An application of self-organising maps for a knowledge base for use in cardiac domain

The conception, project realisation and updating of hospital information systems (HISs) rely on specific standardisation procedures. Particularly important is the momentum that technology is giving to the production of computer-assisted clinical documentation. Developing such a task, which includes issues like specific diagnostic reports as well as a total clinical record, requires making explicit the syntactic and semantic relationships among the data that describe the patient's health. Our study tackles the realisation of a knowledge-based clinical record designed as a combination of modules depending on a set of main medical concepts. We have heuristically defined the knowledge base's content, which has a frame-based structure, through the self-organising map (SOM) technique. We have used the SOM approach for its intrinsic capability of grouping the words of a text according to the semantic categorisations of the natural language. A set of 30 reports, referring to patients afflicted with cardiac diseases and written in the Italian language, gives us the experimental setting that we have used to train the SOM and to obtain the knowledge base. The lists of words which activate the different nodes of the SOM are assumed as frames of the required knowledge base. Performance evaluation shows that this method gives us a representation of the medical knowledge that is efficient for producing meaningful clinical sentences. Moreover, this approach is of general interest: in fact, it works independently of the language used for writing clinical documents.