| Breast cancer is second to lung cancer in the cancer fatality rate among U.S. women today. Early detection is the best defense and mammography is the best tool to detect early stage breast cancer. The National Library of Medicine (NLM) tele-educational system, created at UNC, is a computer-based educational system for radiology residents who interpret digital mammograms. This system includes a data collection tool, an annotation tool, a case demonstration tool, an online query tool, a mammography test design tool, a testing tool, an XML tool, an online survey tool and some utilities for automatically loading digital mammograms. A local Oracle database was designed and implemented at UNC to support those tools. Our system integrates the imaging findings and pathology information together. Radiology residents can get the images, review them and compare pathology reports from breast biopsies with the corresponding mammography examinations that prompted those biopsies. The educational mammography cases in our database can be published as XML files for sharing. Mammography tests can be automatically constructed by using our test design tool, by using a set of default rules or by creating new unique rules. Our testing tool uses the Receiver Operator Characteristic (ROC) curve methodology to evaluate radiology residents' performance in interpreting mammograms. The testing tool not only evaluates skills of test takers in the detection of abnormalities visible on mammograms, but also accesses their skills in the diagnosis of the detected abnormalities utilizing a continuous rating scale. Our testing tool successfully and automatically provides test takers with detailed performance test results, such as sensitivity, specificity, positive predictive value, negative predictive value, diagnostic accuracy, ROC curve, and Az value (Area Under Curve).; In conclusion, mammograms can be integrated with diagnostic clinical information and a skill-assessment test can be automatically constructed. The ROC analysis method is appropriate for a real-time computer application, directly using the raw data from a database, to evaluate the performance of trainees in interpreting mammograms. The Az value from ROC analysis is an ideal index for accessing the performance of radiologists in interpreting mammograms. |
