Electrocardiosignal feature automatic extraction method based on one-dimensional convolution neural network

Abstract

The invention discloses an electrocardiosignal feature automatic extraction method based on a one-dimensional convolution neural network. The extraction method comprises the following steps of electrocardiosignal data acquisition and preprocessing, establishment of the deep neural network, parameter learning of the deep neural network, reconstruction of the deep neural network into a multi-layer feature extractor, and feature extraction. The invention aims to provide the electrocardiosignal feature automatic extraction method based on the one-dimensional convolution neural network. Participation of medical experts is not needed, the step of pattern recognition is not needed, and effective feature values are automatically found from data, so that medical resources are more reasonably dominated and used; based on the one-dimensional convolution neural network, the multi-layer feature extractor is designed, a large amount of electrocardiosignal data is accumulated in the development of medical equipment, on the basis of the great amount of marked electrocardiosignal data, the feature values which are high in distinguishing degree and good in effect are learned, manpower is saved, andthe medical level is improved.

Description

technical field [0001] The invention relates to a method for automatically extracting features of electrocardiographic signals, in particular to a method for automatically extracting features of electrocardiographic signals based on a one-dimensional convolutional neural network. Background technique [0002] The traditional method of ECG signal data analysis relies on the professional knowledge of medical experts and requires the deep participation of medical experts. For example, the QRS complex in the ECG signal is related to the electrical activity of the ventricle. The narrow and fast QRS complex represents normal ventricular activity, while the wide and slow QRS complex indicates that there may be a problem with the conduction of the ventricle, suggesting that there may be a problem with heart disease. There are diseases such as ventricular premature beats and indoor conduction block. Therefore, the width of the QRS complex can be used as the feature value (Feature) o...

AI Technical Summary

Problems solved by technology

[0003] Relying on medical experts to extract eigenvalues ​​and analyze ECG signal data, there are the following problems: the design and extraction of eigenvalues ​​requires the in-depth cooperation between algorithm engineers and medical experts, which consumes a lot of manpower and material resources; the extraction of eigenvalues ​​depends on the model The identification method needs to first identify the heartbeat from a section of ECG signal, and then identify the P wave, QRS wave group, ST segment, T wave and other characteristic waves from each heartbeat, and finally extract the characteristic value

In addition, the differences between many diseases are very small, and if the eigenvalues ​​are not comprehensive enough, the analysis effect will be poor; from language description to eigenvalues ​​is also a huge challenge

For example, the ST-segment changes of ECG signal morphology related to myocardial ischemia are clinically described as "elevation". no precise definition

Images