Marcelo Gómez

An Intelligent Decision System for Jaundice Management in Newborn Babies

Jaundice management in newborn babies is a complex problem faced almost everyday by neonatology services. This suggests the development of a decision support system integrating the beliefs and preferences of members of a neonatology service. In this paper, we introduce one such system called IctNeo. We describe first the basic representation of the problem by means of influence diagrams, including admission and treatment decisions. The corresponding uncertainty model is built with the aid of both historical data and subjective judgements. We then describe the multiattribute preference structure. Finally, we provide some ideas on the system implementation and use.

Multiattribute Utility Analysis in the IctNeo System

We have constructed IctNeo, a decision support system for neonatal jaundice management which integrates beliefs and preferences with an influence diagram. This paper considers the problem of modelling preferences by means of a multiattribute utility function. We describe the objectives hierarchy including a hierarchical structure of attributes, tests of independence which lead to a nested utility function and the assessment of imprecise weights and component utility functions.

Implementation of IctNeo: a Decision Support System for Jaundice Management

IctNeo is a complex decision support system (DSS) for the management of a common disease: neonatal jaundice. This article discusses the system implementation with a wide point of view, which has produced a general tool for DSSs construction with influence diagrams (IDs). The knowledge representation of the problem as an ID, and the user interface are only the specific modules for IctNeo. The other components can be reused for further developments. A grammar is defined for IDs representation, dividing the specification and manipulation of this graphical model. The compiler of IDs (expressed with this grammar) will produce a compiled version of the diagram, once it is correct and prepared for evaluation. Finally, various mechanisms are required to reduce the computational complexity of the evaluation of so large IDs: a genetic algorithm to determine an efficient sequence of node reductions, evidence incorporation from patient data and a recursive computation of the utility function. The evaluator gathers these tools and provides a set of decision tables with each patient proposals.