Background The improvement of medical informatics and the introduction of computers in all medical fields have modified the nature of the information manipulated by diagnosis support systems. Systems must be able to store information, to consult validated clinical tables, and to compare the results for several diseases.1 This suggests the production of strongly inter-connected process modules, sharing a common core of information. We have to problem to solve. The first one is to build a system which can treat several types of information about diseases. Ordinary, such systems try to aggregate information and to compute the diagnosis with a new global procedure. We assume that it is better, in first, to analyse each type of information with each treatment associated to it and after that to aggregate the different results into a global advice. The second part of the problem is to conceive a data organisation which allows to store only one patient description, enough generic to work for all diagnosis procedures.
Methods We have adopted an Object Oriented Design to create the Knowledge Data Base and to obtain an architecture which allows the collaboration between the treatment modules.2 The model of the Knowledge Data Base contains a generic object Description composed from three sets of signs: present signs, missing signs and information signs (to store non-pathological data as sex, age\ldots). Both patient and disease descriptions inherit their structure from the object Description. This organisation ensures the consistency between the patient and disease descriptions and simplifies their comparison. Two methods of diagnosis are defined: evaluation of criteria lists issued from bibliography (as ARA criteria), and scoring calculus from an expert advice. The first one eliminates all diseases for which the patient have exclusion signs. If the patient presents all criteria of a disease, the diagnosis is validated, else the score calculus produced an ordered list of possible. So, we can have a final advice about the patient case which is an aggregation of the two processes.
Results The realised software contains a KBD with more than one hundred signs to define 13 diseases in inflammatory rheumatology. 132 records of patients who present one of these 13 rheumatic diseases, was analysed. 83 (62.9%) patients presented a complete criteria list for one disease. The scoring method classified the right disease in the head of the list in 106 (80.3%) cases, and the aggregation of the two methods leads to make the right diagnostic for 120 (90.9%) cases.
We have shown that it is possible to conceive such hybrid system which can deal with several types of information and can compare diseases for which information is heterogeneous. For example criteria list is missing for three diseases. We think that this way produces better results than when we try to mix all information in only one type of value and after that to compute with.
Beurton-Aimar M, Vernhes JP, Le Blanc B, Dehais J, Salamon R. Creating a knowledge base for diagnosis making in rheumatology. Revue du rhumatisme1997;11:762
Beurton-Aimar M, Le Blanc B, Vernhes JP. LADRI: a decision making system in inflammatory rheumatology. Proceedings of the 7th International Conference on Intelligent Systems, July 1–2, 1998, 51–6
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.