Use of CAD abets increased appreciation as incidence rates for breast cancer inch upward
BY DEBORAH R. DAKINS
The road ahead in breast imaging can only benefit from the clinical and technological improvements being made in mammography, specifically screening. Despite its limitations?and there are several?film mammography remains the gold standard against which all new breast imaging techniques are measured.
As recently as March, the Institute of Medicine reinforced the primary role of film mammography in breast cancer. But the panel also acknowledged that many other technologies have emerged that offer clinical advantages and deserve further study.
Two of the most promising are full-field digital mammography (FFDM) and computer-aided detection (CAD). While the contributions of FFDM have so far remained largely technical, CAD has already been shown to improve breast cancer diagnosis.
"We know that if you have a second reader to interpret a mammogram, there is a 15% increase in breast cancer detection," said Dr. Rachel Brem, director of breast imaging and intervention at Georgetown University. "The expectation is that CAD will be able to take the place of that second reader and decrease the number of breast cancers not initially appreciated on a mammogram."
CAD has also received an economic boost that could influence its practical use. In July, the American Medical Association approved a Category I CPT code for use of CAD in conjunction with screening mammography. The code will enable radiologists to bill for and receive supplemental reimbursement, beginning in January 2002, for CAD, beyond what they already receive for performing and interpreting a screening mammogram.
"CAD takes a little more time and effort," Brem said. "But now that there's reimbursement from Medicare, the hope is that we use CAD to create improved diagnosis of breast cancer that is fiscally reasonable and attainable. Before the reimbursement code, using CAD involved a financial investment without getting anything back, which made it more difficult to justify."
BATTLE IN THE BREAST
As much as mammography has improved over the years, it's clear that it could use the help of complementary tools like CAD. After rising about 4% per year in the '80s, breast cancer incidence rates continue to increase slightly, according to the American Cancer Society.
An estimated 192,000 new cases of invasive breast cancer are expected to occur in the U.S. this year. In addition to these cases, 46,400 new cases of in situ breast cancer are expected to occur during 2001. About 40,600 women will likely die this year from breast cancer. It's estimated that 34 million screening mammograms and 3.3 million diagnostic mammograms will be performed in the U.S. this year.
 |
CAD systems rely on computer algorithms and neural
networks to highlight suspicious microcalcifications and masses on digitized
mammograms, as shown on this laser-printed report produced by the CADx Second
Look system. Second Look is currently for investigational use only and not
available for sale in the U.S. |
The high sensitivity of screen-film imaging makes it a powerful diagnostic tool for early detection of nonpalpable breast cancer. Randomized clinical trials have demonstrated that screening mammography can identify small, nonpalpable breast cancers and reduce breast cancer mortality by 29% to 45%. Most lesions detected by mammography are benign, and the positive predictive value of mammography for breast cancer ranges from 10% to 40%, depending upon age and type of population examined.
But mammography, despite its gold standard status, has inherent limitations. While the sensitivity of screen-film is reported to range between 80% and 85%, it may be considerably lower in dense breasts. Researchers have long been troubled by a perplexing problem with mammography: perceptible findings so nonspecific that women who present with them are not recalled for immediate follow-up and then eventually develop breast cancer.
The types of lesions that mammography is most likely to miss have been the subject of much study. Calcifications and mass lesions with spiculated or irregular edges are the primary culprits, reported Dr. Robyn Birdwell, an assistant professor of radiology at Stanford University, in a study published earlier this year (Radiology 2001;219:192-202).
The multicenter study was based on a retrospective review of 1083 consecutive cases of breast cancer detected at screening mammography. Prior mammograms available in 427 cases were evaluated, and 286 cases had lesions visible in retrospect, according to Birdwell. Of these, 110 patients were diagnosed with 115 cancers.
The value of CAD was made evident in this study. Birdwell and colleagues found that when computer-aided detection was applied to the relevant prior mammograms, CAD picked up 86% of the calcifications and 73% of the missed masses. The findings dovetail with other studies, including those conducted at Georgetown.
"A number of studies show that CAD improves detection of breast cancer," Brem said. "What is less well known is that radiologists can also detect cancer early by using CAD. When used in studies to look at the previous year's films of women with mammographically detected breast cancer, CAD was able to detect the cancer in 25% of the cases. So there is a lot of potential."
In a study presented at the 2000 RSNA meeting, Brem and colleagues reported a 92% sensitivity in detecting breast lesions of 5 mm or less. The team also reported a sensitivity of 84% for breast masses, 98% for microcalcifications, and 89% for mixed mass/ microcalcification lesions.
So far, the FDA has approved only one commercial device for breast cancer CAD; 140 sites are equipped with the commercially approved CAD device developed by R2 Technologies, and an estimated 1.25 million women have had their mammograms reviewed using the device. Clinical trials being conducted by Brem, Dr. Janet Baum of Beth Israel Deaconess Medical Center in Boston, and Dr. Wendie Berg at the University of Maryland, among others, may result in FDA approval of another device, called Second Look, marketed by CADx Medical Systems.
TWEAKING CAD
 |
Radiologists may find cancers earlier when aided by
CAD. In one study, CAD was applied to year-old mammograms of women with
clinically proven breast cancer, and it detected 25% of lesions initially missed
on mammography. |
The study, conducted by Dr. Carl J. Vyborny, an assistant professor of radiology at the University of Chicago, reviewed 677 consecutive cases of breast cancer to determine presence of spiculation and to assess the sensitivity of a CAD algorithm that measures mass spiculation as a way to detect such lesions.
Radiologists interpreted 55% of the masses as spiculated. When the CAD algorithm was applied, it correctly marked 322 of the 375 clearly spiculated masses (86%). With a looser definition of spiculation, 585 of the 677 masses (86%) were called spiculated by at least one radiologist on one view. The algorithm correctly marked 464 (79%) of the 585 lesions as spiculated or possibly spiculated.
Studies like these help increase CAD's specificity, which has been its weak point. Pattern-matching algorithms have traditionally marked everything as suspicious, leading radiologists to complain that CAD systems did little more than redirect them to do a second read. But the technology has improved markedly, said Vyborny, who spoke on CAD advances at the Society for Computer Applications in Radiology annual meeting in May. In the hands of a trained radiologist, CAD systems can now detect 90% of all malignancies, while producing as few as 0.5 marks per image.
"In 25% of cases, CAD produces no marks at all," Vyborny said. "And the chance that a cancer is present but the computer makes no mark is only one in 10,000."
Such findings may prove important in winning over CAD skeptics such as the Institute of Medicine. In its March report, the IOM noted that while CAD holds the potential to decrease false-positive rates, more extensive studies are called for.
"We need to ensure that CAD does not lead to more false-positive or -negative results and to define more clearly the value and appropriate use of this technology," said Dr. Joyce Lashof, chair of the IOM committee that authored the report, Mammography and Beyond: Developing Technologies for Early Detection of Breast Cancer.
INTERPRETIVE CHALLENGE
Balancing high sensitivity for abnormalities with the need to keep recall rates low is one of the greatest challenges in interpreting screening mammograms. It's estimated that 20% to 40% of breast cancer goes undetected at screening due in large part to the high volume of exams (of which about 95% are expected to be normal) and resulting radiologist fatigue.
 |
The pairing of digital mammography and CAD will
likely boost both technologies. Similarly, the integration of information
systems that link images to patient history is expected to allow advanced CAD
applications such as data mining?interpreting and tracking clinical findings in
specific populations to direct future research and development. |
Numerous studies affirm the value of having prior mammograms. In one paper presented at the American Roentgen Ray Society meeting in May, researchers found that evaluating prior exams reduces recall rates and increases biopsy yield. Patients with a prior exam on file had a recall rate of 3.9%, compared with 6% of patients without a prior exam, according to Dr. Elizabeth Burnside, an assistant professor of radiology at the University of California, San Francisco. Biopsy yield in screening patients with a prior exam was 42%, while those without priors had a biopsy yield rate of 25%.
When broadened to incorporate findings from UCSF's 12-year study of screening mammography, the recall rate in screening patients with prior exams shrank from 7.2% to just 3% during that time. Recalling patients to repeat an exam is not only time-consuming, it may not necessarily lead to better cancer detection. Depending on the radiologist, increased callback rates may not have a higher sensitivity for cancer, according to Dr. R. James Brenner, chief of breast imaging at Eisenberg Keefer Breast Center in Santa Monica, CA. But those physicians who miss a higher than average number of cancers might benefit from increasing their callback rates?within reason, he said.
One of the best ways to increase breast cancer detection that doesn't involve a technology investment is to become an expert, according to Brem.
"There's no question that the more mammograms you read, the better you are at interpreting them," she said. "And second readings on screens is obviously going to be very helpful as well."
Radiologists at the University of Vermont-Fletcher Allen Health Care in Burlington found their cancer detection increased 7.9% when they deployed a double-reading strategy. The retrospective study, presented at the American Roentgen Ray Society meeting, evaluated 25,638 mammograms. The review led to a 1.5% increase in patient recall. Because the second read is not billed, expenses generated to interpret as well as perform additional imaging and related tests totaled $111,000, an average of $10,000 per additional cancer detected.
The socioeconomic costs of double reading may put it out of reach for many radiology practices, however, particularly in a time when the field is short-staffed by an estimated 20%. At the same time, about 800 radiologists a year are retiring, compared with a historical rate of about 400, according to Vyborny at the University of Chicago. That's yet another reason why CAD could emerge as mammography's most valued complementary tool.
"Computers are different from humans and can give reliable, consistent assistance," Vyborny said during his presentation at the SCAR meeting. "They never forget or regress, they don't get tired, and they will only get better."
In addition to its clinical advantages, CAD could also provide radiologists with a liability edge. Radiologists are more frequently targeted for breast cancer litigation than any other specialists. The average judgment is $300,000, and multimillion-dollar judgments are not uncommon. Missed cancer is a common reason for such litigation, as is delayed diagnosis.
Concerns have been raised that if it fell into the wrong hands, CAD could be used as a litigation tool against radiologists. That fear, triggered by a false allegation on an Internet news group that R2 Technologies has sold its ImageChecker device to lawyers, raises the issue of unintended consequences of new technology.
The Los Altos, CA, company has affirmed that no such sales to law firms have been made, nor does R2 plan to target its marketing efforts outside radiology. The reason is simply that the clinical claims made about the product are based on its use by trained radiologists, said Michael Klein, president and CEO of R2.
There's little question that CAD could prove to be a useful risk management tool for radiologists, however, according to Dr. Richard Anderson, chair of the board of governors of the The Doctors Company, a national medical malpractice insurance carrier. In an editorial penned for the June edition of The Doctor's Advocate, the organization's member newsletter, Anderson noted that CAD offers substantial benefit.
"A 20% increase in the cancer detection rate involving the most common cause of cancer in women is impressive enough," he said. "The promise of reducing the incidence of the most common malpractice suit in the U.S. is added incentive for widespread evaluation."
ROAD AHEAD
The transition in mammography from screen-film to digital could create new opportunities for CAD. While development of the two technologies has been independent, a natural interaction exists between them.
The primary advantages of digital mammography include increased speed of image acquisition, image enhancement through advanced algorithms, and the potential to better store and transmit images for archiving and consultation. To date, full-field digital's contribution to mammography has been in these largely technical realms, posing a marketing challenge for vendors in various stages of clinical testing and FDA approval (see sidebar).
A study that began enrolling patients in September is expected to quantify the clinical value that digital mammography adds. The three-year trial, conducted under the auspices of the American College of Radiology Imaging Network (ACRIN), will employ digital devices from several competing vendors and compare the clinical performance of digital and conventional screen-film mammography. Dr. Laurie J. Fajardo, director of breast imaging at Johns Hopkins University, is principal investigator in the $25 million National Cancer Institute-funded study.
The ACRIN trial will involve 49,500 asymptomatic women at 19 participating centers in the U.S. and Canada. All women will undergo both digital and screen-film mammography and will be followed for up to 15 months.
In addition to measuring clinical efficacy, the trial will also evaluate two related issues: the relative cost-effectiveness of both technologies, and the impact on patient quality of life that results from an expected reduction of false-positive mammograms with digital mammography.
"There's enormous excitement about digital mammography but no data to show it has improved diagnostic capability," Brem said. "The ACRIN trial will, hopefully, provide us with some answers."
Meanwhile, research continues apace to increase breast cancer detection using technologies with features mammography doesn't provide. One of these, scintimammography, offers a physiological approach to diagnosis.
"Mammography and ultrasound examine anatomy. Scintimammography looks at breast cancer in a fundamentally different way. It evaluates not only how the tissue looks, but also how the tissue is working," Brem said.
The potential to better evaluate women with dense breasts using scintimammography is the topic of a study with Brem at the helm. An estimated 25% of women have dense breasts, and interpreting cancer in these women using mammography and other methods presents a great challenge, she said.
"In women with dense breasts, mammography detects only 65% of cancers," she said. "But studies have shown that scintimammography is not impacted by breast density. So we may be able to overcome a very difficult situation with a physiologic approach to breast imaging."
It is these kinds of studies that the IOM had in mind when it urged researchers to continue their focus on breast cancer detection beyond conventional mammography. No single imaging technology is capable of accurately detecting all breast abnormalities. Ultimately, the best detection may come from using several different tools.
