4 May 2021
José Luis Raya-Povedano et al
AI-based strategies to reduce workload in breast cancer screening with digital mammography and digital breast tomosynthesis: a retrospective evaluation of data collected from the Tomosynthesis Cordoba Screening Trial
Raya-Povedano J L , Romero-Martín S , Elías-Cabot E, Gubern-Merida A, Rodríguez-Ruiz A, Alvarez-Benito M
Aim and Objective
To evaluate if using an artificial intelligence (AI) system could reduce workload without reducing cancer detection in breast cancer screening with digital mammography (DM) or digital breast tomosynthesis (DBT).
Materials and Method
Consecutive screening paired and independently read DM/DBT exams acquired from January 2015 to December 2016 were retrospectively collected from the Tomosynthesis Cordoba Screening Trial. The original reading settings were single or double reading of DM or DBT. An AI system (Transpara, ScreenPoint Medical) computed a cancer risk score for DM and DBT exams independently. Each original setting was compared to a simulated autonomous AI triaging strategy (the least suspicious exams for AI are not human read, the rest are read as in the original setting, and non-recalled exams by radiologists but very suspicious for AI are recalled) in terms of workload, sensitivity, and recall rate. The McNemar test with Bonferroni correction was used for statistical analysis.
15,987 DM/DBT exams (including 98 screen-detected and 15 interval cancers) from 15,986 women (mean age 58 ± 6 years) were evaluated. In comparison with double reading of DBT (568 hours needed, 92/113 cancers detected, 706/15987 recalls), AI with
DBT would result in 72.5% less workload (P<0.001, 156 hours needed), non-inferior sensitivity (95/113 detected cancers, P=0.38) and 16.7% lower recall rate (P<0.001, 588/15987 recalls). Similar results were obtained for AI with DM.
In comparison with the original double reading of DM (222 hours needed, 76/113 cancers detected, 807/15987 recalls), AI with DBT would result in 29.7% less workload (P<0.001), 25.0% higher sensitivity (P<0.001) and 27.1% lower recall rate (P<0.001).
Digital mammography and digital breast tomosynthesis screening strategies based on Transpara AI could reduce workload up to 70%.