New Technology Weighs In On Mammography Debate

Despite proven success in reducing breast cancer mortality, mammography remains one of the most controversial techniques of cancer screening. The recommendation that women aged 50 and older have regular mammograms is universally accepted. But the question over whether women aged 40-49 years should receive regular screenings has sharply divided the medical community. MAMMOGRAPHIC PLOT: Jack Sklansky has developed a method of graphically representing a mammogram, which could improve accuracy. E

Nov 24, 1997
Stephen Hoffert

Despite proven success in reducing breast cancer mortality, mammography remains one of the most controversial techniques of cancer screening. The recommendation that women aged 50 and older have regular mammograms is universally accepted. But the question over whether women aged 40-49 years should receive regular screenings has sharply divided the medical community.


MAMMOGRAPHIC PLOT: Jack Sklansky has developed a method of graphically representing a mammogram, which could improve accuracy.
Earlier this year, the National Cancer Institute (NCI) went against the recommendation of the National Institutes of Health when it endorsed biennial mammograms for women under 50 years old. Critics of the decision have charged that NCI and its National Cancer Advisory Board (NCAB) disregarded scientific evidence and gave in to political pressure with the recommendation (S. Glazer, CQ Researcher, 7[24]:555-64, 1997).

The newest studies on the benefits and risks of mammographic screening for the controversial age group could renew the debate. Meanwhile, researchers in imaging technology are striving to improve the accuracy of mammography and therefore put an end to much of the debate.

Controversy erupted in January when an NIH consensus panel ruled that there was insufficient evidence to recommend mammograms for women aged 40-49 years. NCI supported the NIH recommendation and then wavered, endorsing the recommendation of the presidentially appointed NCAB. The board urged by a vote of 17-1 that women aged 40-49 receive annual to biannual mammograms.

Much of the controversy stemmed from the fact that NIH based its recommendation, in part, on the Canadian National Breast Cancer Screening Survey (CNBSS), a study of more than 90,000 women aged 40-59 that showed an insignificant effect of mammograms in reducing mortality in women aged 40-49 years. The validity of the Canadian study was challenged; critics argued that randomization was tampered with as more suspected and known symptomatic women were assigned to treatment groups (C.J. Baines, Annals of Internal Medicine, 120[4]:326-34, 1994). Although an independent panel found no evidence of tampering (J.C. Bailar III, B. MacMahon, Canadian Medical Association Journal, 156:193-9, 1997), critics are still vocal and persistent.

Daniel B. Kopans, a professor of radiology at Harvard University and director of breast imaging at Massachusetts General Hospital, maintains that most breast cancer screening trials were not designed to evaluate just women aged 40-49, and not enough women from this age group are included to conclusively prove a significant benefit from mammograms (D.B. Kopans, Surgical Oncology Clinics of North America, 6[2]:233-63, 1997). Moreover, he claims that there are no parameters, such as breast density, that suddenly change at age 50 to improve detection rates (D.B. Kopans, CMAJ, 157:247,1997).

Radiologists criticize CNBSS for using substandard mammographic screening techniques (S.A. Feig, Cancer, 75[10]:2412-9, 1995). Stephen A. Feig, a professor of radiology and director of breast imaging at Thomas Jefferson University Hospital and Medical College in Philadelphia, argues that CNBSS results are unreliable because of antiquated methods of mammographic examination were used throughout the seven-year study. After widespread criticism, CNBSS eventually brought in outside radiologists in an advisory role to recommend ways to improve the screenings.


NEW AND IMPROVED: Radiologist Stephen Feig says digital mammography could significantly improve cancer detection rates.
"The study had never included the advice of radiologists, but just epidemiologists," Feig explains. "Without other experts, they could look at mammography in a simplistic way by looking only at numbers. The survey was just a mess because of poor mammographic technique. They called in outside experts in mammography to patch things up. I was the fourth to resign from the position. I felt that they were using my good name and didn't want to implement the improvements that I advised." Feig notes that after improvements were finally made, detection rates for women aged 40-49 nearly doubled.

R. Edward Hendrick, a professor and chief of the division of radiological sciences at the University of Colorado Health Sciences Center in Denver, says his most recent meta-analysis shows a significant reduction in mortality in women aged 40- 49 who receive mammograms (R.E. Hendrick et al., Journal of the National Cancer Institute Monographs, 22:87-92, 1997). The study combined follow-up data from eight randomized mammography screening trials, including a multisite trial in Sweden commonly called the Five County Swedish randomized clinical trials (RCTs) and CNBSS. The meta-analysis showed an 18 percent mortality reduction for women aged 40-49 invited to mammography screenings. Follow-up data from just those women in the Swedish trial yielded a 29 percent mortality reduction.

Of the eight RCTs, two of the Five County Swedish RCTs alone showed a significant benefit in mortality for women aged 40- 49. Three RCTs suggested a insignificant benefit and another three could not demonstrate a benefit, according to Hendrick. But he notes that when combined and analyzed, the eight studies show a significant benefit at a high confidence level.

Critics of Hendrick's meta-analysis remain guarded. Karla M. Kerlikowske, an assistant professor in the department of medicine, epidemiology, and biostatistics at the University of California, San Francisco, argues that studies like Hendrick's prove only a relative, not an absolute, benefit from mammography screening in women aged 40-49. "It's actually ironic," Kerlikowske says. "In Sweden, the county trials have ended recommendations for mammograms in women 40- 49 years old. Only when the data is exported to the United States does anyone consider recommending mammograms to women aged 40-49." But Hendrick notes that counties in Sweden set their own screening policies, and some still recommend mammograms for women aged 40-49.

Mammograms also have been shown to increase the risk of receiving treatment for a tumor that may not become malignant. Since the 1980s, when mammography screening became widespread, the diagnosis of ductal carcinoma in situ (DCIS) has significantly increased (V.M. Ernster et al., JAMA- Journal of the American Medical Association, 275[12]:913-8, 1996). Compared with the previous decade, the diagnosis rate from 1983 through 1992 climbed from 0.3 percent to 12 percent for women aged 30-39 years. Among women 40-49 and 50-59 years old, the diagnosis rate jumped, respectively, from 0.4 percent to 17.4 percent and from 5.2 percent to 18.1 percent. The total number of DCIS cases in the U.S. in 1992 was 23,368, or 200 percent higher than the 1983 rate.

The boom in DCIS cases found as a result of mammograms has prompted concern from some researchers, as 30 percent to 75 percent of all DCIS does not become malignant cancer (K.M. Kerlikowske, JNCI, 89[1]:76-82, 1997). While risk factors for DCIS nearly parallel those for invasive cancer, Kerlikowske concludes in the study that "more research is needed to better understand the malignant potential of DCIS lesion and factors that predict which lesions will become invasive cancer if left untreated."

Given the uncertainty about the connection between DCIS and invasive cancer, researchers fear that many women are receiving aggressive cancer treatment when it is not necessary. But radical treatment rates for DCIS have fallen since 1983, when 71 percent of all cases were treated with mastectomy. This rate fell to 43.8 percent by 1992 as the favored treatment became lumpectomy.

INVISIBLE TO THE NAKED EYE: A standard film image mammogram, top left, is limited when compared with Klaus-Ruediger Peter's digital enhancement technique, which produces three new views: The major contrast view, bottom left, shows the "organ component" of the breast; the minor-contrast view, bottom right, shows the fibrous component of the breast; and the medium-contrast view, top right, shows the glandular and ductal structures in the breast.

"Mammograms sometimes do save lives," Kerlikowske says. "But factors like the likelihood of false positive screenings and aggressive treatment for DCIS suggest that the benefit might not be so great or worthwhile. There is a risk that you're treating people for something which may not ever develop into cancer."

Various studies have attempted to explain why mammogram quality is lower for younger women. Cornelia J. Baines, a professor in the department of public health sciences at the University of Toronto, recently suggested that particular phases of the menstrual cycle might have an adverse effect on mammography quality (C.J. Baines et al., Cancer, 80[4]:720-4, 1997). Baines reports a significantly higher risk factor for false negative screening results during certain phases of the menstrual cycle and in those who had ever used oral contraceptives or replacement estrogen. Citing earlier research that found evidence of changes in breast composition during menstruation (P.A. Fowler et al., British Journal of Obstetrics and Gynecology, 97:595-602, 1990), Baines writes, "Although it cannot be concluded that these changes cause an increased likelihood of false negative mammograms, they are changes that occur in synchrony with an apparent increased likelihood of [false negatives]."

Feig does not dispute that breast composition changes during menstruation and notes other studies that suggest how breast composition interferes with mammographic quality. But he argues that these physical changes are not great enough to produce the upsurge in false negative readings that Baines found.

"As a radiologist, I've performed screenings in various stages of the menstrual cycle," Feig says. "I just don't see the exaggerated effect of false negative readings that Baines sees. What is known for certain is that breast compression gives you a sharper image at lower doses of radiation. Of course, menstruation causes breast tenderness, and this can interfere with a woman's tolerance for compression.

"This is one way to explain her results. But I don't think the two-times greater risk is accurate. The study is based on mammographic studies from the 1980s, and the techniques were not good at all. While Baines's study is useful, it is based on poor-quality screenings, and this exaggerates the effect of false negative results."

Mammographic quality has increased significantly over the years. But even with these improvements, radiologists acknowledge that mammography has "inherent limitations" (S.A. Feig, M.J. Yaffe, Seminars in Ultrasound, CT, and MRI, 17[5]:423-43, 1996). As many as 10 percent of all cases of breast cancer go undetected by mammography, according to Feig.


BLIND STUDIES: Klaus-Ruediger Peters says the human eye's limitations result in nearly 80 percent of a mammogram's information not being read.
This should not be surprising, given subtlety of difference between normal and possibly cancerous tissue. Klaus-Ruediger Peters, an assistant professor of diagnostic imaging and therapeutics at the University of Connecticut Health Center, explains that cancer reveals itself in what often amounts to just a single contrasting pixel in the breast image. Denser breast tissue, which interferes with image quality and is common in premenopausal women, makes detection of suspicious tumors even more difficult, according to Feig.

But researchers have been hopeful that new digital mammography will overcome many of the limitations of traditional mammography. Early trials show that digital mammography can increase detection rates by improving spatial and contrast resolution. In using digital images, the creation of mammogram databases and computer-assisted diagnosis (CAD) programs has been facilitated.

Jack Sklansky, a professor of electrical and computer engineering at the University of California, Irvine, has been developing expert computer systems that can classify and analyze many types of images, including mammograms. Sklansky recently completed a promising trial at the King/Drew Medical Center in Los Angeles that shows that the CAD system could reduce both false negative and false positive mammogram readings. He will report the results in a paper submitted in Chicago to the 1997 annual meeting of the Radiological Society of North America, which begins on November 30.

Professional Resources

American Cancer Society
John R. Seffrin, Executive Vice President
1599 Clifton Rd., N.E.
Atlanta, Ga. 30329
(404) 320-3333 or (800) ACS-2345
Fax: (404) 329-7530
http://www.cancer.org
Journal: Cancer

Radiological Society of North America (RSNA)
Michael A. Sullivan, President
2021 Spring Rd.
Oak Brook, Ill. 60523-1860
(630) 571-2670 -- Fax:(630) 571-7837
http://www.rsna.org
Journal: Radiology

Journal
Seminars in Ultrasound, CT, and MRI and Radiologic Clinics
W.B. Saunders Company
Curtis Center
Independence Square West
Philadelphia, Pa. 19106-3399
(215) 238-7300 -- Fax: (215) 238-8483
http://www.hbuk.uk.com

Internet Resources

Brandeis University maintains a digital mammography home page with various links http://www.rose.brandeis.edu/users/mammo/digital.html.

The National Surgical Adjuvant Breast and Bowel Project, headquartered in Pittsburgh, has information on breast cancer trials http://www.nsabp.pitt.edu. The University of Pennsylvania Oncolink provides resources to health-care professionals and basic researchers http://oncolink.upenn.edu/disease/breast

Sklansky's CAD system, or "visual neural network," works by mapping a mammogram onto a two-dimensional grid. Distances between dots represent measures of the dissimilarity of the medical image. So, two dots that lie close together have similar diagnostic value. A decision curve also is placed on the grid. The distance of a dot from the curve measures the uncertainty of the network's diagnosis of benign or malignant.

The CAD system was tested using a database of 86 diagnosed mammograms. This set was divided into a test and a design set. The design set was used as the standard for the CAD system. When given the set of test images, the CAD system caught all malignant cases but recommended biopsies for 21 percent of all benign cases. This experimental reduction would translate into a significant reduction in false positive readings in real medical practice, according to Sklansky.

Peters recently developed a photographic image-processing technique that produces highly detailed and easy-to-read mammograms. His work began when he contemplated the amount of information in a mammogram that is never read by the radiologist. The human eye can discriminate among about 20 percent of all possible shades. Most of these are those with the highest contrast.

"But the most telling information in a mammogram is those less-discriminated regions of gray," Peters explains. "That means about 80 percent of the information contained in a mammogram is never read because the human eye cannot notice it."

Peters discovered an algorithm that duplicates the process of hysteresis, first postulated as a theory of visual perception by the late David Marr, a British psychologist and pioneer in cognitive science and visual processing (Vision: A Computational Investigation Into the Human Representation and Processing of Visual Information, San Francisco, W. H. Freeman, 1983). Hysteresis refers to a mental function in the brain in which visual signals of differing contrast are transferred into a symbolic representation that the brain perceives as an image. Using a variety of computer equipment, Peters turns a continuous film image into a digitized image by converting discrete contrast regions of the film into digital contrast patterns. In this process, each pixel is compared to all of its neighboring pixels in the whole image. The program then assigns the pixel a contrast shade large enough for the human eye to discern.

"With this process, I can give you a digitized image which is perceptually indistinguishable from a photographic image," Peters boasts. "More exciting, the process also yields separate high-, medium-, and low-contrast views which would help in finding smaller, more hidden tumors."

Peters and Gale R. Ramsby, chairman of the department of diagnostic imaging and therapeutics, have begun retrospective studies involving women recently diagnosed with breast abnormalities. All of the women in the study have long mammography histories. Peters is hopeful that digitized images will reveal abnormalities much earlier than traditional film images. If successful, prospective studies will be done to see if digitized images read with the help of a CAD system can detect cancer earlier.