The researchers found that the AI system achieved an area under the receiver operating characteristic curve (AUROC) of 0.976 on a test set consisting of 44,755 exams. The AI system achieved a higher AUROC than the average of 10 board-certified breast radiologists in a retrospective reader study (AUROC, 0.962 for AI; 0.924 for radiologists). Radiologists decreased their false-positive rates by 37.3 percent and reduced the number of requested biopsies by 27.8 percent with the help of AI; the same level of sensitivity was maintained. The system was evaluated on an independent external test dataset to confirm its generalizability and achieved an AUROC of 0.927.
Yiqiu Shen, from New York University in New York City, and colleagues curated a dataset consisting of 288,767 ultrasound exams from 143,203 patients examined at NYU Langone Health between 2012 and 2019 to develop and validate an AI system that achieves radiologist-level accuracy in identifying breast cancer in ultrasound images.
“If our efforts to use machine learning as a triaging tool for ultrasound studies prove successful, ultrasound could become a more effective tool in breast cancer screening, especially as an alternative to mammography, and for those with dense breast tissue,” a coauthor said in a statement. “Its future impact on improving women’s breast health could be profound.”