Application of fuzzy cognitive maps using semantic web approaches to model medical knowledge E.I. Papageorgiou 1 , J. De Roo 2 , C. Huszka 2 and D. Colaert 2 1 Department of Informatics & Computer Technology, Technological Educational Institute of Lamia, Lamia, Greece, epapageorgiou@teilam.gr 2 Agfa HealthCare NV, Moutstraat 100, 9000 Gent, Belgium, {csaba.huszka, jos.deroo, dirk.colaert}@agfa.com Abstract—This study presents our investigation into a generic approach to model medical knowledge, using the semantic web framework, to have explicit knowledge with a clear semantic meaning and cognitive maps. Fuzzy Cognitive Maps are proposed to model the medical knowledge extracting by clinical guidelines as they have been proven by literature as powerful knowledge representation and reasoning tools. A general purpose reasoning engine, Eye, with the necessary plug-ins was developed to be able to perform the reasoning on this knowledge model. The concepts and causal relationships among them are transferred in semantic web using the notation3. Our effort is mainly concentrated to present the main aspects of formalizing medical knowledge using cognitive maps and semantic web. The advantage of this approach is to enable the sharing and reuse of knowledge from databases of guidelines and simplify maintenance. Keywords-fuzzy cognitive maps; knowledge representation; decision support; semantic web; notation3 1. INTRODUCTION Usually, the decision making process of the treating physician is supported by a large number of medical guidelines produced by medical committees for deciding the relevant therapy to apply to the presented patient [1]. Due to the sheer volume of one such guideline and the rigid structure of such guidelines, it is very difficult for the treating physician to apply and strictly follow the workflow presented by such guidelines. Thus there is a challenge to propose a medical decision support system that dynamically guides the physician through the workflow of guidelines. This workflow is not a static, because the workflow itself is adapting to the changing reality of the patient care [2]. Previous works were mainly based on static modeling of simple rules for medical decision support [3,4]. But the nowadays’ medical decision making is focused to more individualized treatment guidelines and the future goal of the research is adjust general guidelines to individual circumstances. In our approach, in order to keep the main characteristics of uncertainty in focus that is inherent in medical domain, we attempt to model medical knowledge using fuzzy cognitive maps which is a different approach from static structures such as decision trees [5]. FCMs have been proved from the literature as efficient medical decision-making and support techniques [6-9]. So currently, we focus on formalization of medical knowledge using dynamic influence graphs, such as FCMs, which create paths, based on the current and constantly changing clinical information of the patient, the environment and the guidelines. The undertaken work is focused on the establishment of the generic knowledge representation model using Fuzzy Cognitive Maps, implemented in the semantic web framework. Notation3 [10] was selected as an open and semantic web language to implement the fuzzy cognitive maps. A general purpose reasoning engine, Eye [11], and the necessary plug-ins to be able to perform the reasoning on these knowledge models were also developed and presented. The suggested approach is implemented to the Urinary Tract Infection problem caused by Escherichia coli and the preliminary trials show its effectiveness in medical decision support. 2. MODELING MEDICAL KNOWLEDGE In order to model medical knowledge it is essential to deal with the causal feature of the knowledge elements. It is well known that medical information is abundant and ever-growing, yet it hardly represents a complete set of knowledge, rather a limited set thereof. Medical knowledge can be represented in several ways. Perhaps the most commonly used is the clinical pathway or clinical guideline [12,13]. It is quite easy and handy to use by the clinician as it provides a simple decision tree algorithm that could be followed in the diagnostic or therapeutic path to aid conclusions. However, many guidelines are not so easy to use in daily practice due to the complexity of the clinical data [1,2]. Thus, there is a need to create a decision support system (DSS) that aims to recognise the appropriate guidelines in front of a medical problem. 2.1 Medical Models Modelling medical knowledge starts with a standard procedure, by defining the problem to be modelled (as a domain) and identifying all variables that represent this domain. Our model follows a goal-drive, need-oriented approach meaning that each model will serve a certain purpose and is custom-made. We identified two basic models to be used in the scope of this study: diagnostic model and therapeutic model.