Electrocardio-beat feature automatic extraction and classification methods based on deep learning method

Abstract

The invention relates to electrocardio-beat feature automatic extraction and classification methods based on a deep learning method. The electrocardio-beat feature automatic extraction method comprises the following steps: (1) removing high-frequency noise and baseline drift by utilizing biorthogonal wavelet transform; (2) detecting R waves by utilizing maximum and minimum values generated by binary sample wavelet transform; and (3) detecting a QRS wave group and P and T waves on the basis of the R waves in step (2). Deep learning classification (heart beat learning classification) is carriedout on the detected wave form data information by utilizing a bi-directional long short-term memory (Bi-LSRM) network. The methods provided by the invention have the advantages that the feature extraction procedure is effectively simplified, the wave forms are accurately located, the electrocardio signals are accurately classified, etc.

Description

technical field [0001] The invention belongs to the technical field of electrocardiogram detection, and in particular relates to a method for automatically extracting and classifying electrocardiogram beat features based on a deep learning method. Background technique [0002] Among the various parameters of the human body, heart activity analysis is a key part of intelligent judgment, and electrocardiography (ECG) is an important means of non-invasive examination and diagnosis of arrhythmia and other heart diseases widely used in the world. It is an important indicator of cardiac periodic activity and has been widely used in clinical practice. Arrhythmia is an important group of diseases in cardiovascular diseases. It is an extremely common and very important symptom of abnormal cardiac electrical activity caused by the origin of cardiac activity and (or) conduction disturbances resulting in abnormal cardiac beat frequency and (or) rhythm. It can occur alone or in combinat...

AI Technical Summary

Problems solved by technology

The existing diagnostic models are basically based on the low-level characteristics of abnormal ECG signals in a limited time period. The complexity of some difficult diseases makes it difficult to describe them with some rules. This diagnostic model has great limitations.

Although it has a certain auxiliary effect on the judgment of common types of cardiovascular diseases, the diagnosis of complex diseases still relies heavily on the doctor's experience and diagnostic level

The main reason for the poor diagnostic effect of the traditional model is that the traditional model has limited learning ability and cannot establish an organic connection between the low-level features of the ECG signal and the high-level semantic features of the experience and diagnostic knowledge of cardiovascular doctors.

In other words, traditional models cannot Figure 1 In this way, comprehensive knowledge and experience can be used to fully mine all the useful information of ECG signals.

[0005] Recently, some researchers have done a lot of deep learning work to detect abnormal ECG signals, such as using 1-D convolutional neural network (CNN) for ECG classification or using 34-layer cnn for arrhythmia detection, but these technologies are mostly based on CNN , most of these improvements focus on designing more complex, deeper and wider cnn networks, and aim to learn feature representations based on large and diverse datasets, these methods further improve the convolutional neural network (CNN) The disadvantage is that it is only efficient and effective under certain specific data structures, and cannot extract and classify ECG signals in depth and accurately.

Images