"In this study, we analyzed 49 full-field digital mammograms, 19 of which showed cancer," said Rich Rana, a medical student at the University of Chicago. Four mammography specialists read the images and electronically put a box around the suspicious calcifications. The computer then automatically detected the calcifications within the box, analyzed them and calculated the probability of cancer, Rana said.
The system proved to consistently achieve performance comparable to the radiologists in classifying malignant and benign calcifications, regardless of who was using it, Rana added. One technique for rating the computer's effectiveness is to give it one malignant case and one benign case and then test its ability to determine which is which, Rana said. Using this technique, the radiologist had a 72% chance of making the correct diagnosis, and the computer had a 79% chance.
This study was one of the first to test the effectiveness of computer-aided diagnoses on full-field digital mammograms versus plain film mammograms. In addition, this system employs artificial intelligence in that the computer "learns" how to automatically locate the calcifications and predict whether they are benign or malignant, Rana said. In the future the radiologist's assessment could be compared with the computer's assessment as a "double-check" for the diagnosis of breast cancer, Rana said.
This research was lead by Dr. Yulei Jiang, assistant professor of radiology at the University of Chicago. The research was funded by the National Cancer Institute, the U.S. Army Medical Research and Material Command, and the National Institutes of Health. Rana will present the stu
'"/>
Contact: Keri Sperry
703-858-4306
American Roentgen Ray Society
4-May-2004