THE FORMAL DESIGN AND EVALUATION OF A VARIETY OF MEDICAL DIAGNOSTIC PROGRAMS

Abstract

One of the most successful and most widely studied applications of computer-aided diagnosis is that of acute abdominal pain. However, widespread introduction of computers into clinical practice seems to be hindered by their limited diagnostic accuracy. This monograph documents some experiments we have carried out to investigate the effect on diagnostic accuracy of using various statistical and knowledge-based methods to take dependencies between clinical observations into account. We present detailed specifications of a variety of statistical and knowledge-based programs within a common formal framework using the Z specification language. We describe how we collected a retrospective database of 1270 cases of abdominal pain of suspected gynaecological origin. This we used to evaluate all methods.