A comprehensive review on encoder–decoder architectures in ECG signal compression and denoising: opportunities, challenges, and prospects

Purpose Cardiovascular disease diagnosis relies on the accurate interpretation of electrocardiogram (ECG) signals, which are sensitive to noise and artifacts. Furthermore, with ubiquitous long-term monitoring technology, portable, lightweight wearable devices have limited memory and energy resources...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Research on biomedical engineering Jg. 41; H. 4; S. 57
Hauptverfasser: Das, Maumita, Sahana, Bikash Chandra
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Cham Springer International Publishing 01.12.2025
Springer Nature B.V
Schlagworte:
Algorithms
Artificial neural networks
Biomaterials
Biomedical Engineering and Bioengineering
Biomedical Engineering/Biotechnology
Cardiovascular diseases
Coders
Comparative analysis
Comparative studies
Compression
Deep learning
EKG
Electrocardiography
Energy resources
Energy sources
Engineering
Flexibility
Fourier transforms
Generative adversarial networks
Heart
Mathematical models
Memory devices
Neural networks
Noise
Noise reduction
Noise sensitivity
Peripheral equipment (computers)
Portable equipment
Real time
Review
Signal processing
Signal transmission
Surveys
Telemetry
Wavelet transforms
Wearable technology
Deep learning models
Detection
Autoencoder
Data preprocessing
Electrocardiogram
Denoising
ISSN:2446-4732, 2446-4740
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Purpose Cardiovascular disease diagnosis relies on the accurate interpretation of electrocardiogram (ECG) signals, which are sensitive to noise and artifacts. Furthermore, with ubiquitous long-term monitoring technology, portable, lightweight wearable devices have limited memory and energy resources, as well as high signal transmission costs. Therefore, the signal must be denoised and compressed before it is transmitted to the clinic. For effective real-time ECG signal transmission via wearables or telemetry systems, the denoising autoencoder (DAE) architecture is quite popular. This article examines the utilization of learning-based DAE techniques in ECG signal denoising and compression. It also emphasizes the potential of these algorithms to mitigate various ECG noises while minimizing overall system complexity and computation time. This survey aims to identify the most effective strategies based on comparative studies on the state-of-the-art encoder–decoder architectures for ECG signal compression and denoising. Methods Numerous researchers have developed a variety of efficient DAE algorithms for portable, lightweight peripheral devices. Some of the most frequently used DAE architectures are the convolutional denoising autoencoder (CDAE), variational denoising autoencoder, stacked denoising autoencoder, adversarial denoising autoencoder, and hybrid DAE. The operating principle, advantages, and disadvantages of each architecture are explained in this article. Results It has been observed that the powerful nonlinear mapping capabilities of the DAEs allow them to provide a high compression rate together with improved filtering capabilities. The comparative analysis of the recent research indicates that approximately 65% of the literature uses CDAE-based ECG denoising architecture. The inclusion of skip connections and attention modules in CDAE can improve its performance by reducing training time and enhancing denoising ability. Additionally, the conditional generative adversarial network and adversarial denoising convolutional neural network show comparatively higher SNR with lower RMSE for denoising the ECG signal. However, the U-Net DAE architecture exhibits better denoising performance with fewer parameter requirements compared to other DAE techniques. Moreover, it provides a low-power, lightweight architecture that is simple to implement on mobile devices. Furthermore, the hybrid DAE model is a remarkably efficient and high-speed option, making it especially suitable for edge devices. Conclusion This paper surveys various research on ECG signal compression and noise reduction systems. It discusses the step-by-step framework of DAE architectures and presents a comparative study on their performance to suppress different ECG noises and artifacts. This study also covers corresponding challenges, limitations, and future directions.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2446-4732
2446-4740
DOI:10.1007/s42600-025-00436-0