Diagnostic performance of two versions of an artificial intelligence system in interval breast cancer detection.

Abstract

Various versions of artificial intelligence (AI) have been used as a diagnostic tool aid in the diagnosis of breast cancer. One of the most important problems in breast screening progmrams is interval breast cancer (IBC). To compare the diagnostic performance of Transpara v1.6 and v1.7 in the detection of IBC. Reports of screening mammograms of a total 2,248,665 of women were evaluated retrospectively. Of 2,129,486 mammograms reported as Breast Imaging Reporting and Data System (BIRADS) 1 and 2, the IBC group consisted of 323 cases who were diagnosed as having cancer on mammography and were correlated with pathology in second mammogram taken >30 days after first mammogram. Four hundred and forty-one were defined as the control group because they did not change over 2 years. Cancer risk scores of both groups were determined from 1 to 10 with Tranpara v1.6 and v1.7. Diagnostic performances of both versions were evaluated by the receiver operating characteristic curve. Cancer risk scores 1 and 10 in v1.7 increased compared to v1.6 (P < 0.001). In all cases, sensitivity for v1.6 was 56.6%, specificity was 90%, and, for v1.7, sensitivity was 65.9% and specificity was 90%, respectively. In all cases, area under the curve values were 0.812 for v1.6 and 0.856 for v1.7, which was higher in v1.7 (P < 0.001). Diagnostic performance of v1.7 was higher than v1.6 at the 7-12-month period (P < 0.001). The present study showed that Tranpara v1.7 has a higher specificity, sensitivity and diagnostic performance in IBC determination than v1.6. AI systems can be used in breast screening as a secondary or third reader in screening programs.