Despite known clinical value, CAD needs more validation

May 1, 2007

The Elizabeth Wende Breast Clinic in Rochester, NY, has used computer-aided detection for seven years. While CAD's value in the screen-film arena is well documented, its implementation can be challenging, and further research still needs to be performed in the digital world.

The Elizabeth Wende Breast Clinic in Rochester, NY, has used computer-aided detection for seven years. While CAD's value in the screen-film arena is well documented, its implementation can be challenging, and further research still needs to be performed in the digital world.

Medicolegal issues also exist, such as whether a facility must keep a paper print in the patient's chart and whether the radiologist is responsible for the marks should he or she discard the results or save them for later documentation. Many questions and concerns have arisen about possible lawsuits and how to handle the CAD marks.

Screen-film CAD is a lot more work for a clinic or office. When we first implemented CAD for all our screening and diagnostic mammograms, workflow was an issue, as we had to sort, scan, and re-sort the films in order to digitize the images. Our clinic added a staff member to perform this task, and radiologists initially took longer to interpret mammograms with the addition of CAD. The learning curve may differ from radiologist to radiologist.

Digital CAD is easy. You press a button on your workstation, and the CAD marks appear. The digital data are sent automatically to the viewer, and the radiologist can then look at the CAD marks and the mammogram. Digital CAD has not been proven or tested as much as screen-film, but our clinic has performed some interesting retrospective studies presented at the Computer Assisted Radiology and Surgery conferences in Berlin in 2005.1

This review of 81 biopsy-proven invasive lobular cancers revealed that CAD marked 65% of the cancers the year of diagnosis. In these patients, the prior mammograms, when available, were reviewed. In the patients for whom we had priors (31), 26% were marked by CAD the year before diagnosis. At the 2006 RSNA meeting, we presented a series of 111 patients with biopsy-proven malignancy imaged with full-field digital mammography.2 We had CAD available in 73 patients. Overall, CAD marked the cancer in 84% of these cases. It marked 93% of the calcifications, 65% of the masses, and 100% of architectural distortions.3

We use CAD on all our screen-film and digital mammograms. Most insurance carriers didn't reimburse for CAD initially, but soon the studies validating its use appeared, and with them came reimbursement.4,5 As reimbursement became available, using CAD has been an easy decision for the doctors. Standard of care in our clinic since 1996 has been to double-read all screening mammograms. Radiologists double-read the mammogram first and then review the CAD results. CAD doesn't "talk us out of" working up a lesion; rather, it often confirms a worry or concern we have after reading the mammogram.

We always read the films first and then go back and double-check areas that are marked by CAD. It is very helpful for subtle calcifications. Readers have to be aware of the fact that CAD predominantly marks areas of asymmetry or normal breast tissue and areas of overlap that aren't real. Radiologists have to make a conscious decision every time they use CAD regarding what they will pay attention to and what they will ignore. CAD cannot be used as a human reader.

Studies have shown that, in retrospect, a cancer may be seen on prior-year mammograms as much as 70% of the time.6 We performed an internal review of 318 cancers diagnosed in the year 2000, in which 52 could be seen in retrospect on the prior year's films. CAD marked 81% of those cancers the year the cancer was diagnosed by our radiologists and 71% on the prior year's diagnosis.6

But as the radiologist is reading the mammogram in real-time, can CAD talk that reader into working up an area that he or she thinks is normal? A study at the Elizabeth Wende Breast Clinic of approximately 20,000 screening patients, conducted prospectively between 2000 and 2002, revealed a 7% increased detection rate by reviewing CAD results. Radiologists subsequently changed their minds from calling the mammogram negative/normal to recall/workup based on the CAD marks (see figure).7

FUTURE CHALLENGES

Many different digital gantry vendors operate in the mammography arena. Variances in image size, processing, and workstation display are present. In the digital world, one challenge is to display CAD to radiologists in the format they are familiar with. CAD will have to be the front face that normalizes the image size and presentation in this widely varying vendor population.

Reimbursement cuts that will occur in the next few years call for a decrease of over 50% in CAD reimbursement.8 This reduction is going to hurt clinics and outpatient centers that perform only breast imaging, as the profit margin is already so small. As it is, fewer radiologists want to go into breast imaging because of the decreased reimbursement and medicolegal challenges. Further cuts will leave overburdened breast centers and their radiologists even more strapped for funds and time. The volumes of these centers may increase, but they could still go under financially. As radiologists, we need to get the governing bodies to listen to arguments that these reimbursement cuts may be so devastating that patients may not be able to receive breast care.

Refining neural networks research to develop diagnoses specifically in the breast cancer arena will affect the future of CAD. The ability to do comparisons of prior mammograms with current mammograms so a subtle change (a small tumor) can be detected would be useful. If CAD could compare and detect asymmetries or a subtle mammographic change over prior years, that would be extremely helpful. Many invasive lobular cancers present only as increasing densities over time, and CAD may help detect these subtle changes. Currently, we don't have that ability, but vendors are conducting research in this area.

The number of studies showing the value of CAD is growing. As more studies validate the value of CAD, its use may become a standard of care issue to be considered when providing screening mammography.

At present, a lower volume facility will not be able to afford CAD or even digital or computed radiography mammography. We are in an interesting position, as women and their healthcare providers are trying to do everything to ensure their health and early detection of breast abnormalities. Reimbursement cuts, however, may prevent smaller facilities from even having these newer and improved tools available for earlier detection. Some insurance companies go so far as to include CAD as part of their screening programs. As more studies continue to validate the CAD technology and support its value, it may very well become the standard of care along with the screening mammogram, whether highly reimbursable or not.

The author would like to acknowledge her fellow radiologists at her clinical practice: Dr. Wende Logan-Young, Dr. Posy Seifert, Dr. Patricia Somerville, and Dr. Philip Murphy.

Dr. Destounis is a diagnostic radiologist at The Elizabeth Wende Breast Clinic in Rochester, NY, and a clinical associate professor of radiology at the University of Rochester School of Medicine and Dentistry.

References

  • Destounis S. Retrospective study evaluating the accuracy of CAD in marking invasive lobular cancers. Berlin, Germany: Computer Assisted Radiology and Surgery: Proceedings of the 19th International Congress and Exhibition, June 22 to 25, 2005.

  • Destounis S, et al. The role of computer-aided detection with digital mammography and the detection of carcinoma. RSNA, Chicago, Nov. 26-Dec. 1, 2006.

  • Destounis SV, Hanson S, Roehrig J. Computer aided detection marker value and breast density in the detection of invasive lobular carcinoma. Int J Comp Asst Radiol Surg Accepted, publication pending.

  • Burhenne LJ, Wood SA, D'Orsi CJ, et al, Potential contribution of computer-aided detection to the sensitivity of screening mammography. Radiology 2000;215:554-562.

  • Freer T, Ulissey M. Screening mammography with computer-aided detection: prospective study of 12,860 patients in a community breast center. Radiology 2001;220;871-876.

  • Destounis SV, DiNitto P, Logan-Young W, et al. Can computer-aided detection with double reading of screening mammograms help decrease the false-negative rate? Initial experience. Radiology 2004;232(2):578-584.

  • Destounis S, et al. Can computer aided detection (CAD) decrease the false negative rate of consensus double read screening mammography? Chicago: RSNA 2001.

  • Destounis SV. CMS looks to cut CAD reimbursement. Mammography regulation and reimbursement report. 2006;5(10):1-2.