than human evaluations in prospective live tests and supported simulated second-reader replacement that could reduce overall radiologist effort despite increased arbitration cases.