Decentralized case-based reasoning and Semantic Web technologies applied to decision support in oncology

Mathieu d'Aquin, Jean Lieber, Amedeo Napoli
2013 Knowledge engineering review (Print)  
This article presents the KASIMIR system dedicated to decision knowledge management in oncology and which is built on top of Semantic Web technologies, taking benefit from standard knowledge representation formalisms and open reasoning tools. The representation of medical decision protocols, in particular for breast cancer treatment, is based on concepts and instances implemented within the description logic OWL DL. The knowledge units related to a protocol can then be applied for solving
more » ... ic medical problems, using instance or concept classification. However, the straight application of a protocol is not always satisfactory, e.g., because of contraindications, necessitating an adaptation of the protocol. This is why the principles and methods of case-based reasoning in the framework of description logics have been used. In addition, the domain of oncology is complex and involves several specialties, e.g. surgery and chemotherapy. This complexity can be better undertaken with a viewpointbased representation of protocols and viewpoint-based reasoning, for either application or adaptation of the protocols. Accordingly, a distributed description logic has been used for representing a viewpointbased protocol. The application and the adaptation of the viewpoint-based protocol to medical cases is carried out using global instance classification and decentralized case-based reasoning. 2 MATHIEU D'AQUIN ET AL. "patient" when there is no ambiguity. For most medical cases -about 60 to 70%-the protocol is simply and straightforwardly applied. The remaining non typical medical cases, called out-of-protocol cases, are examined by a committee of physicians called the breast therapeutic decision committee, for adapting the protocol and making a therapeutic decision. More precisely, given an out-of-protocol patient description, say P, the breast therapeutic decision committee generally selects a class of patients whose description is close to the patient description P and adapts the associated therapeutic decision to propose a treatment for the patient. This adaptation of the protocol can be modeled within a case-based reasoning (CBR) process, where the problem to be solved corresponds to a patient description and the solution to a therapeutic decision to be applied to the patient. Oncology has to be regarded as a complex domain where several specialties, e.g. chemotherapy, surgery, and radiotherapy, are involved. For each specialty, different characteristics of the patient are analyzed and taken into account for setting on a specific treatment within the whole treatment process. Accordingly, during a meeting of the breast therapeutic decision committee, each expert provides a view specific to his/her specialty on the treatment, as a part of a collective recommendation. The oncology specialties determine interrelated viewpoints for a patient treatment, i.e. units of information on a given patient in a local viewpoint can be shared and combined with other units of information in another local viewpoint to build a global treatment. Moreover, a decision taken in a local viewpoint, i.e. for a particular oncology specialty, may have an influence on the decision to be taken in another local viewpoint. Hence, knowledge representation and reasoning within the KASIMIR system have to consider -and to take advantage of-the multiple viewpoints involved in the decision. Until now, two versions of the KASIMIR system have been developed. The first version is based on an object-based representation formalism [d'Aquin et al., 2004] , and the second version is based on a semantic portal using description logics (DLs) and knowledge formalisms dedicated to the Semantic Web [d'Aquin et al., 2005a] . Being a standard for knowledge representation and exchange for Semantic Web applications, the Web ontology language OWL 1 , particularly the sub-language OWL DL, provide an adequate formalism for a reusable representation of knowledge contained in a decision protocol. Reasoning mechanisms associated with OWL DL, such as subsumption, classification, and instantiation, can be efficiently used for decision support in oncology. Finally, it should be noticed that there are still new developments in the KASIMIR system [Meilender et al., 2012] . Applying Semantic Web technologies in such a challenging domain leads to interesting issues, emphasizing some of the distinctive strengths and limitations of available formalisms such as OWL. Actually, an originality of the present research work is that the KASIMIR system implements various non-typical elements within a DL framework, namely: • The representation of a decision protocol for associating a treatment recommendation with a patient description is based on a domain ontology implemented within the OWL DL description logic, and used for knowledge representation and reasoning. • The representation of adaptation knowledge for protocol adaptation and CBR within a DL framework uses similarity paths for comparing two problems and adaptation paths for adapting the solution of a source problem to a target problem considered as being close to the source problem. • The representation of viewpoints within the KASIMIR system -one viewpoint corresponding to one oncology specialty-is based on a distributed DL formalism, namely C-OWL. The C-OWL formalism provides means for local representation and reasoning within a viewpoint, and global reasoning across several viewpoints, leading to decentralized CBR. Accordingly, the main contributions of the paper are the following: • We define and use CBR in a distributed environment based on distributed description logics and implemented in C-OWL. This leads to the original notion of "decentralized CBR", which combines case-based reasoning and distributed description logics.
doi:10.1017/s0269888913000027 fatcat:v5qtwne66jhh5hm6fhsd65u7ga