Investigation of coding outputs of artificial neural networks for classification of heart rate variability

Artificial neural network (ANN) is information processing system, which differs from conventional systems with parallel aspect of information transfer process and the presence of self-regulation for a given objective function. These properties contribute to their use in medical diagnostics for decision making. Currently in clinical and preventive medicine is becoming more widely used method of analysis of heart rate variability.

In this paper, the possibility of using ANN for classification of the heart rate variability is considered. Special attention was given to two coding methods of artificial neural networks outputs used in the analysis tasks of heart rate scattergram and

histograms.

The first method is coding outputs of artificial neural networks to 6 etalons, and the second to 18 etalons.

Two databases were formed, one of them contains a histogram value and the second contains the binary matrices of

scattergram. Researches carried out by the frequent teaching of artificial neuron networks of the hidden layer different volume. After teaching, every network passed testing on the independent database. Studies have shown that network which has 6 etalons has the best characteristics of sensitivity. The specificity of network is high in all cases.