Computer-aided detection could hinder image interpretation

Amidst mixed reports about whether CAD improves clinical performance, a study in the October issue of Academic Radiology suggests computer-aided detection could actually hinder image interpretation.

The authors, led by Trafton Drew, Ph.D., at Brigham & Women's Hospital in Boston, found that when readers use CAD they focus more of their attention on areas marked by CAD and spend less time searching for other potential targets. The authors concluded that CAD can "engender a sense of certainty that can lead to incomplete search and elevated chances of missing unmarked stimuli."

According to an article in AuntMinnie, Drew and his fellow authors became interested in investigating the benefits of CAD because of the number of recent studies that have suggested CAD may not improve clinical performance.

In the study, Drew and his colleagues carried out two experiments involving 47 observers. In both experiments half of the observers completed a task without CAD, while the other half used a CAD system that marked 75 percent of targets and 10 of distractors (non-targets).

In one experiment, targets were embedded in a field of noise and CAD was used to aid target detection. In the second experiment the targets and distractors were manipulated in such a way that they were similar to each other and CAD was used to aid target diagnosis. In each case researchers tracked the observers eye-movements as they interpreted the images.

In the first experiment researchers found that while observer sensitivity using CAD was enhanced compared to that without CAD, targets not marked by CAD were missed more often than in the non-CAD group. The second experiment showed no behavioral benefit from CAD and no significant cost on sensitivity to unmarked targets.

In both experiments, those observers using CAD examined a lower total percentage of the search area than the no-CAD observers. This, Drew says in the AuntMinnie article, suggests that CAD can result in an incomplete search for targets by pulling attention away from areas not marked by CAD.

Meanwhile, the jury is still out on whether CAD helps or hinders detection. A study published last month in the journal Investigative Radiology found the software helped reduce the rate of variability for radiologists reading images produced by low-dose CT lung cancer screenings.

A study published in March in the American Journal of Roentgenology found CAD improved the likelihood for radiologists to identify cancer that initially went undetected during a screening. That contrasts with work published in Journal of the National Cancer Institute saying that CAD subjects patients to unnecessary tests.

To learn more:
- read the study
- check out the Aunt Minnie article


The Real Payback of Healthcare Analytics

Tuesday, April 6 | 2pm ET / 11am PT

With the unpredictability of healthcare today, organizations are sharpening their focus on analytics to make more data-informed decisions. Join us for a roundtable session in which thought leaders will discuss how they are maximizing their analytics investments.