AI Assists in Reading of Digital Breast Tomosynthesis Images

Use of a deep learning artificial intelligence (AI) system for digital breast tomosynthesis (DBT) images can improve cancer detection efficacy and reduce image reading time, according to a study published issue of Radiology: Artificial Intelligence.

Emily F. Conant, M.D., from the Perelman School of Medicine at the University of Pennsylvania in Philadelphia, and colleagues developed a deep learning AI system to identify suspicious soft-tissue and calcified lesions in DBT images. They compared the performance of 24 radiologists (13 breast subspecialists) reading 260 DBT examinations (65 cancer cases) with and without AI. The readings occurred during two sessions at least four weeks apart.

The researchers found that based on measurement of the area under the receiver operating characteristic curve (AUC), there was a 0.057 increase in radiologist performance for the detection of malignant lesions with the use of AI, from 0.795 to 0.852. A 52.7 percent decrease was seen in reading time with AI, from 64.1 to 30.4 seconds. There was an increase noted in sensitivity (from 77.0 to 85.0 percent) and specificity (from 62.7 to 69.6 percent) with AI and a decrease in the recall rate for noncancers from 38.0 to 30.9 percent.

"As machine learning methods advance with exposure to larger and larger datasets and the adoption of AI expands, we expect the impact on patient outcomes at the individual level to only improve," the authors write.

Several authors disclosed financial ties to iCAD and Hologic, which partially funded the study.

-----------------------------------------------------------------------
Editor's Note:

If you have any suggestion to the content, 

please email: Julia.Zhang@ubmsinoexpo.com