Computer program to diagnose breast cancer on mammograms
Mammograms are very effective techniques for detection of breast cancer. It is capable of detecting breast cancer even at very early stages known as carcinoma in situ. Reading a mammogram is a tedious process and it often takes years for a physician to acquire the expertise needed to read mammograms. Also the mammogram reading skill of a physician is directly related to the experience of the physician in reading such mammograms. It looks like computers may be able to take over this tedious job of analyzing the small specs of lights on a mammogram film.
A computer algorithm known as likelihood ratio (LRb) classifier is shown to be very effective in detecting breast cancer in mammograms. Likely ratio classifier is capable detecting breast cancer in mammograms with 100 percent accuracy. But when it comes to non-cancerous lesions they were not very good. The program was capable of detecting benign lesions only in 26 percent of mammograms.
These findings are from a study presented by Anna Bilska-Wolack PhD from Duke University, presented at the Era of Hope Department of Defense Breast Cancer Research Program meeting (2005). Likelihood ratio (LRb) classifier uses an optical technique that mathematically calculates the data from signal processing to decide if the presented signal represents a breast malignancy. The computer program was developed from a large database of mammograms which was biopsy proven to have breast cancer. The likelihood ratio (LRb) classifier has been shown to be accurate in sets of mammogram that were not involved in developing this computer program.
This new technology may be used in two ways. This computer program may be used to analyze an abnormality that has been identified by a physician. Since the accuracy of this process is 100 percent in the previous studies, the assessment of the computer may be more accurate than provided by the physician. An alternate use may be to supplement the mammogram reading of the physicians to localized abnormalities on the mammogram. These computer programs are not yet ready to take over the job from the physicians, but this is a real possibility in the near future.