R.V. Isakov, J.A. Lukyanova
Researches on artificial neuron networks are related to that the method of treatment of information a human brain in a root differs from methods, applied by ordinary digital computers.
Long-term researches, conducted with the most different obvious algorithms, rotined that medical tasks, having non-obvious character, decided obvious methods with accuracy and comfort, quite insufficient for the wide practical use in the concrete tasks of diagnostics, prognostication and making decision.
Procedure, used for the process of teaching, is named a teaching algorithm. This procedure ranges in set procedure sinaptic weight of neuron network for providing of necessary neurons intercommunications structure.
It is possible to offer two basic variants of neuron network organization for the automated analysis of electrocardiography signals.
It is possible to consider architecture the first variant similar with classic description of multi-layered perceptron, and second - modular construction of neuron network, which consists of a few parallel located neuronetwork modules, built on the basis of multi-layered perceptron structure.
As a source of entrance data for the neuronetwork analyzer of electrocardiography signals the single channel recorder of electrocardiogram, productive registration of electrocardiosignal in one of the standard leads, was chosen.
The structures of teaching databases were offered for two variants of organization of neuron network. For filling of the developed databases the open informative system of «PhisioNET» was used.
The created and trained neuron network it is assumed to use jointly with a single channel electrocardiosignals recorder in the domestic systems of the automated analysis of the functional state of the cardiac-vascular system, and also during a mass express-researches for the selection of «risk groups»