AI and computer-aided detection improves breast cancer diagnosis

Register now

A study evaluating the performance of an AI-based computer-aided detection system showed fast interpretation times and diagnostic accuracy.

The study, at the University of Chicago, was said to be the first of a concurrent-read AI/CAD system. Researchers tested the approach for screening automated breast ultrasound (ABUS) exams.

Results showed that the use of automation reduced reader interpretation times for screening the exams by 33 percent.

Also See: Three-year grant to fund CHOP study of contrast ultrasound

In the study, “Interpretation time using a concurrent-read computer-aided detection system for automated breast ultrasound in breast cancer screening of women with dense breast tissue,” researchers conducted a reader study to compare diagnostic accuracy and interpretation time of screening ABUS for asymptomatic women with dense breast tissue, with and without the use of the concurrent-read CAD system.

The study, published in the American Journal of Radiology, had 18 radiologists interpret 185 screening ABUS studies of women with dense breast tissue. Each reader interpreted each case twice, once without a CAD system and once with the system, from QView Medical. Case interpretation time was recorded in each instance.

The results demonstrate that the use of the concurrent-read CAD system can make interpretation of screening ABUS studies significantly faster without negatively affecting diagnostic accuracy, says Yulei Jiang, associate professor of radiology and principal investigator for the school.

“A reduction in interpretation time would provide little value to radiologists and patients if accuracy suffered,” he adds. “However, sensitivity was preserved, and the results also showed significant specificity improvements in some readers, which could translate to improved clinical confidence as well as productivity gains when used in a clinical environment.”

To improve reader productivity, the QView Medical system provides synthetic 2D images of all six volumetric datasets in a standard ABUS exam to provide an immediately visual overview of the case. The C-thru images, which are minimum intensity projections, summarize each 3D ABUS volume in a 2D image and bring attention to specific areas of interest by enhancement of radial spiculations and retraction patterns in coronal reconstructions, which are highly suggestive of breast cancer in ABUS.

The results are important because clinicians believe ABUS can enhance diagnosis of breast cancer detection in women with dense breasts. However, the interpretation of ABUS exams, which often result in as many as 2,000 images per case, is complex and time consuming, particularly for new users.

For reprint and licensing requests for this article, click here.