Dr. Linda Moy from New York University will talk about how her team developed and trained an AI model that they say automatically identified malignant and benign lesions without the need for manual annotations from radiologists.
The model was trained with nearly 289,000 ultrasound exams. This includes over 5.4 million total images from over 143,000 patients between 2012 and 2019.
The model achieved an area under the curve (AUC) of 0.976 for identifying malignant lesions in a test set of nearly 45,000 exams. It also had an AUC of 0.962 on a set of 663 reader study exams, higher than the average radiologist.
"AI decision support improves diagnostic accuracy by decreasing unnecessary biopsies," the team said.
Find out what else the model was able to do in this presentation.
