Structure of the knowledge base

↑

⇐ ПредыдущаяСтр 5 из 5

 Structure of the knowledge base

Knowledge base of ESTHER system consists of 19 knowledge bases for the decision classes

(different classes of poisonings) and 12 knowledge bases for differential diagnostics.

After the construction of complete and noncontradictory  knowledge base, methods of

questionnaire theory [6] have been used to produce an optimal decision tree for each class of

medicines. Criterion of optimality here is a minimum average number of questions needed to arrive to a conclusion about the existence and severity of poisoning by the drug of this class [7]. Decision tree presents a strategy of putting questions about values of clinical signs with the goal to ascertain the fact of intoxication and state of a patient. Each node of the tree is a question about a value of a diagnostic sign. Depending on the answer the following node is chosen. One of the decision trees embedded into the ESTHER knowledge base is presented on Fig 1.

Fig. 1.The constructed optimal decision tree for diagnosing of poisoning by one of the medicines.

System architecture

Principal modules of ESTHER system and their interactions are represented at the diagram (see

Fig.2).

There are two basic modes of a user’s work with ESTHER system – data input mode and consultation mode. In the data input mode a user sees a list of clinical signs grouped under the titles: “External

examination data”, “Anamnesis”, “Cardiovascular system”, etc. All information about a patient

including his/her state, illness history, complaints and so on is stored in the “Case Box” module.

“Hypotheses Advancement” module monitors this information and makes

suppositions about a possible poisoning cause.  If there are several concurrent hypotheses it selects one that is supported by the largest number of typical signs’ values. When a hypothesis for checking is formed, it is transferred to the “Hypotheses Checking” module when system is in the consultation mode.

In the consultation mode the system successively checks and puts forward new hypotheses.

While checking a hypothesis about a poisoning by some drug, the system uses the constructed optimal decision trees. A pass through the tree is accomplished and at the nodes, where values of clinical signs are still unknown, a user is questioned about values of appropriate signs to reach a conclusion. As a result of this pass one of the following conclusions is drawn: “Serious poisoning by medicine A”, “Mild poisoning by medicine A” or “No poisoning by medicine A”. The last means that either the poisoning was caused by the medicine other than X or there is no drug poisoning at all. In general case it may happen as a result of hypotheses checking that several hypotheses about poisonings by different medicines are confirmed. This may be the case if patient actually has taken several medicines as well as when different medicines result in similar clinical pattern. In order to refine the diagnosis the “Differential Diagnostics” module is applied.

An oriented graph is constructed in the “Differential Diagnosis” module. Nodes of the graph correspond to confirmed hypotheses and arcs correspond to the relations of poisonings’ probabilities between different medicines. For example, if differential diagnostics rules state that poisoning by medicine A is more probable than poisoning by medicine B, then nodes A and B in the graph are connected by the arc, which originates at node A and enters node B. After that, the graph is condensed, that is a kernel of undominated nodes is extracted. For each node A belonging to the kernel the string “Poisoning by medicine A is most probable” is written to the final diagnosis (module “Diagnosis”). For the other nodes B the string “Poisoning by medicine B is not excluded” is written to the diagnosis. Our discussion with expert revealed that these two wordings are sufficient for practical purposes and there is no need in more precise distinguishing among poisonings’ probabilities.

There is also the module of recommendations on treatment as a constituent part of the ESTHER system (“Treatment” on the diagram). It can be used as a reference manual for getting acquainted with basic methods of treatment of poisonings by one or another medicine. Another, more important feature is its ability to give recommendations on the particular given case of poisoning. Using a diagnosis made by other modules of the system and extracting clinical pattern of poisoning from the “Case Box” module it is able to recommend treatment with regard to the gravity of poisoning, degree of damage to the systems of human organism. It can also prescribe a symptomatic therapy. Moreover, when prescribing treatment methods and dosage of antidote medicines it is possible to take into account age, weight and illness history of the patient.

 Conclusion

From our point of view the ability of expert system to behave like a human is the critical factor

for its application. The developed ESTHER system includes large knowledge base that closely imitates

expert knowledge and reasoning in the problem of drug poisonings diagnostics.

It is important to note that the open architecture of ESTHER provides a possibility to add new

decision classes (causes of intoxications) and new diagnostic signs for such classes. A revision of

individual knowledge bases is also possible for any decision class using knowledge of a  more

experienced (for this class of poisonings) physician.

The expert system for the diagnostics of acute drug poisonings is the example of “knowledge

distribution” for large groups of specialists. We believe that appearance of new diseases and new

drugs make the problem of “knowledge distribution” for large areas more important.

References

1. Pople, H.E. : The Formation of Composite Hypotheses in Diagnostic Problem Solving: An

Exercise in Synthetic Reasoning, The Fifth International Joint Conference on Artificial

Intelligence, Boston, MIT, 1977.

2. Shortliffe, E.H.: Computer-based Medical Consultations: MYCIN, New York, American

Elsevier, 1976.

3. Larichev, O.I., Mechitov, A.I., Moshkovich, H.M., Furems E.M.: The Elicitation of Expert

Knowledge, Nauka, Moscow, 1989 (in Russian).

4. Larichev, O.I., Moshkovich, H.M., Furems, E.M., Mechitov, A.I., Morgoev, V.K.: Knowledge

Acquisition for the Construction of the Full and Contradictions Free Knowledge Bases, IEC

ProGAMMA, Groningen, The Netherlands, 1991.

5. Larichev, O.I., Bolotov, A.A.: The DIFKLASS System: Construction of Complete and

Noncontradictory Expert Knowledge Bases in Problems of Differential Classification, Automatic

Documentation and Mathematical Linguistics, Vol. 30, N. 5, Allerton Press, New York, 1996.

6. Parhomenko, P.P.: The Questionnaires Theory: Review, Automatics and Telemechanics, N. 4,

Moscow, 1970.(in Russian)

7. Naryzhnyi, E.V.: Construction of the Optimum Questioning Strategy for a Complete Expert Data

Base, Automatic Documentation and Mathematical Linguistics, Vol. 30, N. 5, Allerton Press,

New York, 1996.

Познавательные статьи:



Где возникла философия и почему?

Относительная высота сжатой зоны бетона

Сущность проекции Гаусса-Крюгера и использование ее в геодезии

Тарифы на перевозку пассажиров