In most countries of the world in recent years significantly increased morbidity and mortality associated with cardiovascular diseases. Today all become more widespread automated methods of analysis and interpretation of the ECG.
As a recognition system core neural network method was chosen. The use of neural network analysis in clinical practice improves the accuracy of the cardiovascular diseases diagnosis. In this paper, for the ECS pretreatment was chosen an approach based on ECS segmentation on 3 key areas that are responsible for depolarization of the atria, ventricles depolarization and repolarization of the ventricles. Was chosen as the neural network – multilayer perceptron with the logistic activation function, which is layered (3 layers) and without feedback. Such structure is characterized by a consistent feature extraction from the source image, which contributes to more efficient recognition.
To find the optimal number of neurons in the hidden layer has been proposed objective function, which will tend to peak at the positive correlations of sensitivity and specificity in the zone of minimum training error.
This approach makes it possible to select such a combination of sensitivity and specificity values for which the increase of one parameter does not cause the fall of another.In the future expected to conduct research on the definition of assessment methodologies the quality of training databases for constructing an optimal training set, as well as the development of neural systems analysis of ECS for the household, screening and day-aided analysis of the functional state of the cardiovascular system.