Computer-aided detection (CAD) software can improve radiologist efficiency when interpreting chest CTs, reducing reading times by as much as 44 percent, University of California, Los Angeles, researchers wrote in Academic Radiology this summer.
The team, led by first author Matthew Brown, PhD, of the Center for Computer Vision and Imaging Biomarkers at UCLA, said in the journal the drive to improve reader sensitivity for smaller nodules, reduce reader variability and minimize radiologist fatigue have motivated the development of automated lung nodule detection algorithms, a handful of which have been successful. Theirs isn’t the first paper to study the utility of those algorithms, Brown et al. said, but it is the first to apply them to everyday clinical workflow.
“The few studies that included reading time in nodule detection showed similar times or modest speed-up in the detection task with CAD, but did not include measurements or characterization as required in clinical radiology reporting,” the authors wrote. “This lack of integration and proven benefit in terms of radiologist efficiency may be a contributing factor to the lack of adoption of CAD in current clinical practice.”
Brown and his colleagues tested CAD’s effectiveness in daily practice by integrating a commercial chest CT CAD software known as Virtual Resident with a commercial radiology report dictation application. The CAD software automatically prepopulated a radiology report template.
“The scope of the study design was limited to whether integrated CAD enables time savings by providing automated detections and measurements that are acceptable to a radiologist,” the researchers wrote. “Given that the readers had to consider whether to accept the prepopulated measurements, and may have elected to delete and remake the measurements, it was possible that CAD ‘assistance’ could have actually increased reading times.”
That’s the problem with most currently available CAD studies, the authors said. Because they’re standalone projects rather than integrated ones, clinicians actually end up spending more time figuring out the system than they would interpreting a study through traditional means.
Brown and co-authors tested the CAD software using 40 scans from a publicly available lung nodule database. Each scan was read twice by three separate radiologists, all of whom read the images with and without the CAD analytics.
CAD assistance reduced reading times by 7 percent to 44 percent relative to conventional methods for the panel of radiologists, the authors reported, suggesting CAD could be successfully integrated into clinical workflow for more efficient imaging.
“At the end of the day, we believe an important metric for adoption of CAD in clinical practice will be the improvement in the efficiency of the radiologist, defined as time spent between opening a case and signing a final report,” Brown et al. said. “It may be that CAD integration and measurement accuracy and acceptance are just as important as sensitivity and specificity in efficiency and adoption.”