CAD software marks potential masses with ovals and microcalcifications with rectangles. Source: iCad

When computer-aided detection (CAD) technology for mammography first came to market, some radiologists had an almost visceral reaction against the software. Their concern was that these applications would supplant their role as diagnostician. However, as CAD achieved greater market penetration, many radiologists realized that the technology was not a threat but a powerful tool to aid their practice.

“Complex relationships exist between humans and computers,” says Robyn Birdwell, MD, from the department of radiology at Brigham and Women’s Hospital in Boston. Birdwell, writing in the October issue of Radiology, believes the preponderance of clinical evidence suggests that computer-aided detection (CAD) is a useful tool for a radiologist reading mammograms.

Yet, as she points out in her article, CAD can raise a number of issues for radiologists, ranging from sheer annoyance at the number of marks presented by the application that have to be dismissed, to the question of whether the “barrage” of false-positive marks limits the ability of a reader to see any tangible benefit from the software.

“It is only from the carefully performed clinical studies [prospective sequential read or historical control studies]…that the true impact of CAD be measured,” writes Birdwell.

Laurie Margolies, MD, director of breast imaging at Mount Sinai Medical Center in New York City, agrees that the preponderance of the literature in the field suggests that CAD provides enough of a clinical benefit that if it hasn’t already been widely adopted, it soon will be.

“The majority of the literature shows a benefit to CAD with an increased detection of cancers,” says Margolies, whose department uses iCAD SecondLook version 7.2. “So it is helpful.”

Further north in Rochester, the Elizabeth Wende Breast Care imaging center happened to be an early adopter of CAD.

“We’re breast imagers,” says Stamatia Destounis, MD, “That’s what we do. A lot of us aren’t in academic facilities—we’re in private practices and we’re reading a lot of screenings, whether they’re online, or batch reading, or in large volumes, and we’re looking for very subtle changes in mammograms from year to year.

“Breast cancer very early on can be very difficult and hard to find,” adds Destounis, “so when the Freer study came out [an article in the September 2001 issue of Radiology by Timothy Freer, MD, and Michael Ulissey, MD, that reported on a study that found 19.5 percent more cancers with CAD than without], it really made people stop and take a second look at CAD.”

So, as she puts it, Elizabeth Wende Breast Clinic “jumped on the CAD bandwagon early.”

Destounis and her colleagues at the center did their own study, published in the August 2004 issue of Radiology, that found the use of CAD helped detect breast cancers earlier. She says they found a lot of cancers they would have missed without CAD, particularly subtle calcifications that could be early cancer.

Where are the limitations?
Destounis says CAD is particularly useful in practices like hers where she screens large populations of younger women with dense breasts and difficult patterns. But she’s also careful to point out that despite the evidence of CAD’s usefulness in helping physicians identify breast cancer, she understands its limitations as well.

“It can’t do the reading for you,” says Destounis. “It needs to be used as an additional tool like everything else you have when you’re doing breast imaging, like ultrasound or biopsy.”

Margolies agrees with Destounis that CAD is a tool that a radiologist should use as one would use a magnifying glass, “or click a button on a mouse to invert the color or change the background of a digital image.” That said, she believes that a major impediment to wider adoption of the technology is the need for physicians to understand exactly how it works and how it should be used to maximize its benefits.

In the same October 2009 issue of Radiology in which Birdwell’s article appeared, Liane Philpotts, MD, department of diagnostic radiology at Yale University School of Medicine, posed the question of whether CAD could actually be detrimental to mammographic interpretation.

Philpotts argues that while CAD has potential benefits, it has important and “potentially serious limitations” as well. She suggests that in order to avoid any detrimental effect from using CAD on screening interpretation, a physician should:
Never use CAD as a prescreener.

Never use CAD as an initial step in mammogram interpretation.

Use CAD as a last step in a reading protocol.

Destounis agrees there is a “correct” way to use CAD. “We read the mammo first, then look at the CAD marks,” she says. “Then we go back and look at the mammo again to see if there is anything there that you need to change your mind about, if your initial impression was that everything was normal.”

Most of the time, she adds, the marks are discounted by looking at them a second time.

Margolies points out that the most notorious downside of using CAD—its false/positive rate and the number of marks physicians need to disregard before finding a useful one—“can, I think, lead people to discard things that they really shouldn’t be discarding.”

Destounis puts forward a case in which a patient develops cancer and a look back at previous film finds that CAD marked the cancerous area, but that it was never worked up.

“Why,” she asks. “Perhaps it was below the threshold of something [the readers] would have been concerned about, or there were so many other false marks on that same breast, and on the same view, that all the marks were discounted, even the one they should have paid attention to.”

The specificity is “just not there” with CAD, Destounis says, and when you have radiologists who may not be experienced with it, “what could happen is that either they are going to work up everything, or they are just not going to work up anything.”
The future

“Everybody wants the same improvements,” Destounis says. “We want to mark the cancers and not the false/positives.” What she would like to see is that “computer-aided detection move toward computer-aided diagnosis.”

Margolies would like to see significant improvement in specificity. “I also would like to see CAD be able to look at old mammography films so it can make comparisons and look for changes,” she says. “And it might be nice to know why CAD has marked a particular area—how suspicious is the computer and why is it suspicious?”

Economic factors also could impact CAD’s future, says Destounis. Concerns about reimbursement rates could suppress the willingness of physicians to adopt CAD if they haven’t already done so, she speculates. However, as for practices like hers that were early adopters, “we’re not going to get rid of it. We’re used to having it and we’re going to continue to use it.”

After all, many breast imagers do find comfort in the support CAD gives them in interpreting mammograms. “Having a system to aid the human eye that does not take vacations, is not vulnerable to fatigue or environmental distractions, is without emotion, and is designed specifically to assist the very human eye to “look over here” seems like a good idea,” writes Birdwell in her October Radiology article. “Speaking for myself, in this busy and emotional area of imaging, I am happy to take all the help I can get.”