Content uploaded by Peter J. Lucas
Author content
All content in this area was uploaded by Peter J. Lucas on Apr 16, 2013
Content may be subject to copyright.
A preview of the PDF is not available
Knowledge Acquisition for Decision-theoretic Expert Systems
Abstract and Figures
In this paper, the construction of decision-theoretic expert systems in collaboration with domain experts is discussed. In particular, the role of domain models in guiding the knowledge-acquisition process is reviewed, and various techniques that may help in the design of a decision-theoretic expert system are presented. Treatment planning in patients with a congenital heart disease is described as an example domain. The development of a decision-theoretic expert system for this domain is taken as a running example. 1 Introduction Decision-theoretic networks offer a mathematically sound collection of formalisms for building knowledge-based (expert) systems for domains in which uncertainty is of central concern. In this paper, such systems will be called decision-theoretic expert systems. The main applications of the formalisms are in classification, e.g. diagnosis (cf. [4]), and in decision-making under uncertainty, e.g. optimal treatment management of a patient (cf. [1]). The ...
Figures - uploaded by Peter J. Lucas
Author content
All figure content in this area was uploaded by Peter J. Lucas
Content may be subject to copyright.
... No structure learning methods, however, were used in the context of our previous research. Dr Lucas and Prof. van der Gaag have wide experience in building Bayesian-network based and other decision-support systems for clinical problems in collaboration with clinicians [29] [33] [35] [36] [19]. Dr Lucas has 18 years experience in dealing with the methodological and technical questions which arise in research in medical decision support; having been trained both as a computer scientist and an MD, he has the necessary background to lead this project. ...
... No structure learning methods, however, were used in the context of our previous research. Dr Lucas and Prof. van der Gaag have wide experience in building Bayesian-network based and other decision-support systems for clinical problems in collaboration with clinicians [29, 33, 35, 36, 19]. Dr Lucas has 18 years experience in dealing with the methodological and technical questions which arise in research in medical decision support; having been trained both as a computer scientist and an MD, he has the necessary background to lead this project. ...
2) Summary Although research on learning Bayesian networks from data started only about 10 years ago, significant theoretical progress has been made since that time. However, the community involved in this research has placed relatively little emphasis on gaining insight into the usefulness of this technology in solving real-life problems. In fact, the issue of tailoring Bayesian-network learning methods to the characteristics of problem domains has not even been addressed. The ProBayes project's primary aim is to investigate whether the Bayesian-network formalism oers a suitable framework for learning prognostic models from clinical datasets. Prognostic models play an important role in oncology, and the quality of the clinical management of cancer in patients may profit considerably from deploying medical decision-support systems incorporating such models. Decision support in clinical oncology is therefore taken as an experimental setting for ProBayes. Since it is expected to be essential to exploit background knowledge to guide data-mining and learning in the context of real-life problems, the major goal of the requested research is to obtain insight into the form of the required knowledge in constructing prognostic Bayesian networks from data. Within the empirical setting of the project, the problem of learning Bayesian networks will be studied along the spectrum from rare to common disorders, not simply by using publicly available datasets of unclear quality, but in its real-life context with considerable input from expert clinicians. There will be a major emphasis in the project on learning intuitive, understandable Bayesian networks, such as those that can be given a causal reading. Finally, the clinical usefulness of the developed Bayesian-network models will be studied in the context of clinical management of cancer patients. Overall, the ProBayes project aims to extend Bayesian-network technology so that it comes nearer to fulfilling its promise as a practical technology for clinical decision support.
... [1]. Bayesian networks have been used to create medical models before, as Lucas [29] causal model of aortic coarctation. The Bayesian network is meant to be used to represent dependencies between state-variables in the clinical decision support system. ...
... Later more emphasis was placed on treatment selection and making a prognosis, e.g. [8,[17][18][19]. Markov decision processes and dynamic influence diagrams are examples of temporal probabilistic models that have been used in [35] for selection of treatment strategies. ...
... The qualitative knowledge is acquired via composition of constrained natural language sentences in a linguistic model editor, whereas the quantitative knowledge is compiled from fuzzy, qualitative statements about stochastic relations. In his report on the knowledge acquisition process for a particular medical expert system, Lucas [6] emphasizes the use of specific domain models in guiding the process, and he presents various other techniques that may be helpful in designing a Bayesian network. Lacave and Díez [5] reports on the process of constructing PROSTANET, a Bayesian network for diagnosing prostate cancer. ...
We present a practical and gen- eral methodology that simplifies the task of acquiring and formulating qualitative knowledge for construct- ing probabilistic graphical models (PGMs). The methodology effi- ciently captures and communicates expert knowledge, and has signifi- cantly eased the model development process for three real-world prob- lems in the domain of robotics.
... For example , if death results from a temperature over 108°F, then 108°F and above is assigned a utility of -1000. If 104°F 1.By way of reference, a DDN modeled the decisions and utilities of forms of treatment for an aortic coarctation (Lucas 1996). For more information on the DBN to DDN transformation see Dean and Wellman (1991). ...
As autonomous mobile robots (AMRs) begin living in the home, performing service tasks and assisting with daily activities, their actions will have profound ethical implica-tions. Consequently, AMRs need to be outfitted with the ability to act morally with regard to human life and safety. Yet, in the area of robotics where morality is a relevant field of endeavor (i.e. human-robot interaction) the sub-discipline of morality does not exist. In response, the Utilibot project seeks to provide a point of initiation for the implementation of ethics in an AMR. The Utilibot is a decision-theoretic AMR guided by the utilitarian notion of the maximization of human well-being. The core ethical decision-making capac-ity of the Utilibot consists of two dynamic Bayesian net-works that model human and environmental health, a dynamic decision network that accounts for decisions and utilities, and a Markov decision process (MDP) that decom-poses the planning problem to solve for the optimal course of action to maximize human safety and well-being.
... The selection of the relevant variables is generally based on interviews with experts, descriptions of the domain, and an extensive analysis of the purpose of the network under construction. Often, knowledge about the (patho)physiological processes concerned is used to guide the identification of the relevant variables [24,29]. (2) Identification of the relationships among the variables: Once the variables to be included in the network have been decided upon, the dependence and independence relationships between them have to be analysed and expressed in a graphical structure. ...
Due to the uncertainty of many of the factors that influence the performance of an emergency medical service, we propose using Bayesian networks to model this kind of system. We use different algorithms for learning Bayesian networks in order to build ...
The paper proposes a new approach to hierarchical classification based on condition-action rules that represent expert knowledge in a given domain. The approach adopts a voting metaphor: each rule is regarded as a voter that expresses a preference for a given category to be assigned to an item to be classified; the category that receives more votes wins. Novel performance measures of hierarchical classifiers are also introduced that aim at overcoming the limitations of the current concepts of precision and recall. The proposed approach can be applied to any hierarchical classification task, for which expert knowledge is available. The viability of the approach and its performance are shown through a real-size application concerning the e-mail dispatching task inside a large public administration. The results obtained demonstrate that the proposed knowledge-based approach to hierarchical classification can reach a performance level comparable to that of human experts, if not even better.
The central role played by uncertainty in medical decision making explains why medicine was amongst the first areas where
applications based on Bayesian networks were developed. Biomedical research is firmly grounded on statistical methods; new
methods are adopted only slowly by the field. During the past decade, however, Bayesian networks have become important tools
for building decision-support systems in medicine and are now steadily becoming main stream. More recently, Bayesian networks
have also been adopted as analytic tools in human biology, mainly in research that aims to elucidate the biological mechanisms
underlying disease. In this paper, we review some of the applications in both medicine and human biology and we make an attempt
to unravel some of the characteristics of Bayesian networks in biomedicine.
Medical patient management is a complicated process, usually involving a large amount of, possibly uncertain, information. Clinicians may, therefore, require some form of decision support to deal with complicated situations; assistance in exploring various clinical questions, e.g., concerning prognosis and optimal treatment, may be valuable in this respect. Decision-theoretic expert systems provide a suitable framework for such assistance due to the flexibility of the underlying formalisms, with inherent potentials of knowledge reuse. In this paper, the development of a decision-theoretic model of non-Hodgkin lymphoma of the stomach is described, and examined for its clinical usefulness. Central to the model is a probabilistic network that offers an explicit representation of the uncertainties underlying the decision-making process.
Congenital Heart Disease (CHD) represents the most common group of congenital malformations of the heart and of its blood
vessels. In this paper, we present an ontology-based approach to detect abnormalities and malformations due to CHD. In particular,
we propose a formal and well-defined model to represent the anatomy of the cardiovascular system, based on the SNOMED vocabulary.
The model defines either the anatomy of the cardiovascular system in normal patients or the anatomy characterized by malformations
and abnormalities in CHD patients. We have formalized this model in OWL ontologies and SWRL rules and, then, we have used
a logic reasoner to identify either CHD patients or the heart abnormalities and malformations they are affected by.
Book on knowledge-based (expert) systems, published in 1991. Contains a description of principal methods and techniques and implementations in Prolog and Lisp. It still offers a useful description of some of the basics of artificial intelligence, although developed from the beginning of 1990s on are of course missing. Some people might like the algorithmic description of rule-based systems and reasoning with uncertainty. It also includes description of early systems hard to find nowadays.
A causal network is used in a number of areas as a depiction of patterns of ‘influence’ among sets of variables. In expert systems it is common to perform ‘inference’ by means of local computations on such large but sparse networks. In general, non‐probabilistic methods are used to handle uncertainty when propagating the effects of evidence, and it has appeared that exact probabilistic methods are not computationally feasible. Motivated by an application in electromyography, we counter this claim by exploiting a range of local representations for the joint probability distribution, combined with topological changes to the original network termed ‘marrying’ and ‘filling‐in‘. The resulting structure allows efficient algorithms for transfer between representations, providing rapid absorption and propagation of evidence. The scheme is first illustrated on a small, fictitious but challenging example, and the underlying theory and computational aspects are then discussed.
We develop an algorithm that can evaluate any well-formed influence diagram and determine the optimal policy for its decisions. Since the diagram can be analyzed directly, there is no need to construct other representations such as a decision tree. As a result, the analysis can be performed using the decision maker's perspective on the problem. Questions of sensitivity and the value of information are natural and easily posed. Modifications to the model suggested by such analyses can be made directly to the problem formulation, and then evaluated directly.
Bayesian networks are defined, and the chain rule for Bayesian networks is stated. Outlines of algorithms provided: inference in Bayesian networks, sensitivity analysis, EM for parameter learning, and learning structure. Copyright © 2009 John Wiley & Sons, Inc.
For further resources related to this article, please visit the WIREs website.
A model of carbohydrate metabolism has been implemented as a causal probabilistic network, allowing explicit representation of uncertainties involved in the prediction of 24-h blood glucose profiles in insulin-dependent diabetic subjects. The parameters of the model were based on experimental data from the literature describing insulin and carbohydrate absorption, renal loss of glucose, insulin-independent glucose utilisation and insulin-dependent glucose utilisation and production. The model can be adapted to the observed glucose metabolism in the individual patient and can be used to generate predicted 24-h blood glucose profiles. A penalty is assigned to each level of blood glucose, to indicate that high and low blood glucose levels are undesirable. The system can be asked to find the insulin doses that result in the most desirable 24-h blood glucose profile. In a series of 12 patients, the system predicted blood glucose with a mean error of 3.3 mmol/l. The insulin doses suggested by the system seemed reasonable and in several cases seemed more appropriate than the doses actually administered to the patients.