Artificial Intelligence (AI) improves breast cancer screening performance for both digital mammography and digital breast tomosynthesis

Purpose

To evaluate the added value of AI for digital mammography (DM) and digital breast tomosynthesis (DBT) reading as support for human double reading in a breast cancer screening program.

Methods and Materials

We included all DM or DBT screening examinations (Selenia and Selenia Dimensions, Hologic, USA) between March 2021 and March 2022. These were double read (without consensus) by radiologists with concurrent AI support. We compared the screening performance in this period with the same 12 months period one year earlier, prior to the implementation of Transpara in clinical practice. Transpara (version 1.7.1) categorises each exam representing the probability of cancer and highlights suspicious areas.

We compared the overall cancer detection rate (CDR), recall rate (RR) and positive predictive value (PPV) during the study period for DM and DBT separately. A Chi2 test was performed for statistical analysis.

Results

We identified 12 000 screening examinations (DM=5028, DBT=6972) that were double-read with concurrent AI-support. Overall, 105 cancers were detected (DM: 44, DBT: 61). Before the introduction of Transpara, 92 cancers were detected (DM: 37, DBT: 55) in overall 16 555 screening examinations (DM: 7229 and DBT: 9326).

For DM, the CDR increased by 71% from 5.1 to 8.7/1000 (p = 0.014) with a concomitant decrease in RR from 8.3 to 6.7% (p < 0.001) and an increase in PPV from 6.1 to 13% (p < 0.001). For DBT, the CDR increased by 50% from 5.8 to 8.7/1000 (p = 0.032) with an increase in RR from 4.5% to 5.7% (p < 0.001) and in PPV from 13.1% to 15.3% (p = 0.37).

Conclusion

Using Transpara concurrently in screening practice contributes to an increase in CDR and PPV for both DM and DBT. A larger improvement is observed for DM. Artificial Intelligence can improve breast cancer screening performance in both DM and DBT.