Uncertainty Management in Medical Applications

In biology and medicine, as well as in many other domains, imperfect knowledge cannot be avoided. It is difficult to construct automatic systems to provide classification or pattern recognition tools or help specialists make a decision. There exist two kinds of difficulties: (1) those related to the type of imperfection we have to consider (partial information, uncertainties, inaccuracies) and (2) those due to the type of problem we have to solve (e.g., images to process, expert rules, databases). Which mathematical model are we supposed to choose to manage this imperfect knowledge? What is the best knowledge representation for a given problem? The answers to such questions are not obvious, and our purpose is to present several frameworks available to represent and manage imperfect knowledge, particularly in biological and medical domains. We indicate principles, interest and limits of these frameworks. We give more details about numerical approaches that have given rise to more practical applications than about symbolic approaches, which will be mentioned only briefly.