A framework for medical decision support systems - A case study of EMG signal interpretation.

This thesis contains a framework of medical decision support systems (DSS) and a case study of signal interpretation in electromyography (EMG). Methodologies required for architecture of such knowledge-based systems (KBS) are explored at both epistemological level and computational level in the thesis. Firstly, a research scheme is proposed under which system development is explicitly defined as a process of two stages, conceptual design and computational design. The conceptual design is mainly concerned with the medical domain. Its task is to abstract the problem-solving logic of that domain into symbolic forms of models. The computational design deals with more concrete issues related to system implementation and information technology (IT). Its task is to create applicable computing models in light of the conceptual system models. Object-oriented analysis (OOA) is introduced into the process of system analysis and design. At conceptual level, three fundamental inference patterns, forward reasoning (FR), backward reasoning (BR) and induction reasoning (IR), are abstracted as basic building blocks of domain problem-solving. Medical domain problem-solving can therefore be modelled as a hierarchical process made from the three basic patterns. A scheme extracted from a number of blackboard systems is determined as the framework's general problem-solving scenario. The blackboard scheme is then defined as a complex of interacting objects. At computational design, various computing algorithms are reviewed which range from object-oriented design (OOD), knowledge representation, uncertainty-handling to user-system interface. Decision strategies are also investigated in the context of EMG interpretation. The logic of medical decisions is revealed as a process of heuristic search and probabilistic thinking by Bayes' theorem. Decision-making is therefore modelled as a hierarchical process that combines inference reasoning and uncertainty-handling. An inference engine is designed using fuzzy logic and a decision network. It employs the decision principles of Bayes' theorem, but applies more realistic strategies of uncertainty-handling and inference. A preliminary EMG interpretation system is built under the framework. It is evaluated via a simulation study. The general methodology presented by the framework is shown to be a promising approach to system transparency, efficiency and performance.