Abstract
(English)
|
The contribution of the Geneva University Hospital to the GALEN-IN-USE project has been mainly concerned with the improvement and extension of the previously developed natural language processing (NLP) tools for clinical medicine. Indeed, while most of competitive work on medical terminologies is conducted in English, there is a real need to be multilingual in order to meet the various needs of different linguistic and cultural communities. That's why, GALEN-IN-USE has led to the creation of a new Europe-wide multilingual and multicultural infrastructure for cooperative development and maintenance of medical terminology. However, the GALEN's technology mainly centred around a common reference model for medical concepts (so-called CORE Model) that is supported by a formal language for concept representation (so-called GRAIL), has led to a knowledge representation which is necessarily complex and rather unfamiliar to most potential users. This is where NLP techniques play a significant role and have been promoted by the Geneva partner as follows:
· On the one hand, a natural language generator has been developed. It allows knotted and complex GRAIL expressions to be naturally displayed to the user through sentences formulated in his native (or at least known) natural language. Such a tool has been especially improved for the assessment of surgical procedure classifications, thus allowing professional language proposed by expert clinicians to be compared to the generated controlled vocabulary in order to finalise the linguistic labels of the coding system.
· On the other hand, the multilingual analyser, so-called the RECIT system (initially developed as part of the European HELIOS project), has been successfully adjusted for the analysis of surgical procedure rubrics. It takes as input natural language sentences written in English or French, which have been previously reformulated by the domain expert in order to produce unambiguous and self contained natural language expressions. This allows the analyser to yield as output valid representation which can then be (nearly) automatically mapped to the GALEN CORE model.
· Finally, the development in parallel of the generator and the analyser, in a multilingual context, has relied on a strong reuse of linguistic as well as conceptual knowledge maintained in modular knowledge bases. In particular, extensive annotations of the model, performed in parallel for various latino-greek languages (e.g. English, French, Italian, German,...) that share in common linguistic features, has led to the elaboration of multilingual dictionaries.
All these NLP techniques have been successfully applied for the elaboration and assessment of a new French coding system for surgical procedures called CCAM. This classification is currently developed by the GALEN's French collaborating centre at the University of Saint Etienne and is planned to cover 15 sections each containing about 500 rubrics. Up to now, half of the classification has been evaluated by our NLP tools and a contract with the French government is underway to achieve its full assessement.
|
