Workload reduction of digital breast tomosynthesis screening using artificial intelligence and synthetic mammography

Purpose

A synthetic mammogram (SM), equivalent to digital mammography (DM), can be generated from digital breast tomosynthesis (DBT). To achieve the higher sensitivity of DBT, a time-consuming reading is necessary. However, SM is faster to read and might be sufficient in many cases. This study investigates using artificial intelligence (AI) to stratify examinations for reading either SM or DBT to minimise workload and maximise accuracy.

Methods or Background

This is a retrospective study based on double-read paired DM and one-view DBT from the Malmö Breast Tomosynthesis Screening Trial. DBT examinations were analysed with the cancer-detection AI system ScreenPoint Transpara 1.7, resulting in a score (1-10). For low-risk examinations (score 1-6), SM reading was simulated by assuming equality with DM reading. For high-risk examinations (score 7-10), the DBT reading results were used.

Results or Findings

By reading the DBT of 22% (3267/14772) of the cases with highest risk, and SM for the rest, 124 cancers were detected. That is, 31% (29/95) more cancers than with DM double reading, and 92% (124/135) of the cancers detectable with full DM and DBT double reading.

Conclusion

In a DBT-based screening programme, AI could be used to select high-risk cases where reading of DBT is valuable, while SM is sufficient for low-risk cases. Substantially more cancers could be detected compared to DM only, with only a limited increase in reading workload. Prospective studies are necessary.

Limitations

SM results have been approximated with DM results. Interaction between readers and AI was not studied. Single-centre study with a single vendor. Single-view DBT was used.