Computer-aided detection (CAD): the case against
- R Given-Wilson1
© BioMed Central 2004
Published: 14 July 2004
CAD places prompts to alert readers to potential lesions. Readers can decide whether to initiate further assessment. Currently, commercially available CAD systems are highly sensitive and will prompt 94% of detected cancers (98.4% microcalcifications, 89% masses). Prompt rates on missed cancers are lower but the majority are prompted (52–77%). A number of published evaluations of CAD benefit extrapolate from these rates. They assume that readers will respond appropriately to every correct prompt and initiate recall. Studies in vitro and in real life show variable reader response, with increases in cancer detection ranging from 0% to 22%.
There is lower specificity with one to two false prompts per case. In UK screening, readers will dismiss over 1000 false prompts for every true prompt, which should change their outcome decision. This makes discrimination difficult. When investigating missed cancers we have found readers are more likely to dismiss than act on true prompts. In addition, CAD increases reading time and can increase recall rate, thus increasing cost. This is offset in the US by higher reimbursement for CAD reading but would be cost increasing in the UK.
CAD technology is improving and may shortly be beneficial but at present its role in the UK is unproven.