S. V. Zhernakov, M. A. Shulakova

There is a problem of development of Clinical Decision Support System (CDSS) for hypertension diagnosis which considers features of diagnostics data. First of all this data are defined as nonformalized, uncertain and fuzzy. When we interpret diagnostics data, we can-t use all-purpose formal rules because of uniqueness of diagnostic objects. Therefore modern CDSS are based on Artificial Intelligence Technology and Soft Computing. Another keyword problem of automated data processing is the clarity of decisions - generation. The Hybrid System Technology is applied to solve described problems. Two data processing levels are distinguished - interpretation level and computing level. Interpretation is performed by Expert System. Computing is performed by Neural Network (NN). The purpose of the research is the adaptation of NN technology to decision support in clinical diagnosis of hypertension. Firstly, comparative analysis of four NN models is carried out. As a result the Hopfield NN is chosen as the simplest tool for CDSS. Secondly, NN module structure is designed. It allows to process heterogeneous information by using different transfer functions while increasing responsiveness and precision of the system. Thirdly, Degree Hypertension Detection Module was developed step-by-step: NN target vector is described, methodology of input data pre-processing is designed, output data interpretation mechanism is described. Target vector represents points of stability, which describes centers of classes. Six points in tridimensional space are designated for hypertension degree detection. They represent six classes of patients - state. Input data pre-processing methodology is developed for transformation of different input variables to NN computing form. Output signal enters the Interpretation module. There it is transformed according to Knowledge Base rules to user-readable form. An NN Module implementation is performed by Matlab. As a result it is confirmed that NN module of Expert System of hypertension diagnosis system allows to diagnose in "non-factors" conditions. It also successfully processes diversified input data and has the ability to interact with Knowledge Base of Interpretation system to achieve output results transparency.