Why Mammography CAD Matters

Western Missouri Radiological Group for Encompass Medical Group is using the Kodak Carestream CAD for mammography system with film mammography studies.

Computer assisted detection for mammography first entered the scene in the late 1990s. Approval by the FDA with subsequent granting of increased reimbursement for the use of CAD by Medicare launched adoption of this technology into hospitals and imaging centers across North America. Many subsequent studies have proven its worth in increased detection rates of breast cancers at an earlier stage, many of which could otherwise be missed.  

CAD’s value, however, is now being challenged by some. The publication in the April 5th issue of the New England Journal of Medicine of a large, multi-center study that reported that the use of CAD reduced accuracy of mammographic interpretation sent shockwaves through women’s imaging centers and industry, and has sparked lively debate about the study’s results. “Influence of Computer-Aided Detection on Performance of Screening Mammography” by Fenton, et al has raised numerous questions about the use of CAD in mammography.

This study “linked data from surveys that were mailed to mammography facilities and affiliated radiologists to data on mammograms and cancer outcomes for women screened between 1999 and 2002.” It involved a total of 43 facilities across three states, all members of the Cancer Surveillance Consortium and compiled data from a total of 429,000 screening mammograms performed on 222,000 women between 1998 and 2002. The results suggested that within the seven institutions where CAD was implemented during the study period, not only did CAD not improve the detection of breast cancer; it resulted in a higher rate of false-positive mammogram results and increased the recall and biopsy rate. The study did, however, suggest that CAD could be helpful in finding ductal carcinoma in situ.

Not surprisingly, experienced mammographers who have used CAD for many years have questioned some of the study findings. Issues such as lack of CAD utilization experience in those radiologists who were using newly installed systems, tumor size at the point of detection, and the age of the women at time of diagnosis have been raised. Even the study authors acknowledge that since “computer-aided detection is used in the screening of millions of healthy women, larger studies are needed to judge more precisely whether benefits of routine use of computer-aided detection outweigh its harms.”


CAD finds smaller lesions

Judy C. Dean, MD, a radiologist in private practice in Santa Barbara, Calif., observes that this study spanned years that would necessarily mean all of the sites were using an older version of CAD because the seven institutions who provided CAD data had installed the equipment during the study period (the average amount of time the facilities had been using CAD was seven months). While those CAD applications were not inferior in quality, the newer versions of CAD, especially those that are used with FFDM systems, include refined algorithms.

When Dean’s practice first installed CAD in 2002, she preferred to read cases, make a decision as to whether or not to recall a patient, wrote that conclusion down, then applied the CAD marks to the image. Following that procedure, she then tracked the number of additional patients who were recalled due to the results provided by CAD as opposed to those who would have been recalled based on her initial interpretation.

“Recall rates are impacted the most when you first start using [CAD],” she notes. However, they found that recall rate with CAD decreased with experience. The recall rate in her practice doubled during the first two months of use, but then returned to a modest 9 percent above the pre-CAD rate. Dean cites another study published in Radiology in 2006 that revealed that women would prefer to be called back for more procedures if it might result in earlier detection of cancer.

More importantly, Dean found that in a prospective evaluation of CAD that they were detecting significantly smaller cancers. Invasive, non-palpable breast cancers detected with CAD had a mean size of 5 mm, compared with 10.5 mm for similar cancers detected without CAD, which she assesses to be a huge improvement since early detection improves outcome. While the Fenton study did not provide data about the size and stage of cancers detected, Dean considers that to be of utmost importance.  

The most significant issue in the use of CAD relates to the manner in which it is applied. Experts recommend the radiologist should first read the images to detect any abnormalities. Then CAD marks should be applied, and if there is anything marked by CAD, those areas should receive special attention. “You can never use CAD to dismiss a finding,” she asserts. If in the initial review, the radiologist sees something that CAD fails to mark, there should be no questioning of the initial finding. Experienced radiologists must always interpret first and then look to see if CAD adds anything deserving of closer observation.

One of the other issues from the study she raises is the increased biopsy rate. If CAD results suggest that a patient should be recalled for further evaluation, and perhaps intervention, usually additional studies and further interpretation would determine the course of action. Therefore, if a biopsy were indicated, that would be due to the additional work-up such as ultrasound or MRI, not the CAD markings. 

Dean, who uses the Siemens Novation Full Field Digital Mammography (FFDM) system with iCAD Second Look loaded onto the workstation, suggests that when physicians are experienced mammographers, CAD will only make them better. “The thing I like most about it is that you never miss microcalcifications, and many of the small invasive cancers are found because of microcalcifications.” (The Fenton study did find that CAD is relatively more sensitive in detecting microcalcifications than in detecting masses.)

Tommy Cupples, MD, a CAD researcher and medical director of Women’s Health at ImageCare in Columbia, S.C., has been using CAD since 1998. By early the following year, they began reviewing their data and auditing their results compared to previous, pre-CAD years. At that point, they were completing between 12,000 to 15,000 mammograms per year, with about 8,000 to 10,000 of those being screening studies using film with analog CAD marks.

Cupples published the audits of their regional mammography CAD use in the American Journal of Radiology in October 2005, where they reviewed all positive biopsy results including confirming the histopathology with studies with CAD from 1999 and 2000 compared with their historical control data. Considering that detecting smaller lesions in younger women were the two most important aspects in terms of patient outcome, they focused attention on those two parameters.

While they did experience an increase in recall rate (from 7.7 percent to 8.3 percent) and biopsy rate (from 1.4 percent to 1.5 percent), they felt those factors were offset by an “increased detection rate, younger age at diagnosis and significantly earlier stage of invasive cancer detection…Our findings support the hypothesis that screening with CAD significantly improves detection of the specific cancer morphologies that CAD algorithms were designed to detect.”

They found smaller lesions that often were more subtle in presentation. “We found that we had a 164 percent increase in the detection of cancers that were 1 cm or less in size, and that there was a 5.3 year decrease in age at diagnosis in their CAD population. Those two findings are not typically reported.” In addition, Cupples asserts that it is illogical to believe that one could lower the threshold and find earlier cancers without some degradation of specificity or effect on the biopsy rate.

He concludes that by CAD detecting microinvasive T1a cancers that are 5 mm or less in size, they would not be detected as a mass because they simply are not large enough to create a mass. They found that about 38 percent of their invasive malignancies had some calcifications in them. Cupples uses a Siemens Novation FFDM system with Hologic CAD.


CAD with analog images

Mark Lavin, MD, staff radiologist at Western Missouri Radiological Group for Encompass Medical Group is using the Kodak Carestream CAD for mammography system with film mammography studies. The original films are couriered to a central location for reading. Technologists scan the films, and submit a print copy (not diagnostic quality) with CAD marks applied, in addition to the film to the radiologists for their interpretation.

His workflow includes reading the film first to look for abnormalities, then reviewing the CAD print for marked areas. He then looks back at the film to determine whether the markings denote any potential pathologic finding. “CAD marks calcifications, but it cannot distinguish vascular calcifications where I would not be concerned from calcifications that could be pathologic.”

While he describes this system as slowing down his reading, he believes it is well worth the extra time to make sure they do not miss something. They store the CAD print in the same jacket as the film. They place annotations on the film with anything that looks questionable that they want the next radiologist to make certain to review. He explains that with the 30 to 40 studies they read per day, they never read from the print copy, only from the film.


Further evaluation indicated

Since early detection of breast cancers has been shown to clearly improve outcomes, some might argue that if CAD can find smaller lesions and thus save more lives, it could be considered worth the cost of more recalled patients. The recently published study that questions CAD’s benefits has increased dialogue about these issues—but believers still value CAD’s contribution to breast cancer detection and future studies will bring more details to light.