6 July 2021
Pinto MC, Rodriguez-Ruiz A, Pedersen K, Hofvind S, Wicklein J, Kappler S, Mann RM, Sechopoulos I
Impact of artificial intelligence decision support using deep learning on breast cancer screening interpretation with single-view wide-angle digital breast tomosynthesis
The high data volume of digital breast tomosynthesis (DBT) and the lack of agreement regarding its implementation in screening programs makes its use challenging. In a retrospective observer study, we compared radiologist performance when reading single-view wide-angle DBT images with and without an artificial intelligence (AI) system for decision and navigation support, aiming to determine whether AI support (Transpara) could facilitate the implementation of DBT in screening programs.
Materials and Methods:
Bilateral mediolateral oblique examinations and corresponding synthetic two-dimensional images of 190 women (median age: 54, range: 48-63) were acquired between June 2016 and February 2018 with a wide-angle DBT system. Fourteen breast screening radiologists interpreted the 190 DBT examinations (including 90 normals, 26 benigns, and 74 malignant findings), with the reference standard being verified by using histopathologic analysis or at least one year of follow up. Reading was performed with a random mix of examinations being read with and without Transpara as decision and navigation support. Each reader provided BIRADS scores and level of suspicion (1–100) scores per breast. The area under the receiver operating characteristic curve (AUC) and sensitivity and specificity were compared between conditions. Average reading times were compared by using the Wilcoxon signed rank test.
The examination-based reader-averaged AUC was significantly higher when interpreting results with Transpara than when reading unaided (0.88 vs 0.85; p = .01). Also, average sensitivity significantly increased with Transpara support (64 of 74, 86% vs 60 of 74, 81%; p = .006), whereas no differences in specificity (85 of 116, 73.3% vs 83 of 116, 71.6%; p = .48) or reading time (48 seconds vs 45 seconds; p = .35) were detected.
Reading single-view digital breast tomosynthesis (DBT) with the support of Transpara could allow for a more effective screening program with higher performance, especially in terms of an increase in cancers detected, than using single-view DBT alone.