Cancers Missed on Mammography

Observer error in reading screening mammograms has been identified as a significant factor in delayed diagnosis of breast cancer. The magnitude of the problem is estimated to be about 30 of potentially detectable cancers are overlooked for one or more years before detection. Computer-aided diagnosis CAD programs have been developed to aid radiologists in the detection task, and pre-clinical studies have shown that CAD applied to digitized mammography films can flag about 50 of radiologists observational oversights. Preliminary study has also shown a wide variability in radiologist observer performance. The purpose of this investigation is to test how many additional cancers are detected by radiologists using CAD, in an observer study using an enriched mixture of cancers. A pilot study has been completed and the results were used to plan the full observer study. Using more sophisticated statistical power analysis, we estimate that we need 400 cases containing 70 cancers with 12 radiologists. In addition, we will use a more accurate detection scheme that has a higher sensitivity and lower false-positive rate than was used in our pilot study. Final case selection is being done and observers are being recruited.