ECR 2023
1 March 2023
Authors
E Elías Cabot, S Romero Martin, JL Raya Povedano, A Gubern-Merida, AK Brehl, M Álvarez Benito
Artificial Intelligence (AI) in Breast Cancer Screening Programs in Cordoba (AITIC): Introduction and first interim results
Purpose
To prospectively evaluate Transpara for safe workload reduction by applying double-reading only to high risk cases in screening with 2D and 3D mammography.
Methods and Materials
Participants in a breast cancer screening program in Córdoba (women, age 50-69) are included upon signed informed consent. Two reading strategies are independently applied to all exams: Double blind and non-consensual reading of all exams (control arm) and an AI-based triaging (AI arm), where Transpara (version 1.7.1) evaluates the cancer risk of all exams. Cases identified by AI as low risk are automatically assessed as negative, while cases with intermediate to elevated risk are double read with AI-support. The operating point of Transpara was selected to approximately identify 70% of the exams with lowest cancer risk. Readers are randomly assigned to each reading and blinded to other reading outcomes. We hypothesize that an AI-based screening workflow allows for substantial workload reduction without presenting inferiority in terms of cancer detection (CDR).
Results
Between March and July 2022, 4852 women participated. In total, 42 cancers were detected. AI-based triaging led to equivalent screening performance with non-inferior CDR compared to double-reading of all cases (7.8/1000 (38/42 cancers) vs. 7.2/1000 (35/42), p=1) and similar recall rates (5.7% vs. 6.1%, p=0.5). In the AI arm, 3276 exams were automatically assessed as negative, resulting in 67% workload reduction.
Conclusion
Triaging with Transpara safely reduces workload and increases the overall effectiveness in breast cancer screening.
Limitations
Results are preliminary. Final results are expected after inclusion of 27000 participants by March 2024.