Electrocardiosignal noise reduction method based on variational self-coding and Pixel CNN model

Abstract

According to the electrocardiosignal noise reduction method based on the variational autoencoder and the Pixel CNN model, the variational autoencoder depends on probability distribution, reconstruction errors can be reduced, and the model can learn useful potential representation of data and effectively simulate the global structure of signals. The self-regression decoder based on Pixel CNN further optimizes the potential variables compressed by the variational self-encoder, and can capture a large number of potential features and boundary small-scale features at the same time. The Pixel CNN model facilitates modeling of local features that are complementary to global features of the VAE with a decomposed output distribution model. The model is expanded to potential variable hierarchical structures with different scales, so that the receptive field is increased, the calculation cost is reduced, and the extraction of detail feature information is facilitated.

Description

technical field [0001] The invention relates to the field of electrocardiographic signal noise reduction, in particular to an electrocardiographic signal noise reduction method based on variational autocoding and PixelCNN model. Background technique [0002] With the rapid development of the Internet of Things and artificial intelligence technology, remote monitoring and auxiliary diagnosis and treatment of heart disease have become a hot issue in the current medical field. The effective means of preventing and detecting heart disease is the electrocardiogram, but the weak, low frequency and instability of the electrocardiogram make the electrocardiogram signal highly susceptible to noise interference. Common noises include muscle artifact (MA), electrode motion (EM) and baseline. Drift (BW). MA destroys the details of the electrocardiogram (ECG), causing some of the hallmarks of heart disease to disappear. ST-segment deviation from baseline due to EM or BW may be misdiagn...

AI Technical Summary

Problems solved by technology

[0003] Existing ECG signal denoising techniques can be divided into traditional ECG denoising methods and deep learning denoising methods, but the existing denoising methods still face various problems, such as wavelet transform (WD), empirical Traditional noise reduction methods such as modal decomposition (EMD) and Wiener filtering (WF) generally only consider a certain type of noise, and cannot remove multiple noises at the same time

Denoising methods based on deep learning, such as denoising autoencoder (DAE), fully convolutional neural network, generative adversarial network, etc., have excellent results in denoising ECG signals, but there are still problems such as large amount of calculation, high complexity, and overkill. Problems with fitting and vanishing gradients

In addition, due to the large overlap between the noise spectrum and the ECG signal spectrum, these methods will lead to the loss of key information of the signal after noise reduction and the phenomenon of signal distortion.

Images