Studies conflict on value of CAD with mammography
Experts stress: Use CAD appropriately
Depending on which study you're looking at, the use of computer-aided detection (CAD) with mammography improves accuracy slightly, boosts inaccuracy slightly, or does little to impact accuracy either way. While clinicians may be scratching their heads over these mixed results, investigators suggest that they may be explained in part by study design as well as by how study participants actually used CAD in their practice. The experience level of the radiologist, both in clinical practice as well as in the use of CAD, may impact the results as well.
Some of the latest data regarding the use of CAD with mammography comes from Salt Lake City, UT-based Intermountain Healthcare, a group of hospitals and clinics that serves more than half of Utah. To determine whether to make an investment in CAD, a study team, lead by Brett Parkinson, MD, the imaging director for breast care services at Intermountain, looked at the mammographic findings on more than 43,000 women — first without CAD, and then with CAD — to determine whether the addition of CAD improved accuracy.
In the study, participating radiologists would look at each screening mammogram. They then would have to state whether they thought the mammogram was normal, or whether it was abnormal and the woman needed to be called back. In every case, this first impression would be recorded by a scribe, and then the radiologists would view the mammogram again with the addition of CAD. "We would have to commit ourselves before the addition of CAD to make sure it was an honest assessment. After we saw what the CAD marks were, then we would have a second assessment," explains Parkinson.
What the investigators ultimately determined was that the use of CAD resulted in their finding a total of seven early-stage malignancies that otherwise would have been missed. "For every cancer that we found, we had to call back 24 additional people, which we did not feel was an undue burden to the system," says Parkinson, noting that the callback rate was slightly higher than the year before. "We felt that for us the study showed that CAD was worthwhile."
In another study, researchers in Charlotte, NC, found that the single reading of screening mammograms with CAD is more efficient and yields a higher sensitivity than the first reading in a double-reading program.1 The study, by Matthew Gromet, JD, MD, a radiologist at Charlotte Radiology, involved a review of more than 230,000 mammograms that were interpreted in a community-based mammography program between 2001 and 2005. About half were read by sub-specialized mammographers a single time with CAD. The other half were read first by a sub-specialized mammographer and then a second time by a specialist or a general radiologist who is certified in mammography.
Gromet concluded that given the manpower and cost constraints that limit the use of double-reading, CAD appears to be an effective and more widely accessible alternative that also provides a lower recall rate than double-reading. He noted that that the recall rate of mammography with CAD was only slightly higher than the recall rate associated with a single reading without CAD.
The findings of Parkinson and Gromet are in contrast to the results of a large study, published in the New England Journal of Medicine in April 2007, which suggested that the use of CAD results in reduced accuracy in the interpretation of screening mammograms.2 That study considered the use of CAD with mammography at 43 facilities in three states between 1998 and 2002. Investigators considered the results of more than 429,000 mammograms (some conducted with, and some conducted without CAD) to reach their conclusion that an increased rate of biopsy with CAD is not associated with improved cancer detection.
Accuracy in the study was measured in a statistical sense, says Joshua Fenton, MD, MPH, the lead author of the study and a family physician and researcher at the University of California, Davis, in Sacramento, CA. "In particular, it was the likelihood that a radiologist, when presented with one mammogram with cancer and one mammogram without cancer, would classify those mammograms accurately," Fenton says. "What we showed was that, on average, radiologists that are using CAD are more likely to make the wrong classification in those two mammograms."
In particular, Fenton points out that the rate of false positives, leading to callbacks when there was no cancer detected, was 30% higher at the facilities after they installed CAD. Further, the increased rate of biopsy with the use of CAD did not result in improved detection of cancer.
Fenton acknowledges that the results from his study are in conflict with some earlier studies that showed that CAD can improve the detection of early-stage malignancies without significantly boosting the recall rate.3 And he speculates that the mixed results may have something to do with the way in which different radiology practices actually use CAD tools. "If it is true that radiologists in the community are using CAD in sort of an idiosyncratic way — for example, either looking at the output before making a BI-RADS [Breast Imaging Reporting and Data System] assessment or using the output to sort of reassure themselves — then the effect of CAD on performance could be quite unexpected," Fenton says.
Parkinson agrees that if CAD isn't used appropriately — as an adjunct to the radiologist's assessment, but never as a substitute — then it clearly can do more harm than good.
"Even if you look at the results of our study, we found that CAD missed quite a few cancers that our radiologists saw," he says. "There was a significantly greater number of cancers missed by CAD than was missed by us, the radiologists, so to rely on CAD first could actually cause you to miss cancers."
This is where experience may play a role in determining how effective CAD is in improving cancer detection rates, according to Parkinson. He points out that an inexperienced radiologist who lacks confidence in his own assessment may rely on the CAD findings more than he should.
"The only thing that CAD is going to do, and that CAD should do, is assist you in calling somebody back that you may not have called back. So it will increase your callback rate a little bit," Parkinson says. "It should not decrease your callback rate because if it decreases your callback rate, then you are using it as a crutch. Never ignore one of your own judgments, because if you do ignore that judgment, you will miss things."
References
- Gromet M. Comparison of computer-aided detection to double reading of screening mammograms: Review of 231,221 mammograms. Amer J Roentgen 2008; 190:1-6.
- Fenton J, Taplin S, Carney P, et al. Influence of computer-aided detection on performance of screening mammography. New Eng J Med 2007; 356:1,399-1,409.
- Freer T, Ulissey M. Screening mammography with computer-aided detection: prospective study of 12,860 patients in a community breast center. Radiology 2001; 220:781-786.
You have reached your article limit for the month. Subscribe now to access this article plus other member-only content.
- Award-winning Medical Content
- Latest Advances & Development in Medicine
- Unbiased Content