Contents and Knowledge aggregation for case-based reasoning in the field of fœtal dysmorphology – ACCORDYS

Many organizations harbor unexploited repositories of tacit knowledge. These repositories can take the form of unstructured files which record the memories of past cases at various levels in the organization. Examples include business files in companies or patient records in hospitals. Accessing the explicit knowledge held in these memories would often help to handle present cases better. Their under-use is however often caused by a difficulty to access them in a relevant way, either because they are not available in digital form, or because of the inefficiency of standard access to these unstructured files through databases or full text search. To unlock this issue, project ACCORDYS tackles this specific point by a combined approach based on Knowledge Engineering, Natural Language Processing, and Information Retrieval. ACCORDYS tests these methods in a domain where access to past cases is especially useful and where its impact on society is significant: the health domain, more precisely that of rare diseases.
Project ACCORDYS more specifically aims at putting to use biomedical data which have been accumulated over the years by specialists of fetopathology, a branch of medical embryology and genetics whose topic is the study of fetus abnormalities. These data are heterogeneous (photographs, echography images, images obtained by multiple radiology techniques, biological test results, etc.). They are interpreted by specialists, most of whom author a text report. In the context of the prenatal diagnosis of an abnormality, knowledge of “resembling” past situations that are “solved” (i.e., checked after an autopsy of the fetus) is instrumental in diagnostic decision. This knowledge is precious since a majority of the abnormalities diagnosed in the prenatal period are rare affections which concern less than one out of 2000 people.
The issue addressed by project ACCORDYS is the present impossibility of making use in a simple way of the whole set of known cases. A large part of the data that were collected along the years remain on their original media and cannot be used to help solve new cases unless they are made accessible and linked to more general domain knowledge. A large proportion of the relevant documents (texts or images) are on non-digital media, and many of those on digital media are not correctly indexed. This makes it currently impossible to retrieve patient cases with the same pathology. The goal of the project thus meets an immediate need to gather all these data and to make them easily accessible so that these collective memories are put to good use.
Analysing the case reports (if needed after digitization) will endow them with added value in terms of knowledge, thus creating a domain knowledge base. The reports will be structured and enriched through indexing with termino-ontological resources and through linking to external knowledge (scientific literature, databases). Such a knowledge base will help access similar cases.
The framework for this pluridisciplinary project is fundamental research. From a medical point of view, it provides a way to cluster knowledge on rare diseases and to model it. From the point of view of access to information content, disregarding the domain, the project targets progress in methods of semantic access to and enrichment of documents.