Classification method for arrhythmia based on one-dimension convolution neural-network and S transformation

A convolutional neural network and arrhythmia technology, which is applied in the measurement of pulse rate/heart rate, medical science, sensors, etc., can solve the problems of large amount of calculation, ECG signal cannot express the relationship between time and frequency domain, and the classification effect is not obvious. , to achieve the effect of improving accuracy, speeding up convergence, and fast convergence

Inactive Publication Date: 2018-03-20
TIANJIN UNIV
View PDF11 Cites 60 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, through experimental verification, although some hidden features can be extracted, the classification effect is not obvious, and the amount of calculation is large.
[0008] The transformation-based method converts the signal from the time domain to the frequency domain. For example, the Fourier transform can obtain the frequency domain characteristics of the signal, but it cannot be used to represent the time-frequency for the non-stationarity of the ECG signal. Interrelationships between domains

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Classification method for arrhythmia based on one-dimension convolution neural-network and S transformation
  • Classification method for arrhythmia based on one-dimension convolution neural-network and S transformation
  • Classification method for arrhythmia based on one-dimension convolution neural-network and S transformation

Examples

Experimental program
Comparison scheme
Effect test

specific example

[0072] The ECG signal data of this example comes from public databases. The specific process of the example is as follows:

[0073] (1) Preprocessing: In terms of noise removal, the median filter method is used to remove baseline drift, and the low-pass filter is used to remove power line interference and high-frequency noise. R wave detection uses amplitude threshold and wavelet threshold.

[0074] (2) Feature extraction: such as figure 2 As shown, according to the R position detected by the preprocessing, the 90 sampling point signals before the R wave position, the 197 sampling point signals after the R wave position, a total of 288 sampling points of the ECG segment, extract 1D-CNN features and time-frequency domain characteristics.

[0075] The 1D-CNN features of this example are: build a 3-layer 1D-CNN network, and each layer of CNN network contains a convolutional layer and a pooling layer. The size of the convolution kernel is 3, and the numbers are 64, 128, and 2...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

Provided is a classification method for arrhythmia based on a one-dimension convolution neural-network and S transformation. The method includes the steps of pre-possessing electrocardiosignal; extracting the deep nonlinear characteristic of the electrocardiosignal by using the one-dimension convolution neural-network; extracting the time-frequency domain characteristic of the electrocardiosignalby using the S transformation; combining the deep nonlinear characteristic of the electrocardiosignal with the time-frequency domain characteristic of the electrocardiosignal, continuing to conduct characteristic learning after passing through a whole-connection layer, and obtaining the output characteristic of the whole-connection layer; conducting classification after the output characteristic of the whole-connection layer is connected to the softmax layer of the one-dimension convolution neural-network; outputting a classification result. According to the method, it is not necessary to conduct compression and bilinear interpolation on the electrocardiosignal to obtain the picture form of fixed pixel points to extract the characteristics. According to the method, in the aspect of characteristic extraction, the advantages of a deep learning characteristic and the time-frequency domain characteristic are combined to be composed into more complete characteristics, convergence can be quickened, over-fitting is controlled, and the insensitivity of the interwork to initialization weight is lowered. The accuracy rate of multiple kinds of arrhythmia identification is improved.

Description

technical field [0001] The invention relates to a method for classifying arrhythmia. In particular, it relates to an arrhythmia classification method based on one-dimensional convolutional neural network and S-transform. Background technique [0002] Arrhythmia is an abnormal phenomenon of irregular heart function caused by disturbance of the rate, rhythm or conduction of the heart's electrical signals, and is considered the most common heart disease. It can be detected by analyzing the recorded ECG waveform, which is formed by different potentials associated with the depolarization and repolarization patterns of the atria and ventricles, and provides important information about the condition of the heart. For doctors, it is difficult to analyze long-term ECG signals in a short period of time and observe the small morphological changes of ECG signals, which may cause doctors to lose important information during diagnosis. Another problem faced by ECG signal analysis is the...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
Patent Type & Authority Applications(China)
IPC IPC(8): A61B5/024A61B5/0452A61B5/00
CPCA61B5/024A61B5/7267A61B5/349
Inventor 吕卫王粟瑶褚晶辉
Owner TIANJIN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products