MIT-BIH Noise Stress Test Database Research Articles

Elimination of Random Mixed Noise in ECG using Convolutional Denoising Autoencoder with Transformer Encoder.

Electrocardiogram (ECG) signals frequently encounter diverse types of noise, such as baseline wander (BW), electrode motion (EM) artifacts, muscle artifact (MA), and others. These noises often occur in combination during the actual data acquisition process, resulting in erroneous or perplexing interpretations for cardiologists. To suppress random mixed noise (RMN) in ECG with less distortion, we propose a Transformer-based Convolutional Denoising AutoEncoder model (TCDAE) in this study. The encoder of TCDAE is composed of three stacked gated convolutional layers and a Transformer encoder block with a point-wise multi-head self-attention module. To obtain minimal distortion in both time and frequency domains, we also propose a frequency weighted Huber loss function in training phase to better approximate the original signals. The TCDAE model is trained and tested on the QT Database (QTDB) and MIT-BIH Noise Stress Test Database (NSTDB), with the training data and testing data coming from different records. All the metrics perform the most robust in overall noise and separate noise intervals for RMN removal compared with the baseline methods. We also conduct generalization tests on the Icentia11k database where the TCDAE outperforms the state-of-the-art models, with a 55% reduction of the false positives in R peak detection after denoising. The TCDAE model approximates the short-term and long-term characteristics of ECG signals and has higher stability even under extreme RMN corruption. The memory consumption and inference speed of TCDAE are also feasible for its deployment in clinical applications.

QRS Detection Based on Discrete Wavelet Transform for ECG Signal with Motion Artifacts

Motion artifacts in ECG signals recorded during physical exercises activities can affect the diagnosis of arrhythmia. To minimize the faults in arrhythmia detection, it was important to choose accurate algorithm for detecting QRS in ECG signal with noises produced during physical movements of the patients. Therefore, choosing the QRS detection algorithm with good competency for the signal affected by noises and motion artifacts is needed for arrhythmia detection analysis. The QRS detection based on Discrete Wavelet Transform was implemented and presented in this paper. The performance of the algorithm was assessed using the MIT-BIH Arrhythmia Database and MIT-BIH Noise Stress Database. For the MIT-BIH Arrhythmia database, the average Sensitivity (Se) and positive Predictivity (+P) of the algorithm were 98.24% and 98.61%, respectively. The algorithms had a lower average false negative rate (FNR) than the pan Tompkins algorithm when applied to the MIT-BIH noise stress test database, which was 0.033% for record 118 and 0.032% for record 119, respectively. The results demonstrated that the algorithms perform well when dealing with arrhythmia data and motion artifact at various levels of signal to noise ratio.

DeScoD-ECG: Deep Score-Based Diffusion Model for ECG Baseline Wander and Noise Removal.

Electrocardiogram (ECG) signals commonly suffer noise interference, such as baseline wander. High-quality and high-fidelity reconstruction of the ECG signals is of great significance to diagnosing cardiovascular diseases. Therefore, this paper proposes a novel ECG baseline wander and noise removal technology. We extended the diffusion model in a conditional manner that was specific to the ECG signals, namely the Deep Score-Based Diffusion model for Electrocardiogram baseline wander and noise removal (DeScoD-ECG). Moreover, we deployed a multi-shots averaging strategy that improved signal reconstructions. We conducted the experiments on the QT Database and the MIT-BIH Noise Stress Test Database to verify the feasibility of the proposed method. Baseline methods are adopted for comparison, including traditional digital filter-based and deep learning-based methods. The quantities evaluation results show that the proposed method obtained outstanding performance on four distance-based similarity metrics with at least 20% overall improvement compared with the best baseline method. This paper demonstrates the state-of-the-art performance of the DeScoD-ECG for ECG baseline wander and noise removal, which has better approximations of the true data distribution and higher stability under extreme noise corruptions. This study is one of the first to extend the conditional diffusion-based generative model for ECG noise removal, and the DeScoD-ECG has the potential to be widely used in biomedical applications.

Power-line interference and baseline wander elimination in ECG using VMD and EWT

Electrocardiogram (ECG) is a critical biomedical signal and plays an imperative role in diagnosing cardiovascular disorders. During ECG data acquisition in clinical environment, noise is frequently present. Various noises such as powerline interference (PLI) and baseline wandering (BLW) distort the ECG signal which may lead to incorrect interpretation. Consequently, substantial emphasis has been dedicated to ECG denoising for reliable diagnosis and analysis. In this study, a novel hybrid ECG denoising method based on variational mode decomposition (VMD) and the empirical wavelet transform (EWT) is presented. For effective denoising using the VMD and EWT approach, the noisy ECG signal is decomposed within narrow-band variational mode functions (VMFs). The aim is to remove noise from these narrow-band VMFs. In current approach, the centre frequency of each VMF was computed and utilized to design an adaptive wavelet filter bank using EWT. This leads to effective removal of noise components from the signal. The proposed approach was applied to ECG signals obtained from the MIT-BIH Arrhythmia database. To evaluate the denoising performance, noise sources from the MIT-BIH Noise Stress Test Database (NSTDB) are used for simulation. The assessment of denoising performance in based on two key metrics: the percentage-root-mean-square difference (PRD) and the signal-to-noise ratio (SNR). The findings of the simulation experiment demonstrate that the suggested method has lower percentage root mean square difference and higher signal-to-noise ratio as compared to existing state of the art denoising methods. An average output SNR of 24.03 was achieved, along with a 5% reduction in PRD.

Convolutional block attention autoencoder for denoising electrocardiograms

Electrocardiograms are commonly used to detect cardiovascular diseases, so it is important that they are of high quality. However, various sources of noise, such as baseline wander, muscle artifact, and electrode motion can obfuscate the signal of the heart recorded by different monitors. In this work, we propose a novel algorithm, convolutional denoising autoencoder with block attention module (CDAE-BAM), for removing noise from electrocardiograms by leveraging attention in a convolutional denoising autoencoder. We propose an attention block including both spatial and channel attention. Spatial attention captures the location of relevant features within channels in a signal, and channel attention captures the most relevant channels in a signal. We validate our proposed algorithm’s performance in removing noise from the MIT-BIH Noise Stress Test Database from electrocardiogram signals in the QT Database, the Computing in Cardiology Challenge 2017 Database, and the Medical Information Mart for Intensive Care Database. We show that this method outperforms eight other state-of-the-art methods with respect to sum of squared distances, mean absolute distance, and cosine similarity.

An ECG denoising method based on adversarial denoising convolutional neural network

The Electrocardiogram (ECG) is widely used to diagnose heart disease. However, ECG recordings are often polluted by different noises in real situations. It is of great significance to propose an effective method for noise rejection of ECG signal. With the development of deep learning, neural networks have been widely utilized in signal processing. Most of the existing models of deep neural networks have millions of parameters, which places higher requirements on medical computing power. Based on this, this paper proposes a new denoising framework named the Adversarial Denoising Convolutional Neural Network (ADnCNN). We adopt the ADnCNN model to learn the residual signals in the noisy signal to obtain clean signal. A discriminator network is used for two-class classification of the denoised and clean signals, which are fed back to the ADnCNN model for parameter adjustment. The proposed method is validated by dividing three different types of datasets on the MIT-BIH arrhythmia database, the MIT-BIH noise stress test database, and the QT database for experiments. The denoising performance of the ADnCNN model is represented by the Signal-to-Noise Ratio (SNR), Root Mean Square Error (RMSE) and Percentage Root Mean Square Difference (PRD). The generality and adaptability of the model are also demonstrated. Simulation results show that the proposed method has advantages compared to recently proposed models.

ECG Signal Denoising Method Based on Disentangled Autoencoder

The electrocardiogram (ECG) is widely used in medicine because it can provide basic information about different types of heart disease. However, ECG data are usually disturbed by various types of noise, which can lead to errors in diagnosis by doctors. To address this problem, this study proposes a method for denoising ECG based on disentangled autoencoders. A disentangled autoencoder is an improved autoencoder suitable for denoising ECG data. In our proposed method, we use a disentangled autoencoder model based on a fully convolutional neural network to effectively separate the clean ECG data from the noise. Unlike conventional autoencoders, we disentangle the features of the coding hidden layer to separate the signal-coding features from the noise-coding features. We performed simulation experiments on the MIT-BIH Arrhythmia Database and found that the algorithm had better noise reduction results when dealing with four different types of noise. In particular, using our method, the average improved signal-to-noise ratios for the three noises in the MIT-BIH Noise Stress Test Database were 27.45 db for baseline wander, 25.72 db for muscle artefacts, and 29.91 db for electrode motion artefacts. Compared to a denoising autoencoder based on a fully convolutional neural network (FCN), the signal-to-noise ratio was improved by an average of 12.57%. We can conclude that the model has scientific validity. At the same time, our noise reduction method can effectively remove noise while preserving the important information conveyed by the original signal.

Deep Convolutional Generative Adversarial Network with LSTM for ECG Denoising.

The electrocardiogram (ECG), as an essential basis for the diagnosis of cardiovascular diseases, is usually disturbed by various noise. To obtain accurate human physiological information from ECG, the denoising and reconstruction of ECG are critical. In this paper, we proposed an ECG denoising method referred to as LSTM-DCGAN which is based on an improved generative adversarial network (GAN). The overall network structure is composed of multiple layers of convolutional networks. Furthermore, the convolutional features can be connected to their time series order dependence by adding LSTM layers after each convolutional layer. To verify the effectiveness and the denoising performance of the improved network structure, we test the proposed algorithm on the famous MIT-BIH Arrhythmia Database with different levels of noise from the MIT-BIH Noise Stress Test Database. Experimental results show that our method can remove the single noise and the mixed noise while retaining the complete ECG information. For the mixed noise removal, the average SNRimp, RMSE, and PRD are 19.254 dB, 0.028, and 10.350, respectively. Compared with the state-of-the-art methods, DCGAN, and the LSTM-GAN methods, our method obtains the higher SNRimp and the lower RMSE and PRD scores. These results suggest that the proposed LSTM-DCGAN approach has a significant advantage for ECG processing and application in complex scenes.

An efficient ECG denoising method by fusing ECA-Net and CycleGAN.

For wearable electrocardiogram (ECG) acquisition, it was easy to infer motion artifices and other noises. In this paper, a novel end-to-end ECG denoising method was proposed, which was implemented by fusing the Efficient Channel Attention (ECA-Net) and the cycle consistent generative adversarial network (CycleGAN) method. The proposed denoising model was optimized by using the ECA-Net method to highlight the key features and introducing a new loss function to further extract the global and local ECG features. The original ECG signal came from the MIT-BIH Arrhythmia Database. Additionally, the noise signals used in this method consist of a combination of Gaussian white noise and noises sourced from the MIT-BIH Noise Stress Test Database, including EM (Electrode Motion Artifact), BW (Baseline Wander) and MA (Muscle Artifact), as well as mixed noises composed of EM+BW, EM+MA, BW+MA and EM+BW+MA. Moreover, corrupted ECG signals were generated by adding different levels of single and mixed noises to clean ECG signals. The experimental results show that the proposed method has better denoising performance and generalization ability with higher signal-to-noise ratio improvement (SNRimp), as well as lower root-mean-square error (RMSE) and percentage-root-mean-square difference (PRD).

A Conv -Transformer network for heart rate estimation using ballistocardiographic signals

ObjectiveContinuous heart rate (HR) monitoring has great implications for the prevention of chronic diseases, and we use non-contact ballistocardiography(BCG) technology to estimate HR. MethodsIn this paper, overnight BCG data are acquired from 78 patients using a 10-channel piezoelectric sensor matrix with a sampling rate of 50 Hz. 200 sets of non-overlapping 10 s quiet period data are selected for follow-up work, with a total of 15,600 segments. A Conv-Transformer network with Pyramid input (HRCTP-net) is used to estimate HR values for these segments, where local features are obtained by CNN and global features are calculated by the transformer. This is an end-to-end network without additional post-processing. During the experiment, electrocardiogram (ECG) noise which from the MIT-BIH noise stress test database is also introduced for data augmentation to further improve the network generalization ability. ResultsTaking the synchronously collected ECG as the ground truth, the results in the 6-fold cross-validation show that the proposed method achieves the best results on mean absolute error (MAE), standard deviation of absolute error (SDAE) and pearson correlation coefficient (PCC) with 0.93 bpm, 1.31 bpm and 0.97, respectively. ConclusionTo the best of our knowledge, this paper is the first time to introduce transformer and ECG noise into BCG signal analysis and demonstrate their effectiveness. SignificanceOur proposed HRCTP-net has potential and promise in healthcare applications.

A lightweight U-net for ECG denoising using knowledge distillation

Objective. Electrocardiogram (ECG) signals are easily polluted by various noises which are likely to have adverse effects on subsequent interpretations. Research on model lightweighting can promote the practical application of deep learning-based ECG denoising methods in real-time processing. Approach. Firstly, grouped convolution and conventional convolution are combined to replace the continuous conventional convolution in the model, and the depthwise convolution with stride is used to compress the feature map in the encoder modules. Secondly, additional identity connections and a local maximum and minimum enhancement module are designed, which can retain the detailed information and characteristic waveform in the ECG waveform while effectively denoising. Finally, we develop knowledge distillation in the experiments, which further improves the ECG denoising performance without increasing the model complexity. The ground-truth ECG is from The China Physiological Signal Challenge (CPSC) 2018, and the noise signal is from the MIT-BIH Noise Stress Test Database (NSTDB). We evaluate denoising performance using the signal-to-noise ratio (SNR), the root mean square error (RMSE) and the Pearson correlation coefficient (P). We use the floating point of operations (FLOPs) and parameters to calculate computational complexity. Main Results. Different data generation processes are used to conduct experiments: group 1, group 2 and group 3. The results show that the proposed model (U Lde -net) can improve SNRs by 10.30 dB, 12.16 dB and 12.61 dB; reduce RMSEs by 9.88 × 10−2, 20.63 × 10−2 and 15.25 × 10−2; and increase Ps by 14.77 × 10−2, 27.74 × 10−2 and 21.32 × 10−2. Moreover, the denoising performance after knowledge distillation is further improved. The U Lde -net has parameters of 6.9 K and FLOPs of 6.6 M, which are much smaller than the compared models. Significance. We designed a lightweight model, but also retain adequate ECG denoising performance. We believe that this method can be successfully applied to practical applications under time or memory limits.

Novel QRS detection based on the Adaptive Improved Permutation Entropy

Detection of the QRS complex is the most important step in analyzing ECG signals for heart monitoring and diagnosis. There have been several QRS-peak detection methods reported in the literature. Most of these methods have low performance under noisy conditions. In this paper, we propose a novel QRS detection algorithm based on a new Permutation Entropy (PE) method that we developed and referred to as the Adaptive Improved Permutation Entropy (AIPE) method. The parameters of the AIPE method are determined based on the specific signal properties. Implementing the AIPE method leads to prominently preserving the QRS complex and eliminating noises of the ECG signal without smoothing the ECG signal. Our simulations show that the proposed QRS detection algorithm is effective and robust under noisy conditions. The algorithm is validated on the MIT-BIH Noise Stress Test Database for various SNR values. In addition, we examined the algorithm’s performance under motion noise conditions, mimicking a practical scenario. We used the metrics of sensitivity, positive predictive, and F1 score to evaluate the performance of our algorithm and compare it with several other algorithms explained in the literature. Our investigation shows that the proposed algorithm is efficient and effective. More importantly, it is robust under noisy conditions providing superior performance over other recent and popular QRS detection algorithms, including the popular Pan–Tompkins and the recent Advanced Adaptive Multilevel Thresholding (AAMT) algorithms.

A Robustness Evaluation of Machine Learning Algorithms for ECG Myocardial Infarction Detection.

An automatic electrocardiogram (ECG) myocardial infarction detection system needs to satisfy several requirements to be efficient in real-world practice. These requirements, such as reliability, less complexity, and high performance in decision-making, remain very important in a realistic clinical environment. In this study, we investigated an automatic ECG myocardial infarction detection system and presented a new approach to evaluate its robustness and durability performance in classifying the myocardial infarction (with no feature extraction) under different noise types. We employed three well-known supervised machine learning models: support vector machine (SVM), k-nearest neighbors (KNN), and random forest (RF), and tested the performance and robustness of these techniques in classifying normal (NOR) and myocardial infarction (MI) using real ECG records from the PTB database after normalization and segmentation of the data, with a suggested inter-patient paradigm separation as well as noise from the MIT-BIH noise stress test database (NSTDB). Finally, we measured four metrics: accuracy, precision, recall, and F1-score. The simulation revealed that all of the models performed well, with values of over 0.50 at lower SNR levels, in terms of all the metrics investigated against different types of noise, indicating that they are encouraging and acceptable under extreme noise situations are are thus considered sustainable and robust models for specific forms of noise. All of the methods tested could be used as ECG myocardial infarction detection tools in real-world practice under challenging circumstances.

Electromyography Parameter Variations with Electrocardiography Noise

Electromyograms (EMG signals) may be contaminated by electrocardiographic (ECG) signals that cannot be easily separated with traditional filters, because both signals have some overlapping spectral components. Therefore, the first challenge encountered in signal processing is to extract the ECG noise from the EMG signal. In this study, the EMG, mixed with different degrees of noise (ECG), is simulated to investigate the variations of the EMG features. Simulated data were derived from the MIT-BIH Noise Stress Test (NSTD) Database. Two EMG and four ECG data were composed with four EMG/ECG SNR to 32 simulated signals. Following Pan-Tompkins R-peak detection, four ECG removal methods were used to remove ECG with different compensation algorithms to obtain the denoised EMG signal. A total of 13 time-domain and four frequency-domain EMG features were calculated from the denoised EMG. In addition, the similarity of denoised EMG features compared to clean EMG was also evaluated. Our results showed that with the ratio EMG/ECG SNR = 10 and 20, the ECG can be almost ignored, and the similarity of EMG features is close to 1. When EMG/ECG SNR = 1 and 2, there is a large variation of EMG features. The results of our simulation study would be beneficial for understanding the variations of EMG features upon the different EMG/ECG SNR.

A Novel ECG Denoising Scheme Using the Ensemble Kalman Filter.

Monitoring of electrocardiogram (ECG) provides vital information as well as any cardiovascular anomalies. Recent advances in the technology of wearable electronics have enabled compact devices to acquire personal physiological signals in the home setting; however, signals are usually contaminated with high level noise. Thus, an efficient ECG filtering scheme is a dire need. In this paper, a novel method using Ensemble Kalman Filter (EnKF) is developed for denoising ECG signals. We also intensively explore various filtering algorithms, including Savitzky-Golay (SG) filter, Ensemble Empirical mode decomposition (EEMD), Normalized Least-Mean-Square (NLMS), Recursive least squares (RLS) filter, Total variation denoising (TVD), Wavelet and extended Kalman filter (EKF) for comparison. Data from the MIT-BIH Noise Stress Test database were used. The proposed methodology shows the average signal to noise ratio (SNR) of 10.96, the Percentage Root Difference of 150.45, and the correlation coefficient of 0.959 from the modified MIT-BIH database with added motion artifacts.

Muscle Artifact Removal in Single-Channel Electrocardiograms using Temporal Convolutional Networks.

The electrocardiogram (ECG) is a vital diagnostic tool used in many health applications. In practice, interference by muscle artifacts is very common and may significantly complicate interpretation of the ECG waveform. In this work we investigate the removal of muscle artifacts in single-channel ECG signals using neural network models. To this end, we compare two neural network architectures which were previously used for ECG denoising and propose a novel third method based on the ConvTasNet. The neural networks are trained on simulated data using artificial mixtures of single-channel ECG (lead II) and surface EMG signals taken from publicly available datasets. ECG data were sampled from the MIT-BIH Arrhythmia and the PTB-XL database. The former provides recordings of ambulatory ECGs while the latter contains a large variety of cardiac pathologies recorded in a clinical setting. The muscle artifacts were sampled from the MIT-BIH Noise Stress Test database and the TaiChi database. In the past, most denoising methods were only tested on the two smaller MIT-BIH datasets. In this work, we report performances on larger datasets and thus provide stronger evidence for a clinical use-case. We also report out-of distribution performance of the three methods by switching the ECG dataset between training and test. The herein investigated variant of the ConvTasNet substantially reduces interference by muscle artifacts, outperforms state-of-the-art methods and thus, may support clinical decision making.

Channel-Wise Average Pooling and 1D Pixel-Shuffle Denoising Autoencoder for Electrode Motion Artifact Removal in ECG

This paper presents a channel-wise average pooling and one dimension pixel-shuffle architecture for a denoising autoencoder (CPDAE) design that can be applied to efficiently remove electrode motion (EM) artifacts in an electrocardiogram (ECG) signal. The three advantages of the proposed design are as follows: (1) In the skip connection layer, less memory is needed to transfer the features extracted by the neural network; (2) Pixel shuffle and pixel unshuffle techniques with point-wise convolution are used to effectively reserve the key features generated from each layer in both the encoder and decoder; (3) Overall, fewer parameters are required to reconstruct the ECG signal. This paper describes three deep neural network models, namely CPDAELite, CPDAERegular, and CPDAEFull, which support various computational capacity and hardware arrangements. The three proposed structures involve an encoder and decoder with six, seven, and eight layers, respectively. Furthermore, the CPDAELite, CPDAERegular, and CPDAEFull structures require fewer multiply-accumulate operations—355.01, 56.96, and 14.69 million, respectively—and less parameter usage—2.69 million, 149.7 thousand, and 55.5 thousand, respectively. To evaluate the denoising performance, the MIT–BIH noise stress test database containing six signal-to-noise ratios (SNRs) of noisy ECGs was employed. The results demonstrated that the proposed models had a higher improvement of SNR and lower percentage root-mean-square difference than other state-of-the-art methods under various conditions of SNR.

Attention-Based Convolutional Denoising Autoencoder for Two-Lead ECG Denoising and Arrhythmia Classification

This paper presents a fast and accurate electrocardiogram (ECG) denoising and classification method for low-quality ECG signals. To achieve this, a novel attention-based convolutional denoising autoencoder model is proposed that utilizes skip-layer and attention module for reliable reconstruction of ECG signals from extreme noise conditions. Skip-layer connections are used to reduce information loss while reconstructing the original signal, and a lightweight, efficient channel attention module is used to update relevant features retrieved via cross-channel interaction efficiently. The model is trained and tested using four widely available databases. For evaluation, the signals are mixed with simulated additive white Gaussian noise ranging from -20 to 20 decibels (dB) and MIT-BIH Noise Stress Test Database (NSTDB) noise ranging from -6 to 24 dB. The model outperformed the most cited published works, achieving an average Signal-to-noise ratio (SNR) improvement of 19.07±1.67 and percentage-root-mean-square difference (PRD) of 11.0% at 0 dB SNR noise. The model classification performance on 60, 000 beats are 98.76 ± 0.44% precision, 98.48 ± 0.58% recall, and 98.88 ± 0.42% accuracy, respectively using a stratified five-fold cross-validation strategy.

DeepRTSNet: Deep Robust Two-Stage Networks for ECG Denoising in Practical Use Case

In this paper, we develop a low-cost cellular internet of medical things (IoMT)-based electrocardiogram (ECG) recorder for monitoring heart conditions and used in practical cases. In order to remove noise from signals recorded by these non-clinical devices, we propose a cloud-based denoising approach that focuses on utilizing deep neural network techniques in the time-frequency domain through the two stages. Accordingly, we exploit the fractional Stockwell transform (FrST) to transfer the ECG signal into the time-frequency domain and apply the deep robust two-stage network (DeepRTSNet) for noise cancellation. Due to the practical use case, the various heart physiologies and noise levels in different amplitudes and frequencies are needed to be robust against wide-range noises in actual conditions. We utilize the MIT-BIH Apnea-ECG database (APNEA-ECG) with several different heart physiologies. Next, the different noises consisting of muscle artifacts (MA), baseline wander (BW), and electrode motion (EM) from the MIT-BIH Noise Stress Test Database (NSTDB) and random noise, are added to the signals. The main focus of the noise generation part is the fast Fourier transformation (FFT) of the simulated noisy signal and the practical noisy signal has a maximum cross-correlation to gain a better morphological resemblance between realistic signals and the prepared datasets. Based on the results, DeepRTSNet outperforms prior learning-based methods and conventional non-learning approaches in terms of signal-to-noise ratio (SNR), root mean square error (RMSE), and percent root mean square difference (PRD). Moreover, outcomes reveal that DeepRTSNet has an extraordinary performance with a certain amount of further complexity than others.

Two-stage ECG signal denoising based on deep convolutional network

Background. An electrocardiogram (ECG) is an effective and non-invasive indicator for the detection and prevention of arrhythmia. ECG signals are susceptible to noise contamination, which can lead to errors in ECG interpretation. Therefore, ECG pretreatment is important for accurate analysis. Methods. The ECG data used are from CPSC2018, and the noise signal is from MIT-BIH Noise Stress Test Database. In the experiment, the signal-to-noise ratio (SNR), the root mean square error (RMSE), and the correlation coefficient P are used to evaluate the performance of the network. The method proposed is divided into two stages. In the first stage, a U de -net model is designed for ECG signal denoising to eliminate noise. The DR-net model in the second stage is used to reconstruct the ECG signal and to correct the waveform distortion caused by noise removal in the first stage. In this paper, the U de -net and the DR-net are constructed by the convolution method to achieve end-to-end mapping from noisy ECG signals to clean ECG signals. Result. In SNR, RMSE and P indicators, U de -net + DR-net proposed in this paper can achieve the best performance compared with the other five schemes (FCN, U-net etc). In the three data sets, SNR can be increased by 11.61 dB, 13.71 dB and 14.40 dB and RMSE can be reduced by 10.46 × 10−2, 21.55 × 10−2 and 15.98 × 10−2. Conclusions. Despite the contradictory results, the proposed two-stages method can achieve both the elimination of noise and the preservation of effective details to a large extent of the signals. The proposed method has good application prospects in clinical practice.