Single-lead ECG signal classification method and system based on improved residual network

Abstract

The invention discloses a single-lead ECG signal classification method and system based on an improved residual network, and the method comprises the steps: taking 5 seconds as a window length, and segmenting an original ECG signal; and processing the signal by using an improved residual network, inputting the segmented data into the residual network, and outputting a result after processing as an identification result of the corresponding ECG signal. According to the invention, heart beat extraction processing is not required to be performed on original electrocardiosignals; according to the method, classification and identification of five types of arrhythmias, namely, Normal (N), Superventricular (S), Ventricular (V), Fusion (F) and Unnown (Q), can be realized; and the comprehensive accuracy rate of the test on an MIT-BIH arrhythmia database reaches 99% or above.

Description

technical field [0001] The invention relates to the technical field of information processing, in particular to a single-lead ECG signal classification method and system based on an improved residual network. Background technique [0002] At present, arrhythmia detection mainly uses ECG, that is, electrocardiographic signal for detection and diagnosis. There are many defects in the traditional manual diagnosis method: the workload is heavy. It is greatly affected by the experience and level of readers, and errors are prone to occur. Due to the above problems, many automatic detection algorithms have been proposed, but the existing various algorithms generally have the following two problems: the need to sub-beat the ECG signal, that is, to identify the P wave, QRS wave group, etc., on this basis In order to detect various abnormalities. Errors inevitably exist in the ECG signal sub-beating process. Once an error occurs, it will have an impact on the subsequent arrhythmia ...

AI Technical Summary

Problems solved by technology

There are many defects in the traditional manual diagnosis method: heavy workload

It is greatly affected by the experience and level of readers, and errors are prone to occur

Due to the above problems, many automatic detection algorithms have been proposed, but the existing various algorithms generally have the following two problems: the need to sub-beat the ECG signal, that is, to identify the P wave, QRS wave group, etc., on this basis In order to detect various abnormalities

[0004] In the prior art, the arrhythmia recognition and classification method based on sparse representation and neural network (public number: CN108647584A) requires parameter extraction and complicated preprocessing (the original ECG needs to be sub-shot, and dimensionality reduction and other processing are required), which can Distinguish 6 types of arrhythmia

[0005] In the existing technical literature, Robust ECG Signal Classification for Detection of Atrial Fibrillation Using a Novel Neural Network uses a convolutional neural network, but it can only identify normal, atrial fibrillation, noise and other four categories, and the comprehensive accuracy can only reach 82%

[0006] In the existing technical literature, Cardiologist-Level Arrhythmia Detection with Convolutional NeuralNetworks uses a convolutional neural network and recognizes it in a sequential manner, but it can only recognize 12 types of arrhythmia, and the comprehensive accuracy rate is lower than 80%.

Images