A.N. Iusupov - Ph.D. (Biol.), Wellness Foundation (St. Petersburg)

The nature of the probability distribution of biomedical data and its impact on the efficiency of the application of multivariate statistical techniques was analyzed according to the literature data. The approach to standardize data using the integral function of probabilities was considered. An algorithm for the reconstruction of the integral distribution function on the experimental data with the use of artificial neural network special structure was developed. The structure of the artificial neural network, consisting of several series-connected neurons was proposed. Training sample is empirical curve of the integral of the probability distribution. It is shown that the possibility of application of neural networks a special structure for the approximation of the density function of the probability distribution. The effectiveness of the proposed neural network algorithm of restoration of function of distribution of probabilities in the example of two of biochemical indices of lipid metabolism (cholesterol lipoproteins of high density and triglycerides), received in 272 patients was demonstrated. As part of the demonstration, the probability standardization mentioned biochemical parameters was carried out. It was statistically proved that the distribution of these standardized indicators corresponds to the uniformly distribution.