To demonstrate real-world post-market benefit of artificial intelligence (AI) as an extra reader in breast cancer screening (BCS).
A commercially available AI system was employed as an extra reader (XR) in addition to standard double reading (DR) at a BCS provider in Hungary from Apr-Sept 2021. The XR workflow involved flagging cases the AI suggested to recall which DR did not recall, i.e. positive discordant cases, for arbitration by an experienced radiologist. Detected cancers were pathology-confirmed.
Standard DR for 3746 patients had an arbitration rate of 3.0% (114 patients), a recall rate of 6.7% (250 patients), and a cancer detection rate (CDR) of 12.5/1000 (47 cancer cases). Of the cases that were not recalled by DR, the AI flagged 396 cases to recommend recall (positive discordance rate 10.6%). Extra arbitration resulted in recalling 6 patients, all of whom were diagnosed with breast cancer. This equated to a total arbitration rate of 13.6% and 1.6/1000 increase in CDR (sum 14.1/1000). An exact simulation with a less sensitive AI operating point yielded a total arbitration rate of 5.3%, while still detecting 5 of 6 extra cancer cases.
This real-world deployment of AI increased cancer detection rates without recalling extra false positives, indicating the effectiveness of AI as extra reader. Combining the XR workflow with workflows focused on workload savings will mitigate the increased arbitration rate and optimise clinical and operational benefits. The results provide important, real-world evidence showing the benefit of using an AI reader in breast cancer screening, paving the way for innovative workflows where synergy of humans and AI achieve optimal performance for patients.
Single-centre analysis; Only one extra arbitration reader was included.
Ethics committee approval
Funding for this study
Kheiron Medical Technologies