Study: Mammograms Best Read By Human Eye

Apr 06, 11:30 AM

Current Headlines: By DENISE GELLENE

By Denise Gellene

Los Angeles Times

An increasingly popular technology that uses computers to scan mammograms actually produces worse results than human reviewers using their eyes and experience, according to a new study.

Radiologists using computer-assisted detection software were more likely to interpret a benign growth as potentially cancerous, researchers said Wednesday in the New England Journal of Medicine.

The false-positive readings led to more scans and needless biopsies, adding $550 million to the annual cost of breast cancer screening in the United States, researchers said.

In addition, the computerized detection system, known as CAD, did not help radiologists find more real cancers, according to the report.

Dr. Ferris Hall, a radiologist at Beth Israel Deaconess Medical Center in Boston, who wrote an editorial accompanying the report, said some of the mistakes might have been due to inexperience with CAD. It takes radiologists several years to learn the technology, he said.

Nonetheless, Hall said, the study was a setback for the technology, which is used in 30 percent of the more than 30 million mammograms performed in the United States annually. Many imaging centers aggressively advertise their CAD services, which are potentially more profitable than standard mammography, he said. Medicare pays an additional $20 for each mammogram screened by CAD.

"This will have a major impact on radiology," Hall said. "They were calling people back for more scans and did more biopsies - that is hurting people. And what did they get for it? No significant increase in cancer detection."

The study shows other techniques are needed to detect breast cancer in its earliest stages, said Dr. John E. Niederhuber, director of the National Cancer Institute, which finance d the study.

Breast cancer is the second- most common cancer in women, next to skin cancer. The American Cancer Society estimates that 178,480 women will be diagnosed with the disease this year and 40,460 will die of it.

Mammograms , X-ray images of the breast, have long been the primary tool for detecting breast cancer in its earliest stages - before tumors are large enough to detect in a clinical breast exam. Last week, the American Cancer Society recommended annual magnetic resonance imaging for women with a high breast cancer risk.

Launched in 1998, CAD was designed to improve the accuracy of mammogram readings. A device converts the film mammogram into a digital file that can be analyzed by computer and displayed on a monitor. The software marks suspicious areas on the screen image for the radiologist to review in addition to the areas detected by the radiologist's eyes. Each mammogram contains an average of four marks.

Previous studies assessing the benefits of the technology have produced mixed results.

The latest study looked at mammography results for 220,000 women at 43 imaging centers in Colorado, New Hampshire and Washington. Seven of the centers began using CAD during the study.

The researchers found that after using the software, the rate of women recalled for more imaging tests went up 32 percent, along with a 20 percent increase in the number of breast biopsies. The women were later found not to have breast cancer.

Maryellen Lissak Giger, professor of radiology at the University of Chicago and an inventor of CAD technology, said the study had several flaws. The study was too small to detect a slight improvement in cancer detection, she said.

Moreover, the research, which took place between 1998 and 2002, evaluated an early version of the CAD system, said Giger, who owns stock in Hologic Inc. of Bedford, Mass., which licenses its technology from the University of Chicago.

Lead author Dr. Joshua J. Fenton, assistant professor of family and community medicine at the University of California, Davis, said the software made radiologists' jobs more difficult. Because most women who receive mammograms do not have breast cancer, only one of 2,000 marks made by the software represents an actual tumor, he said.

"The radiologist has a difficult time separating the wheat from the chaff," Fenton said.

(c) 2007 Virginian - Pilot. Provided by ProQuest Information and Learning. All rights Reserved.