Formalizing diagnostic reasoning is a necessary aspect of creating useful diagnostic tools. Even though much progress has been made, there are several deficiencies in the existing formalization. These include the definition of diagnostic reasoning as a single- stage process which is independent of the use of the diagnosis, of tests which can be done to clarify the hypothesized diagnoses, or of the utility associated with the treatment of the abnormalities. This research proposes a sequential decision framework for diagnostic reasoning which overcomes many of these deficiencies. A key new intuition for this approach is modeling diagnostic reasoning as a process which consists not solely as a single step of computing a diagnosis (as there is considerable controversy over what a diagnosis actually is, even given the same input data); instead, it is a multi-stage decision process in which decisions about testing and treatment, as well as scarce resource and utility considerations, are crucial parts of the model. This framework extends most existing formalization of diagnostic reasoning to incorporate important aspects of the diagnostic process. Two applications of the framework to diagnostic systems are investigated in this research. The first application (which has been implemented) is a Markov decision network model for the diagnosis of acute abdominal pain. The second application is a sequential process of consistency-based diagnosis interleaved with tests, for diagnosing Boolean circuits. //
