To evaluate the effectiveness and practical implications of a novel workflow of using AI as a supporting reader for the detection of breast cancer in double reading screening mammography.
AI strategies in breast cancer screening should optimise the interaction between AI and human readers to maximise their combined benefit while ensuring patient safety and minimising clinical and operational risks. Large-scale retrospective data is used to evaluate a new paradigm of AI-supported reading. Instead of replacing a human reader, the AI serves as the second reader only if it agrees with the recall/no-recall decision of the first human reader. Otherwise, a second human reader makes an assessment, enacting standard human double reading. 280,594 participants from seven centres in two countries (UK, Hungary), and four hardware vendors (Giotto, Hologic, GE, Siemens) are included. Synthesised performance was measured via superiority/non-inferiority tests on cancer detection rate, recall rate, sensitivity, specificity, and positive predictive value. Workload was measured as arbitration rate and number of cases requiring second human reading.
The novel synthesised workflow was found to be superior or non-inferior on all screening metrics, almost halving arbitration and reducing the number of cases requiring second human reading by up to 87.5% compared to human double reading.
AI as a supporting reader adds a safety net in case of AI discordance while retaining screening performance of standard of care and drastically reducing workload.
The second human reader would only assess cases where the AI and first human reader disagree. The impact of the change in case mix needs to be investigated further.
Ethics committee approval
UK HRA (REC reference: 19/HRA/0376) and ETT-TUKEB (Medical Research Council, Hungary) approval (Reg no: OGYÉI/46651-4/2020).
Funding for this study
Kheiron Medical Technologies