Using Artificial Intelligence to transition from digital mammography screening to digital breast tomosynthesis screening: a retrospective evaluation

Raya-Povedano J L , Romero-Martín S , Elías-Cabot E, Gubern-Merida A, Rodríguez-Ruiz A, Alvarez-Benito M
ECR2021

Aim and Objective
To retrospectively evaluate if a screening strategy using an artificial intelligence (AI) system could allow the transition from digital mammography (DM) to digital breast tomosynthesis (DBT), without an increase in workload.

Materials and Method
The data was retrospectively collected from a previous prospective study (Tomosynthesis Cordoba Screening Trial) that compared DM and DBT in a paired cohort. In total 15987
DM/DBT exams were included from 15986 women (113 cancers).

All of the DBT exams were analyzed using an AI system (Transpara, ScreenPoint Medical) which assigned them a Score from 1 to 10 according to the probability of cancer. An AI
based screening strategy with DBT was simulated: No human reading in scores 1 to 7 and double reading in scores 8 to 10, plus Transpara as a standalone reader (operating
at the same specificity).
Workload, sensitivity and recall rate were compared between the original screening setting that used unaided double reading of DM and the AI-based scenario with DBT using the McNemar test.

Results
DBT screening compared to DM screening could have been carried out a relative workload reduction of 29.7% (156 hours vs 222 hours, (95% CI: 23.8- 36.2), P<0.001). The sensitivity would have been 25% higher (95% CI: 15.8-36.3, P<0.001); with 95 detected cancers through AI-DBT screening (84.1%, 95% CI: 76.2-89.7) and 76 through unaided DM screening (67.3%, 95% CI: 58.2-75.2). The recall rate would have been 27.1% lower (95% CI: 24.1-30.3, P<0.001); with 588 women recalled in AI-DBT screening (3.68%, 95% CI: 3.40- 3.98) and 807 in unaided DM screening (5.05%, 95% CI: 4.72-5.40).

Conclusion
Transpara could allow the transition from DM to DBT screening with a workload reduction.