Computer aided detection

Computer analysis of digitised mammograms is now remarkably sensitive for the detection of calcifications and masses. This sensitivity, however, is only achieved at low levels of specificity. System developers have tried to get around this difficulty by developing prompting systems, in which the computer algorithms are used to alert film readers to possible areas of abnormality. The idea is that the computer may be able to improve a radiologist's, or radiographer's, sensitivity while relying on his or her professional judgement to maintain acceptable levels of specificity.

Research in the United States has shown that the algorithms are sensitive and that prompting systems do not adversely affect specificity. The evidence that they can improve radiologists' sensitivity is less certain, although there are now studies showing that improvements have been achieved. The key unanswered question is what is the potential impact of these tools on the decision-making of radiologists and radiographers working in the UK screening programme? A large scale evaluation, involving 50 film-readers, is underway to assess this impact. Interim results will be presented at Symposium Mammographicum. Issues include the significance of abnormalities other than masses and calcifications, and the relative importance of errors of detection and errors of decision-making.