Special Issue On Case-based Reasoning In The Health Sciences

Appl Intell (2008) 28: 207–209 DOI 10.1007/s10489-007-0100-0

Special issue on case-based reasoning in the health sciences Isabelle Bichindaritz · Stefania Montani · Luigi Portinale

Published online: 6 October 2007 © Springer Science+Business Media, LLC 2007

1 Introduction This special issue features recent advances in applications of case-based reasoning to the health sciences. Case-based reasoning (CBR) is a problem solving paradigm that exploits the contextual knowledge of previously experienced situations, called cases. It basically consists in retrieving past cases that are similar to the current one and in reusing past successful solutions, followed by, if necessary, revising the proposed solution; the newly solved case can then be retained and added to the system knowledge base, called the case base. The retrieve, reuse, revise, and retain processes are known as the steps of the CBR cycle. In recent years, there has been an explosion of interest in health sciences applications of CBR, not only in the traditional CBR in medicine domain, but also in bioinformatics, in enabling home health care technologies, in CBR integration, and in synergies between CBR and knowledge discovery. Health care and health sciences research are expanding with population aging as well as increased access to health services, which fosters the growing development of computational methods in the health sciences. The rate of generaI. Bichindaritz () Institute of Technology, University of Washington, 1900 Commerce Street, Box 358426, Tacoma, WA 98402, USA e-mail: [email protected] S. Montani · L. Portinale Dipartimento di Informatica, University of Piemonte Orientale, Via Bellini 25/g, 15100 Alessandria, Italy S. Montani e-mail: [email protected] L. Portinale e-mail: [email protected]

tion of new biomedical data in electronic format in particular favors the development of approaches such as case-based reasoning, which has the advantage of being a data driven methodology from artificial intelligence. This editorial presents the high quality contributions selected for this special issue from researchers in case-based reasoning in the health sciences, as well as highlights some of the main research questions and advances in this domain.

2 Special issue contents Four of the six papers collected in this special issue are extended versions of works which were selected among the contributions presented at a workshop on CBR in the Health Sciences, held in Ölüdeniz, Turkey, in September 2006, and hosted by the ECCBR-06 (European Conference on Casebased Reasoning) conference. This workshop built upon progress made at the First Workshop on CBR in the Health Sciences, held at ICCBR-03 (International Conference on Case-based Reasoning), in Trondheim, Norway, at the Second Workshop on CBR in the Health Sciences, held at ECCBR-04, in Madrid, Spain, and at the Third Workshop on CBR in the Health Sciences, held at ICCBR-05, in Chicago, Illinois, USA. The main objectives of these workshops were to identify challenges specific to applying CBR to the health sciences, required methodological improvements to fit this context needs, preferred types and domains of application, and guidelines to better develop CBR systems in this field. The four pre-selected papers, together with the two additional contributions independently submitted and accepted for this special issue, address some of the objectives above. In particular, among the emerging challenges related to adopting CBR in the medical and biological domain, the complexity of case mining, feature representation, and case

208

retrieval is one of the most critical. As a matter of fact, on the one hand, a proper case mining strategy would enable CBR to capitalize on clinical databases, electronic patient records, and biomedical literature databases. Resorting to prototypical cases definition may prove to be a very useful methodological choice to accomplish such a goal, especially in the initial case base set up, but also in the case base maintenance phase. On the other hand, recent technological advances and specific domain needs frequently lead to deal with case data which are intrinsically high-dimensional, making feature representation and case retrieval potentially non-trivial. This holds true e.g. for data represented in the form of time series and of images. Prototype definition is afforded in the paper by M. Atzmueller et al., which proposes a case-based approach for characterizing and analyzing subgroup patterns. It presents techniques for retrieving characteristic factors and a set of corresponding cases for the inspection and analysis of a specific subgroup pattern. Then, the set of factors and cases are merged into prototypical cases for presentation to the user. This enables a convenient retrieval of (meta-)information associated with subgroup objects. This article provides an interesting synergy work between CBR and data mining in the service of data mining. The topic of mining for prototypical cases is central in the paper by I. Bichindaritz, where the close synergy between knowledge discovery, data mining, and CBR is explored for supporting automatic prototypical case learning from biomedical literature. The article validates the approach by presenting a comparison between the prototypical cases learnt from stem-cell transplantation domain with those created by a team of experts in the domain. In comparison with the previous paper, it provides an innovative synergy between CBR, text mining, and knowledge discovery in the service of case-based reasoning in biomedicine. The paper by P. Perner explores the problem of image data management, addressed by a case-based object recognition system reasoning from prototypical cases. The paper details a method of catalogue-based classification for image interpretation that can be easily adapted to different biomedical domains. The performance of the catalogue-based classifier is assessed and successfully validated by studying the accuracy and the reduction of the prototypes after applying a prototype-selection algorithm. High-dimensional data, this time in the form of time series, are also dealt with in the paper by N. Xiong et al., where a knowledge discovery approach to identify significant sequences for depicting symbolic time series cases is presented. Such discovered sequences are highly valuable in case characterization to capture important properties while ignoring random trivialities. Moreover, they serve as a means for reducing data dimensionality. Indexing mechanisms for case retrieval are also proposed, and verified by means of some experiments.

I. Bichindaritz et al.

The paper by J. Lieber et al. identifies a very interesting application domain for CBR: the one of medical protocols adaptation. Protocol adaptation can be seen as a knowledgeintensive case-based decision support process. Several issues need to be addressed while trying to model such process, such as the lack of relevant information about the patient, or the closeness to decision thresholds. As handling these issues requires some additional knowledge, which has to be acquired, different methods are presented to perform adaptation knowledge acquisition either from experts, or in a semi-automatic manner. The topic of proposing guidelines to develop CBR systems in the health sciences is afforded in the paper by S. Montani, which provides a detailed analysis of the reasons why CBR is not more integrated today in mainstream clinical practice. As a solution, the author suggests a closer synergy between CBR and other artificial intelligence methodologies, giving birth to a modular architecture, able to provide decision support. In the resulting framework, CBR, originally conceived as a well suited reasoning paradigm for medical applications, can extend its original roles, and cover a set of additional tasks, such as parameter configuration. In summary, in the opinion of the international review committee, the six papers collected in this special issue represent an excellent sample of the most recent advances of CBR in the health sciences, both as regards methodological enhancements and as regards interesting practical experiences, carried out by an international group of researchers from five different countries.

3 Main accomplishments and future issues The complexity of medical and biological domains has encouraged major advances in CBR methodology over the years. In addition, CBR systems in the health sciences have proved to be efficient at providing decision support recommendations, accurate classifications, and health care quality monitoring. The accomplishments of the collection of papers featured in this special issue encompass providing explanations for clusters learnt in cluster analysis, keeping the knowledge in a CBR system up-to-date, improving image and time series signal classification, providing care protocols more closely adapted to fit individual patients, and improving parameter configuration and other tasks in artificial intelligence systems. All of these accomplishments rely on case-based reasoning for advancing a particular set of problems in a particular application domain. They can be summarized as advancing case-based image and time series signal classification (P. Perner and N. Xiong et al.), improving decision-support (I. Bichindaritz and J. Lieber et al.),

Special issue on case-based reasoning in the health sciences

and expanding further the scope of CBR applications to improving other methodologies, such as data mining (M. Atzmueller et al.) or artificial intelligence methodologies like expert systems (S. Montani). The approaches have been validated through significant accuracy and/or precision and recall ratios, user satisfaction, or expert rating. However encouraging may be this current state of research in CBR in medicine and biology, the complexity of these domains provides many more opportunities for improvement of CBR methodology. Among technical improvements, we can cite advancing how to perform adaptation, evaluating similarity between cases involving continuous data, time series, images, texts, or genetic data, tailoring clinical guidelines and treatment protocols to a patient case, modeling more closely expert reasoning, and a smooth interaction between the user and the CBR system. Synergies between CBR, data mining, and knowledge discovery require more research too given the rate of increase in medical data available from electronic medical records, databases, and literature. Similarly, synergies between CBR and other computational methods such as expert systems, statistical analysis and inference are crucial to the development of CBR. All of these improvements should lead to CBR systems better integrated in clinical settings, and to the opportunity to evaluate these systems in real clinical environments, with the ultimate goal of showing positive clinical outcomes. Long

209

term research objectives are to convince clinicians to routinely adopt CBR methodology integrated with electronic medical records in the clinic. With the rate of increase in quality of CBR systems in biology and medicine as showcased in this special issue, this ultimate objective may be closer than we think. Acknowledgements We would like to thank all the authors for having decided to contribute to this special issue, and Prof. Moonis Ali, the chief editor of Applied Intelligence Journal, for his support in this project. We are also very grateful to the reviewers for their careful work: Syed Sibte Raza Abidi, Dalhousie University, Canada, Klaus-Dieter Althoff, University of Hildesheim, Germany, Paolo Avesani, Instituto Trentino di Cultura, Italy, Riccardo Bellazzi, University of Pavia, Italy, Peter Funk, Malardalen University, Sweden, Daniel Hennessy, University of Pittsburgh, USA, Alec Holt, Department of Information Science, University of Otago, New Zealand, Lakhmi C. Jain, University of South Australia, Australia, Jean Lieber, Loria, France, Cindy Marling, Ohio University, USA, Stefan V. Pantazi, University of Victoria, Canada, Enric Plaza, Spanish Scientific Research Council, Spain, Petra Perner, Institute of Computer Vision and Applied Computer Sciences, Germany, Francesco Ricci, University of Bolzano, Italy, Rainer Schmidt, Institut fur Medizinische Informatik und Biometrie, Germany, Olga Vorobieva, Institut fur Medizinische Informatik und Biometrie, Germany.

Guest Editors