Detection of Breast Cancer with Mammography: Effect of an Artificial Intelligence Support System

Rodríguez-Ruiz A, Krupinski E, Mordang J.J, Schilling K, Heywang-Köbrunner S, Sechopoulos I, Mann R
Radiology 2019 290:2, 305-314

Aim and Objective
To compare the breast cancer detection performance of radiologists reading mammograms unaided versus reading supported by the Transpara® artificial intelligence (AI) system.

Materials and Method
Retrospective multi-reader multi-case study with 240 cases from screening (100 cancers, 40 false-positive recalls, 100 normal). 14 radiologists read each case once with Transpara support and once unaided. They assigned to each case a BI-RADS score and a probability of malignancy under both conditions. The area under the receiver operating characteristic curve (AUC, representing radiologists breast cancer detection performance), specificity and sensitivity, and reading time were compared between conditions.

Results
On average, the AUC of radiologists was higher with Transpara support than with unaided reading (0.89 vs 0.87, respectively; P = .002).
Sensitivity increased with Transpara support (+3%, 86% vs 83%; P = .046), specificity trended toward improvement (2%, 79% vs 77%; P = .06).
Reading time per case was similar (unaided, 146 seconds; supported by Transpara, 149 seconds; P = .15).
The AUC of Transpara alone was similar to the average AUC of the radiologists (0.89 vs 0.87).

Conclusion
When using Transpara for support:
– Radiologists improved their breast cancer detection performance in mammograms.
– Sensitivity increases without increasing false positives
– Radiologists do not lengthen their reading time per exam
– Transpara stand-alone performance is comparable to radiologists

Considering a learning curve of radiologists with Transpara, and the further developments of the algorithms of Transpara, it might be possible in the future that radiologists could benefit even more from Transpara support, and that they can reduce their reading time per case.