As computer-aided detection for mammography becomes more common, breast imaging specialists still have concerns about CAD’s utilization—while also noting its clinical and economic potentials.

CAD is proving to be an important, cost-efficient tool for catching more cancers.

Computer-aided detection (CAD) in mammography has been available for several years now, and unlike double reading, it has been reimbursable by Medicare since 2001. Yet many breast imaging radiologists have not embraced the technology in their facilities. Some are wary of increased false positives, while others believe that CAD is an important and cost-efficient tool for catching more cancers.

Aside from Medicare and many insurers, the technology recently has won over prestigious institutions such as Harvard Medical School and Massachusetts General Hospital in Boston, which just replaced their double-reading protocol with a new CAD system. Daniel B. Kopans, MD, a professor of radiology at Harvard Medical School’s Breast Imaging Division and a radiologist at Massachusetts General, still prefers double reading mammograms; however, he also sees CAD’s advantage as an aid to reading mammograms.

“One of the arguments that radiologists are up against is that we should see all cancers, all the time,” said Kopans. “That’s pretty much impossible. Most people don’t see everything, even when they’re trying really hard.”

Kopans uses the example of a person who is searching his home for his keys. Then, another person walks into the same room and immediately spots the keys, which were in plain sight on a table.

“It’s so common that we have the expression ‘It’s right in front of your nose.’ So, the ability to perceive the abnormality on an x-ray is really no different than looking for your keys,” said Kopans.

As to CAD’s effectiveness at discovering more cancers, there have been conflicting studies. A recent retrospective study of 1998 to 2002 statistics that was published in the New England Journal of Medicine suggested that CAD increased the rate of false positives without significantly discovering more invasive cancers. On the other extreme, a prospective study of more than 12,000 cases, published in Radiology in 2001 by Freer and Ulissey, revealed a 20% increase in women who were recalled and 20% more cancers that were detected. Still, Kopans and other radiologists believe that CAD needs further and wider prospective studies to truly prove its accuracy.

One of the New England Journal study’s authors, Carl D’Orsi, MD, at Emery University Hospital in Atlanta, uses a CAD system with digital mammography and does not discount the usefulness of CAD out of hand.

D’Orsi said, “I think that this is an evolving technology, and that’s always the problem with new techniques: they get stamped with either approval or disapproval too early in the game.”

Although Kopans is a supporter of CAD, given the choice, he said he would prefer to double read. “The reason I say that,” said Kopans, “is that when you use CAD, it points to things, but it’s the same radiologist who is reading the mammogram and looks and decides whether to worry about the thing that CAD points to. Whereas, with double reading, you’ve actually got a different brain that may look at the same thing that the first reader saw and dismissed…. The disadvantage of any human reader is that we’re human and we get tired or whatever reason, and we don’t see things. You can be rushed, distracted, etc.”

Asked why Mass General and Harvard chose to replace CAD with double reading, Kopans said that it came down to efficiency and economics.

Double Reading Versus CAD

Since the 1990s, European studies have suggested that double reading may increase the number of cancers detected by some 9% to 15%. However, double reading also comes with extra time and costs. Double reading delays reporting because a second radiologist must find the time to sit down and look at the studies that have been read. In addition, the second reading is not typically reimbursed by payors.

“Having two radiologists look at every mammogram is time-consuming. There are some practices where it’s not even feasible, and it doesn’t get reimbursed. So, in fact, you lose money by doing it,” said Kopans. “Now, because I’m in a teaching hospital and we’re supposed to be leaders and so on, I was able to prevail and my associates were willing to double read free, essentially because we thought it was better care.”

However, as health care expenses increased while the reimbursement for mammography has not kept up with inflation, Kopans said that he and his colleagues began to seriously consider CAD instead of double reading; CAD not only is reimbursed, but also would not siphon off the diagnosis time of a second radiologist.

“There’s benefit in both CAD in increasing sensitivity and a second reader,” said Ellen Mendelson, MD, section chief for breast imaging at Northwestern Memorial Hospital and professor of radiology at the Feinberg School of Medicine, Chicago.

Mendelson’s department does not currently use CAD, but the residents and fellows accomplish double reading, despite the lack of reimbursement. “I think that if we didn’t have two pairs of eyes looking at our cases, I’d feel more pressure to get CAD and to run it at this point,” said Mendelson. “For example, if I were in a private practice with high volume, it would be nice to have another chance to look at the images.” Mendelson also added that when their new facility is completed in the next few years, it will most likely include some sort of CAD system for mammography.

Additional Advantages

Aside from its economic advantage over double reading, CAD’s proponents say that the technology, when used correctly, is a worthwhile aid to increase a radiologist’s sensitivity.

Some experts say that CAD systems—like this Kodak Mammography Computer-Aided Detection System—can be like a second set of eyes.

Mark Lavin, MD, a radiologist at Western Missouri Radiological Group at Centerpoint Medical Center, In-dependence, Mo, prefers using CAD with his recently installed digital mammography system. He said, “With digital, it will pick up microcalcifications that I didn’t see. Most of the time, it’s because they’re vascular or because they’re some kind of benign calcifications. But once in a while, it’ll alert me to something that I would have missed if I hadn’t used the CAD. … It can be a second set of eyes, basically, not so much expertise, but somebody who can see something and say, ‘Is there really a pattern here?’ ”

Mendelson believes that CAD can be most useful for general radiologists who may be interpreting many types of different modalities than for radiologists who practice high-volume breast imaging. “I have heard that opinion expressed at many conferences, and it’s appeared in the literature as well,” she said. “Of course, you need the knowledge of what is significant and what isn’t to apply to your interpretation and for the patient’s management. For just an observational reminder, CAD’s helpful.”

False Positives

Many physicians are concerned with CAD having an increased false-positive and callback rate, as suggested in the aforementioned New England Journal of Medicine retrospective study.

Lead author of the NEJM article, Joshua J. Fenton, MD, MPH, a family physician at the University of California Davis Health System, commented, “There are clear concerns about its effects in recall rates and false-positive rates and biopsy rates. One then has to question whether the benefits of CAD outweigh those harms. So, we need to clarify whether there are benefits to CAD in terms of detecting the most dangerous breast cancers.”

D’Orsi says that the technology lends itself to the risk of increased false positives because CAD essentially casts a wide net to catch as many abnormalities as possible. “The philosophy behind CAD is to throw out a big net to find as many cancers [as we can], but not too big of a net that you’re calling everybody back. So that’s the idea, and that’s why it’s always going to have a relatively higher false-positive rate,” said D’Orsi.

Some physicians say that they are frustrated by CAD’s “big net” and that it causes more time and effort than it is worth.

Kopans, however, believes that CAD’s wide net is really no different than a radiologist’s normal reading process. “As a radiologist, we look at all kinds of things when our brain says, ‘Look at this, look at that.’ Those are conceivably all false positives if we put a mark on the film for everything we looked at, but instead, we dismiss them very quickly.” Similarly, Kopans said that CAD points to a lot of things that a radiologist can dismiss very quickly, but that some may be things they might not have seen, and some may turn out to be cancers.

Because Mass General has been using CAD for only several months, Kopans does not have any current false-positive data. However, he noted that the hospital’s former double-reading protocol slightly increased its false-

positive and callback rate, and generally believes that when CAD is properly implemented, false-positive increases should be minimal.

The Importance of Proper Implementation

Many CAD proponents and skeptics agree that CAD must not be used as a substitute for their own opinions and diagnosis, but rather, only as an aid to detection.

“CAD misses a lot of cancers. So it’s very important for the radiologist to look at the study before using CAD,” said Kopans. “It’s not a substitute for your own eyes and brain.”

Kopans is also troubled when he hears of radiologists who second-guess themselves when CAD does not point to something previously marked by the radiologist, who therefore dismisses the mark. “That’s absolutely crazy,” said Kopans. “CAD is not there to check the radiologist’s interpretation; CAD is there to point things out that might be overlooked by the radiologists.”

Lavin often marks areas that CAD does not catch. He said, “Sometimes you wonder why it doesn’t mark something that looks pretty suspicious. So it’s not rare. That’s why you can’t rely on it, and in fact, I think that’s a good thing, because if it almost always did [mark the same things], then you would rely on it. But if I know that it doesn’t, then I just see CAD as a little bit of extra information, not as anything having to do with the primary read.”

D’Orsi suggests that radiologists be aware of CAD’s strengths and weaknesses. “It’s strongest with calcifications of certain types, and it’s weak with amorphous calcifications. It’s weak with architectural distortions, and it’s weak on soft tissue masses,” said D’Orsi.

Another concern Mendelson has is the way that some CAD programs are moving from aided detection into aided diagnosis. Some companies, for example, are placing a jagged line around a possibly cancerous mass rather than simply marking it with an X. Mendelson feels that this type of feature—which not only prompts the radiologist to look for a mass, but also [suggests] that the mass has a spiculated border—is heading down a slippery slope to interpretation.

“Once you start defining margins as irregularly distinct or what have you, you’re talking about diagnosis there. You can’t read that as benign any more, unless there’s clinically a reason to have a malignancy mimic—a scar, or a granular cell tumor, for example,” said Mendelson.

The Future of CAD for Mammography

Overall, physicians appear to be cautiously optimistic about the technology’s use in mammography, but none feel that CAD is ready to be a standard of care—at least not yet.

“I still feel that CAD is in its infancy,” Kopans said. “I think that the algorithms are okay, but with more powerful computers, the algorithms will need to be made more powerful, as well. With digital mammography, where the images are already electronic, I hope that the companies will do better and better at it.”

Mendelson said, “In general, I think that CAD’s main purpose is as a detection reminder. Did you look at this part of the mammogram? If you did, great, and if you didn’t, maybe you should take a second look right here, and it just prompts you to make certain that you look at certain areas. Will it mark everything? No. We know that. Will it mark things that are totally clinically insignificant? Yes. But we hope there will be fewer of those as the algorithms are refined. So, I think that’s where we stand with mammography and CAD.”

Tor Valenza is a staff writer for Axis Imaging News. For more information, contact .