We have investigated the feasibility of using a commercial prompting system in our screening centre. Its sensitivity was 92% for two-view screening mammograms and 66% for one-view, with three times as many false prompts as true ones. We attempted to integrate the system into clinical practice, assessing the impact on four readers who each reviewed 800 cases. Prompted reading did not significantly change their assessment of the films. Finally we took 90 incident round screening cases in which cancer was detected, and a similar control group of non-cancer cases. The previous screening mammograms were reviewed by an expert radiologist, then analysed by the system. The radiologist detected very early signs of cancer in 10% of the cancer cases, rising to 14.4% with the benefit of the subsequent diagnostic films. The system recorded a sensitivity of 27.8%, with an average of 1.32 false prompts per film in a one-view case and 1.56 per film in two-view cases.
展开▼
