Skip to main content

An Access System for Medical Records Using Natural Language


Experience of existing Hospital Information Systems (HISs) highlights the fact that most relevant medical information is stored in narrative form, in Patient Discharge Summaries (PDSs). These PDSs aim at the transmission of the minimal but sufficient set of data to be used for the next visit of the patient. They are dictated, and stored on the computer by the secretary. The overall goal of this project is to provide better access to this information through a variety of services called "MENELAS":
- The management staff of the medical unit may consult its own electronic activity board.
- The clinical staff may also access the stored PDSs to retrieve a set of patients having specific characteristics, which allows some form of case based reasoning.
- The same facility allows a limited test of research hypotheses, on the basis of the available patient sample. This retrospective analysis may be pursued by a prospective clinical trial when preliminary results are positive.

These services will be implemented as modular subsystems, which can be realised gradually. Within the scope of this project, two services will be implemented: an electronic activity board system, and a query answering system, thus already providing a large part of the potential benefits of the available information.
The project involves the design and implementation of a pilot system a medical reports through natural languages. This system will be able to analyse t medical reports and to store them in a database as a set of conceptual structure may then be consulted to retrieve any specific information contained in the pati summaries (PDS).

User needs were assessed by investigations in hospitals. Input users of the system was obtained by several means. A pedagogic mock up of the sy in collaboration with medical staff. The major types of information that differe would want to use were determined with the help of these potential users.
Partne art software and competencies were examined and integration has begun.

Specific provided for all communication interfaces and data used by the system: grammar o semantic representation formalism. Software components from partners have been m and adapted: morphological analysers, syntactic analysers and general lexical fo and Dutch; semantic and pragmatic analysers are reusing existing principles to b independent components; general tools for knowledge and system management are be for the project needs. A new information retrieval component has been developed using state of the art techniques on retrieval from knowledge bases.

Data and k components have been collected in the conceptual graph formalism: a concept type taxonomy of domain concepts) and a semantic lexicon (semantic definitions, in co of domain words). These are in an advanced stage of progress.

The language anal will allow the analysis of medical natural language in patient records and provi representation for their automated encoding; this will contribute to the task on standardization, classification and encoding.
Technical Approach

The basis of the project is the design and implementation of a pilot system able to access medical reports through natural languages. This system will be able to analyse the contents of medical reports and to store them in a database as a set of conceptual structures; these structures, which may be seen as the core representation of the narrative, may then be consulted to retrieve any specific information contained in the PDS. An index will also be produced according to existing medical nomenclatures. The seven partners of MENELAS represent three linguistic groups (French, English and Dutch) which will be covered during the project.

Outline of Overall Workplan

There are three main phases in the project. During the first phase (months 1-18), the software tools already owned by the partners are brought together and adapted to achieve a first integration; where necessary, new components are developed. An experimental prototype version of the system is thoroughly tested and evaluated for the two main languages (French and English). The second phase (months 19-29) subsequently extends the prototype system into a pilot system, and completes work on the third language. The third phase of the project (months 30-36) is dedicated to testing and evaluating the system on hospital sites.


Service d'Informatique Médicale de l'Assistance Publique
91 Boulevard De L'hôpital
75634 Paris

Participants (6)

1 Place Des Marseillais
94227 Charenton
United Kingdom
4 Buccleuch Place
EH8 9LW Edinburgh
IBM France SA
3 & 5 Place Vendome
75021 Paris
18 Avoca Avenue, Blackrock
Herestraat 49
3000 Leuven

Dublin 2