Computer Detects Ovarian Cancer Early
Program spots tumors by finding patterns in blood proteins
THURSDAY, Feb. 7, 2002 (HealthDayNews) -- A computer program that looks for distinctive patterns of proteins in blood appears to be a promising new screening tool for early detection of ovarian cancer, researchers report.
In its first test, the computer analysis of "proteomic patterns" was 100 percent accurate in detecting ovarian cancer and 95 percent accurate in determining that women with other diseases did not have cancer of the ovaries, says a report in the Feb. 16 issue of The Lancet.
Such a test -- particularly a noninvasive one -- would be welcome, experts say. Because it doesn't display early symptoms and because the ovaries are deep within the body, ovarian cancer is often caught only in its late stages, when the five-year survival rate is only about 35 percent.
Researchers at the National Cancer Institute, the Food and Drug Administration and Correlogic Systems, Inc., a private company, developed the technique. It uses the kind of pattern recognition software that has a wide variety of applications.
"Using this technology has allowed us to identify a pattern that can detect women who have ovarian cancer vs. healthy people who do not have it," says Dr. David A. Fishman, director of the National Ovarian Early Detection Program at Northwestern University, a study participant. "It is a pattern of protein fragments that looks like a fingerprint for ovarian cancer."
It's not necessary to know where the specific protein pattern originates or what biological significance it has, Fishman says. "All we can say is that it is a pattern that leads to cancer rather than noncancer," he says.
The results of the first test "are certainly good enough to justify further evaluation," Fishman says. In that test, a computer-based artificial intelligence program first identified unusual protein patterns in blood samples from women with ovarian cancer.
That program then was applied to samples from 50 women with ovarian cancer and 66 women who were healthy or had illnesses unrelated to ovarian cancer. The program identified all the cases of ovarian cancer and wrongly identified only 3 of the 66 noncancer samples.
"The next step obviously is a larger national validation study in which we can evaluate thousands of women and determine how accurate this application is," Fishman says. That study, already begun, will first evaluate 300 women and then be enlarged to include 2,000 women.
Parallel to that effort will be work to make the test widely usable. "This is a very sophisticated technology that needs to be modified so that everybody should be able to benefit from it," Fishman says.
The journal report indicates that the proteomic test is better than existing tools, says Robert Smith, director of cancer screening for the American Cancer Society.The new test does offer "markers that have better performance than those we have now," Smith adds. Testing is advisable for women at high risk of ovarian cancer, either because there is a family history of the malignancy or because they have had breast cancer. One BRCA gene is is known to be associated with both breast cancer and ovarian cancer, Smith notes.
A longer-range goal, says Fishman, is to analyze the proteins identified by the computer program, "to determine what they are and what they mean biologically -- what is the significance of these fingerprints, and do they have any biologically meaning."
Understanding the proteins could lead to better treatments of the cancer, he adds.