Researchers at Royal Surrey contributed to a paper exploring how Artificial Intelligence (AI) could be used in breast cancer diagnosis.
The research team, led by Google DeepMind, has proposed a system that learns when predictive AI tools can be relied upon for the most accurate interpretation of medical images, and when they should be given to a clinician.
The paper, which was published in the journal Nature Medicine, used data from OPTIMAM, the breast cancer screening image database developed by staff in Royal Surrey’s Medical Physics team in partnership with Cancer Research UK.
OPTIMAM is jointly led by Professor Mark Halling-Brown, Head of Scientific Computing at Royal Surrey, and Professor Ken Young, Head of Research at the Royal Surrey-based National Co-ordinating Centre for the Physics of Mammography (NCCPM).
Prof Halling-Brown said:
“Being able to offer access to OPTIMAM’s huge de-identified data set, enables researchers like the team at Google DeepMind to create hypothetical simulations of clinical processes to test their hypotheses using historic real-world data.
“Royal Surrey has been working in this area for over a decade now. Our work shows the increasing importance that accessibility to large, correctly anonymised datasets has in delivering the best services to patients.”
The paper says that use of the new system resulted in a 25% reduction in false positives for a mammography dataset. In hypothetical simulations where AI was allowed to act autonomously, researchers say the system was able to reduce the number of cases needing to be read by a clinician by 75%.