Methods, Limitations, and Future Directions in Computer-aided Diagnosis of Breast Cancer

摘要

Identifying breast cancer at screening mammography may be one of the more difficult detection tasks in radiology because of the subtle appearance of some cancers, the low prevalence of disease in a large screening population, and the speed with which studies must be interpreted. To meet this challenge, computer algorithms have been developed to assist radiologists in the task of detection. In 1967, Winsberg et al (1) described a computer-aided detection (CAD) system to assist radiologists in the task of identifying breast cancer. Since that time, the field of CAD for breast imaging has matured, with three systems now commercially available and approved by the Food and Drug Administration (FDA) for routine use. Each of these systems detects a high percentage of breast cancers (2), and CAD systems have proved to help radiologists identify cancers that they would otherwise have missed (3,4). The purpose of this review is to consider some of the techniques used in computer detection and classification of breast cancer, to illustrate remaining limitations of CAD, and to speculate on the future direction of CAD in breast imaging.