Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Wavelet Integrated Convolutional Neural Network for ECG Signal Denoising

Takamasa Terada, Masahiro Toyoura

TL;DR

This work addresses the challenge of denoising wearable ECG signals where noise overlaps with ECG frequency bands. It introduces a wavelet-integrated CNN that includes a Discrete Wavelet Transform (DWT) layer to separate high- and low-frequency components, enabling frequency-band-specific learning through trainable filters (using the Daubechies 6, db6 wavelet). Across MIT-BIH and MIT-NST datasets, the backward-type CNN with a single wavelet layer delivers the best denoising performance in RMSE and $SNR$ improvements over a conventional fully convolutional baseline, with qualitative results showing better preservation of ECG morphology. The approach demonstrates improved noise robustness suitable for edge devices, and future work aims to further reduce model weight while maintaining denoising accuracy.

Abstract

Wearable electrocardiogram (ECG) measurement using dry electrodes has a problem with high-intensity noise distortion. Hence, a robust noise reduction method is required. However, overlapping frequency bands of ECG and noise make noise reduction difficult. Hence, it is necessary to provide a mechanism that changes the characteristics of the noise based on its intensity and type. This study proposes a convolutional neural network (CNN) model with an additional wavelet transform layer that extracts the specific frequency features in a clean ECG. Testing confirms that the proposed method effectively predicts accurate ECG behavior with reduced noise by accounting for all frequency domains. In an experiment, noisy signals in the signal-to-noise ratio (SNR) range of -10-10 are evaluated, demonstrating that the efficiency of the proposed method is higher when the SNR is small.

Wavelet Integrated Convolutional Neural Network for ECG Signal Denoising

TL;DR

This work addresses the challenge of denoising wearable ECG signals where noise overlaps with ECG frequency bands. It introduces a wavelet-integrated CNN that includes a Discrete Wavelet Transform (DWT) layer to separate high- and low-frequency components, enabling frequency-band-specific learning through trainable filters (using the Daubechies 6, db6 wavelet). Across MIT-BIH and MIT-NST datasets, the backward-type CNN with a single wavelet layer delivers the best denoising performance in RMSE and improvements over a conventional fully convolutional baseline, with qualitative results showing better preservation of ECG morphology. The approach demonstrates improved noise robustness suitable for edge devices, and future work aims to further reduce model weight while maintaining denoising accuracy.

Abstract

Wearable electrocardiogram (ECG) measurement using dry electrodes has a problem with high-intensity noise distortion. Hence, a robust noise reduction method is required. However, overlapping frequency bands of ECG and noise make noise reduction difficult. Hence, it is necessary to provide a mechanism that changes the characteristics of the noise based on its intensity and type. This study proposes a convolutional neural network (CNN) model with an additional wavelet transform layer that extracts the specific frequency features in a clean ECG. Testing confirms that the proposed method effectively predicts accurate ECG behavior with reduced noise by accounting for all frequency domains. In an experiment, noisy signals in the signal-to-noise ratio (SNR) range of -10-10 are evaluated, demonstrating that the efficiency of the proposed method is higher when the SNR is small.
Paper Structure (14 sections, 7 figures, 3 tables)

This paper contains 14 sections, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Method
  4. Wavelet Layer
  5. CNN model with wavelet layer
  6. Experimental Condition
  7. Dataset
  8. Model Parameters
  9. Experimental Results
  10. Quantitative Evaluation
  11. Qualitative Evaluation
  12. Discussion
  13. Conclusion
  14. Acknowledgments.

Figures (7)

  • Figure 1: The proposed convolutional neural network (CNN) model with an additional wavelet transform layer in which the features are separated into high and low components.
  • Figure 2: An example of power spectrum for an observed signal.
  • Figure 3: Summary of discrete wavelet transform. $D$ and $D^{\prime}$ indicate decomposed denoised component details. $D$ and $A$ are high- and low-frequency components, respectively.
  • Figure 4: Discrete Wavelet Transform (DWT) and Inverse Discrete Wavelet Transform (IDWT) layers.
  • Figure 5: Wavelet integrated convolutional neural network: (a) Forward(F)-type; (b) Backward(B)-type introduction of DWT and IDWT layers.
  • ...and 2 more figures