[Principles and experiences for modeling chaotic attractors of heart rate fluctuations with artificial neural networks].

The aim of the present paper was, using artificial neural networks, to identify chaotic attractors presumed to be responsible for heart rate fluctuations. Chaotic behaviour is based on low-dimensional deterministic processes which are highly sensitive to initial conditions and therefore of limited predictability. Chaotic attractors produce orbits in the phase space where the points are dense. Following transformation of measured heart rate date into the phase space, such heart rate prediction characteristics were found in 6 adult rabbits. A low-dimensional deterministic model was designed by means of an artificial neural network [2*dimension of the time series +1) input neurons, 1 hidden layer, 1 output neuron and was successfully employed to predict the next 5 heart beats. Training of the neural network in terms of the prediction of the suspected chaotic orbits was dramatically improved by considering all possible prediction intervals within the prediction horizon. The trained models enabled the heart rate fluctuations to be distinguished during consciousness, under anaesthesia and additional vagal blockade.