Application of artificial intelligence-based computer-assisted diagnosis on synthetic mammograms from breast tomosynthesis: comparison with digital mammograms

Objective To compare the diagnostic agreement and performances of synthetic and conventional mammograms when artificial intelligence-based computer-assisted diagnosis (AI-CAD) is applied. Material and method From January 2017 to April 2017, 192 patients (mean age 53.7 +/- 11.7 years) diagnosed with 203 breast cancers were enrolled in this retrospective study. All patients underwent digital breast tomosynthesis (DBT) with digital mammograms (DM) simultaneously. Commercial AI-CAD was applied to the reconstructed synthetic mammograms (SM) from DBT and DM respectively and abnormality scores were calculated. We compared the median abnormality scores between DM and SM with the Wilcoxon signed-rank test and used the Bland-Altman analysis to evaluate agreements between the two mammograms and to investigate clinicopathological factors which might affect agreement. Diagnostic performances were compared using an area under the receiver operating characteristic curve (AUC). Result The abnormality scores showed a mean difference (bias) of - 3.26 (95% limits of agreement: - 32.69, 26.18) between the two mammograms by the Bland-Altman analysis. The concordance correlation coefficient was 0.934 (95% CI: 0.92, 0.946), suggesting high reproducibility. SM showed higher abnormality scores in cancer with distortion and occult findings, T1 and N0 cancer, and luminal type cancer than DM (all p <= 0.001). Diagnostic performance did not differ between the mammograms (AUC 0.945 for conventional mammograms, 0.938 for synthetic mammograms, p = 0.499). Conclusion AI-CAD can also work well on synthetic mammograms, showing good agreement and comparable diagnostic performance compared to its application to DM.