D.V. Andreev, A.V. Sorokin THE RELATOR-BASED NEURAL PROCESSOR FOR IDENTIFICATION RANK VALUES OF AN ANALOG SIGNAL

R. Sadiqov, U. Мамеdova SOFT COMPUTING AND EXPERT - DIAGNOSTIC MEDICAL SYSTEM

Unverbal character of personal experience intuitions of a doctor based only on probable self-appraisals of individual’s behaviour in some hypothetical situations complicates the solution of the problem of development of expert-diagnostic medical system (EDMS) and stipulates the necessity of use of Soft Computing technology. The latter represents a combination of intellectual paradigms such as indistinct logic, artificial neuron network, developmental programming, theory of chaos and training, which have predetermined and catalyzed the processes of extensive development of theory and engineering of artificial intelligence.

Unformalized problems solved while developing EDMS have some pecularities as inaccuracy, ambiguity, incompleteness and inconsistency of knowledges about the problem and its tasks, large dimensionality of decision spaces and also dynamically varying data and knowledge.

It is evident that the gamma of such these pecularities requres qualitative organization of interaction of all components of the system simultaneous providing its functional - structural wholeness.

In EDMS which we have worked out special attention is paid to the representation of knowledges in logic output mechanism and a means of operation (work) with inaccuracy and uncertainty of input data and knowledges. EDMS is created on module technology e. i. based on typical unified programs of Fortran, Pascal, Prolog and is realized on the personal IBM PC/ AT type computer. After definition of causal-сorollary relationship (symptoms-ilnesses) based on monotone and nonmonotone reasoning. Minimum enveloping sets are created. The obtained results are passed through special filters and satisfactory solutions are found.

On the fasis of group selection consulation in selected to diagnose using the theory of vouting and cooperative decision. Adaptive solutions are obtained by deseribing obtained results on indistinct-neuron network. In case of necessity the system gives an extensive picture of causal-corollary relationships, condition and facts on diagnostics of diseases.

EDMS is applied for exspres-examination of occupational diseases.

A.P. Babcin, P.N. Isacov, O.N. Choporov PREDICTION OF DEVELOPMENT OF MACROANGIOPATHY BY SUGAR DIABETES ON THE BASIS OF NEURONETS

The value of information security of medical process engineering’s, as we know, is increasing nowadays. The implicit tasks of medicine and biology are an ideal fields in the sphere of action for neuronets technologies. Exactly in this sphere the brightest practical success of neuroinformational methods is observed.

The main task are creation and training of neuronet for prediction of development of vascular defeats by the patients with sugar diabetes.

A etiology and pathogeny of vascular defeats by sugar diabetes are not finally defined. This neuronets model, with is designed for predication of atherosclerotical development, is based on the following entering index: the general cholestherin, the average arterial pressure, the Quetelet index, daily average doze of insulin, maximum level of glucose, duration of ischemic illness of heart and others. The degree of heaviness of illness is defined on an exit of neuronet by average magnitude of a thickness “intima+media” complex.

It is created the three-layered neuronet with sequential connections containiry 60 neurons: 14 – on the enteriry layer, 39 on interior and 7 on output. The number of entering signals corresponds to analyzable indexes. As a transfer function is used the rational sigmoid. The training of neoronet is carreing out on the basis of “back propagation”.

For formiry of trainiry sample the inspection of 102 patients with sugar diabetes of the 1^st type were carried out. 80 of them come to the training sample, 22 made the control group for quality check of functioning model. The iterative process of turning of syneptical balance, is interrupted by reading an error on an exit of a net, with is smaller than 5% from an average value of an output index.

By testing of the model the average error of the prognosis has made 6,2% that testifies its good capacity for work and possibility of use in practical public health services.

The program realization of a system of neuronets modelliry is carried out on Visual C++. The program complex has the convenient and simple interface. It helps a doctor to use in his researches this neuronets model without complicated additional preparation.