Breast Cancer. 2022 Aug 24. doi: 10.1007/s12282-022-01396-4. Online ahead of print.
ABSTRACT
BACKGROUND: It remains unclear whether original full-field digital mammograms (DMs) can be replaced with synthesized mammograms in both screening and diagnostic settings. To compare reader performance of artificial intelligence computer-aided detection synthesized mammograms (AI CAD SMs) with that of DM alone or in combination with digital breast tomosynthesis (DBT) images in an experimental setting.
METHODS: We compared the performance of multireader (n = 4) and reading multicase (n = 388), in 84 cancers, 83 biopsy-proven benign lesions, and 221 normal or benign cases with negative results after 1-year follow-up. Each reading was independently interpreted with four reading modes: DM, AI CAD SM, DM + DBT, and AI CAD SM + DBT. The accuracy of probability of malignancy (POM) and five-category ratings were evaluated using areas under the receiver operating characteristic curve (AUC) in the random-reader analysis.
RESULTS: The mean AUC values based on POM for DM, AI CAD SM, DM + DBT, and AI CAD SM + DBT were 0.871, 0.902, 0.895, and 0.909, respectively. The mean AUC of AI CAD SM was significantly higher (P = 0.002) than that of DM. For calcification lesions, the sensitivity of SM and DM did not differ significantly (P = 0.204). The mean AUC for AI CAD SM + DBT was higher than that of DM + DBT (P = 0.082). ROC curves based on the five-category ratings showed similar proximity of the overall performance levels.
CONCLUSIONS: AI CAD SM alone was superior to DM alone. Also, AI CAD SM + DBT was superior to DM + DBT but not statistically significant.
PMID:36001270 | DOI:10.1007/s12282-022-01396-4
