AJR March 2018

Richard A. Benedikt Intrinsic Imaging San Antonio, TX

“Concurrent Computer-Aided Detection Improves Reading Time of Digital Breast Tomosynthesis and Maintains Interpretation Performance in a Multireader Multicase Study”

Digital breast tomosynthesis (DBT) is more accurate than full-field digital mammography (FFDM) alone but requires a longer reading time. According to a study published in the March 2018 issue of AJR, the use of concurrent computer-aided detection (CAD) with DBT reduced reading time by 29.2% while maintaining reader interpretation performance.

Compared with FFDM alone, DBT plus FFDM results in increased cancer detection rates, lower recall rates, and fewer false-positives. However, reviewing two-view DBT plus two-view FFDM increases both reading time and breast radiation exposure almost twofold when compared with reading FFDM alone. Meaningful improvements for interpreting DBT screening examinations would shorten the reading time of DBT while maintaining its advantages over FFDM.

The authors, led by Richard A. Benedikt of Intrinsic Imaging of San Antonio, Texas, selected 240 cases from women undergoing bilateral screening or diagnostic tomosynthesis examinations, which included craniocaudal FFDM images and mediolateral oblique DBT images. Selection of the 240 cases used in the study was based on breast density, case type, BI-RADS assessment categories, mammographic appearance, and size and histopathologic profile of malignant lesions. A CAD system was developed to detect suspicious soft-tissue densities in DBT planes. Abnormalities are extracted from the plane in which they are detected and blended into the corresponding synthetic image. Twenty radiologists retrospectively reviewed all 240 cases in a multireader, multicase crossover design to compare reading time and performance with and without CAD. The performance of CAD alone was also evaluated.

Meaningful improvements for interpreting DBT screening examinations would shorten the reading time of DBT while maintaining its advantages over FFDM, the authors stated. CAD is used in approximately 83% of all screening digital mammograms to assist radiologists in detecting breast cancer in the United States. A single reading with CAD replacing a second reader for digital mammography has accuracy for cancer detection similar to that of double-reading examinations. However, using CAD as an alternative to a second reader has also been shown to increase reading time by 19% compared with a single reading without CAD. With the increased use of DBT for screening mammograms, there is a potential benefit for a CAD system that reduces overall reading time while maintaining radiologists’ interpretation performance, the authors said.

“Robust clinical trials such as our expansive multireader multicase (MRMC) crossover are vital for both regulatory purposes and to determine market labelling and clinical indications,” Benedikt said in an email interview with AJR InBrief. “MRMC studies are unique as radiologists interpret the same cases retrospectively with and without use of the CAD software and read exams with no opportunity for adjunct imaging or comparison studies. All cases had been assessed by a truthing radiologist. Study cases were collectively cancer rich (up to 20 times the incidence of cancer in true screening population) to enable more extensive statistical analyses.”

In practice, mammographers interpret myriad imaging studies in the course of a workday and perform many other clinical duties including interventional procedures, Benedikt said. “Following regulatory approval, I had the opportunity to speak with radiologists and visit a major mammography center where this software had been installed in clinical settings. I believe there is a significant increase in both productivity and diagnostic confidence with use of this product,” he said.

“I anticipate that continuous improvements in software algorithms will increase both the accuracy and usefulness in DBT interpretations, one of the goals of artificial intelligence (AI) and deep learning (DL) software development,” he added.

Asked about how these developments might increase radiologists’ workload and contribute to the problem of fatigue, Benedikt noted that one set of DBT studies from a single patient and with only one prior DBT screening exam could require review of over 400 images. However, he said, the study authors found that this product enabled radiologists to establish their own novel reading pattern, using CAD assessment concurrently, but consistently more efficiently than without PowerLook Tomo detection software.

“There has been both considerable optimism and consternation in the imaging community regarding AI/DL applications,” he said. “Unequivocally, we found this CAD product to be a helpful tool to both high and low volume mammographers.”

“Our radiologists reported greater ease in use and comfort in interpreting cases as their experience with this specific product increased. PowerLook Tomo detection and other CAD products have little or no utility without the guidance of an experienced radiologist,” Benedikt said.

Do you have a story idea for InPractice?
Contact Mike Mason, Communications Manager, at mmason@arrs.org for more information.