Evaluation of computer-aided detection devices: consensus is developing.

Computer-aided detection (CAD) devices are of increasing interest to practicing radiologists. Many radiologists are familiar with CAD for detecting microcalcifications and masses on mammograms. The earliest US Food and Drug Administration (FDA) approvals for mammographic CAD devices were in 1998 (1). Since then, more than a dozen CAD systems have been approved for detecting abnormalities such as lung nodules on chest radiography and computed tomography, colon polyps on computed tomographic colonography, and pulmonary emboli on chest computed tomography (1). Cross-sectional imaging studies are being performed with increasing frequency, and the number of images per examination is growing exponentially (2,3). Hence, there is a steadily building need for computerized devices to help radiologists interpret this avalanche of images in an accurate and timely fashion. Despite the great potential for CAD devices to make a positive impact in radiology, new CAD technologies are arriving slowly to the reading room. For example, researchers have developed CAD systems that detect or classify lesions in the solid abdominal organs; detect bone lesions and fractures; make fully automated assessments of bone mineral density, visceral fat, and bone age; and detect neurologic and vascular abnormalities such as cerebral aneurysms (4–15). Yet none of these systems is available as a commercial product in the United States. Why does it take so long for these new technologies to make their way from the bench to the viewing console? Delays and impediments occur at each stage of the CAD