Electrocardiogram anomaly detection method based on VQ-VAE2 and deep neural network method

Abstract

The invention relates to the technical field of electrocardiogram anomaly detection, in particular to an electrocardiogram anomaly detection method based on a VQ-VAE2 and a deep neural network method. The electrocardio anomaly detection method based on the VQ-VAE2 and the deep neural network method comprises the following steps that 1, two training databases, namely an atrial fibrillation training database and a non-atrial fibrillation training database are obtained, and data processing is conducted on the atrial fibrillation training database; 2, VQ-VAE2 training and priori training are carried out on the atrial fibrillation training database after data processing, and a new electrocardiogram image is generated; and 3, atrial fibrillation heart rate type identificationis conducted, namely, the new electrocardiogram data finally generated in the step 2 and the atrial fibrillation training database are mixed together to serve as an atrial fibrillation sample set, then the atrial fibrillation sample set and a non-atrial fibrillation training database are input into a deep neural network for discrimination. The electrocardiogram anomaly detection method based on the VQ-VAE2 and the deep neural network method is provided for a doctor for outputting more accurate evaluation data, improving the diagnosis accuracy and efficiency.

Description

technical field [0001] The invention relates to the technical field of electrocardiogram abnormality detection, in particular to an electrocardiogram abnormality detection method based on VQ-VAE2 and a deep neural network method. Background technique [0002] An electrocardiogram (ECG) is a graph formed by recording the changes in the electrical activity of the heart every cardiac cycle from the body surface. A variety of heart diseases in humans can be characterized by an electrocardiogram. Atrial fibrillation is the most common sustained cardiac arrhythmia. According to statistics, the incidence of atrial fibrillation is 1%-2%, and the prevalence of atrial fibrillation gradually increases with age. Diseases of the heart itself, such as heart failure, valvular disease, and myocardial infarction, are significantly associated with atrial fibrillation. In fact, because physiological signals are affected by individual internal changes, there is currently no unified classific...

AI Technical Summary

Problems solved by technology

Imbalanced samples and small total sample size will lead to poor performance of the trained deep learning model

[0004] GAN is often used to solve the problem of sample imbalance. However, training GAN needs to achieve Nash equilibrium. At present, people have not found a good way to achieve Nash equilibrium. Therefore, compared with VAE (variational autoencoder) or PixelRNN, GAN training is unstable

On the other hand, the samples generated by GAN do not fully capture the diversity in the real distribution

At the same time, it is very difficult to evaluate the generated confrontation network, and there is still a lack of a more general metric for judging whether the model is overfitting in the test set

Images