 systems, Transpara 1.7.0 and ProFound AI 3.0, demonstrated high efficacy in detecting breast cancer using digital breast tomosynthesis (DBT).){kind=link}
A study published in RSNA, Radiology: Imaging Cancer, reveals that two commercially available artificial intelligence (AI) systems, Transpara 1.7.0 and ProFound AI 3.0, demonstrated high efficacy in detecting breast cancer using digital breast tomosynthesis (DBT). However, despite their impressive performance, these AI systems still fall short compared to the gold standard of human double-reading by radiologists.
The retrospective study analyzed mammography data from 419 asymptomatic female patients, with a median age of 60, who underwent DBT between 2019 and 2020. Among these patients, 58 had histologically confirmed breast cancer. The performance of the AI systems was evaluated using receiver operating characteristic (ROC) analysis, focusing on the area under the ROC curve (AUC) to assess their accuracy in detecting malignancies. Results were benchmarked against the Breast Imaging Reporting and Data System (BI-RADS) results from human double-reading.
The AUC for detecting malignancy was 0.86 for Transpara, 0.93 for ProFound AI, and an outstanding 0.98 for human double-reading. Transpara, with a rule-out score of 7 or lower, achieved 100% sensitivity and 60.9% specificity. For a rule-in score higher than 9, it showed 96.6% sensitivity and 78.1% specificity. ProFound AI, with a rule-out score lower than 51, also achieved 100% sensitivity and 67.0% specificity. For a rule-in score higher than 69, it displayed 93.1% sensitivity and 82.0% specificity.
While both Transpara and ProFound AI showcased high accuracy in breast cancer detection, they were outperformed by radiologists using double-reading methods. This study underscores the potential of AI in enhancing breast cancer screening but also highlights the continued need for human expertise in ensuring the highest diagnostic accuracy.