A novel ECG signal compression method using spindle convolutional auto-encoder

doi:10.1016/j.cmpb.2019.03.019

Computer Methods and Programs in Biomedicine

Volume 175, July 2019, Pages 139-150

https://doi.org/10.1016/j.cmpb.2019.03.019 Get rights and content

Highlights

•
A deep learning method based on convolutional auto-encoder is applied in ECG signal compression.
•
A novel spindle structure is designed to achieve a high-ratio and quality-guaranteed ECG signal compression.
•
An end-to-end model can directly compress input into the code without independent encoding method.
•
The model achieves high compression ratio 106.45 and low percentage root mean square difference of 8.00%.

Abstract

Background and objectives

With rapid development of telehealth system and cloud platform, traditional 12-ECG signals with high resolution generate heavy burdens in data storage and transmission. This problem is increasingly addressed with various ECG compression methods. The important objective of compression method is to achieve a high-ratio and quality guaranteed compression. Consequently, to achieve this objective, this work presents a deep-learning-based spindle convolutional auto-encoder. The spindle structure achieves the high-ratio compression by reducing the dimension and guarantees the quality by increasing the dimension and end-to-end framework.

Methods

The spindle convolutional auto-encoder provides a high-ratio and quality-guaranteed ECG compression. It is composed of two parts as convolutional encoder and convolutional decoder with functional layers. By convolutional operation, the local information can be extracted. The spindle structure is increasing dimension in first few layers to obtain sufficient information to guarantee compression quality. And it is reducing dimension in last few layers to merge the information into a code for high-ratio compression. Meanwhile, the end-to-end framework is to obtain the optimum encoding for compression to improve the reconstruction performance.

Results

Compression performance is validated with records from MIT-BIH database. The proposed method achieves high compression ratio of 106.45 and low percentage root mean square difference of 8.00%. Compared with basic convolutional auto-encoder, the spindle structure improves the compression quality with lower losses.

Conclusions

The spindle convolutional auto-encoder performs a high-ratio and quality-guaranteed compression. It can be considered as a promising compression technique used in tele-transmission and data storage.

Introduction

With the rapid development of telecommunication and cloud platform in various fields, the demands for efficient data transmission and storage greatly increase, particularly in the healthcare system. Electrocardiogram (ECG) is a bioelectrical signal reflecting the heart state according to the potential changes of heart [1]. An important objective of telehealth system is the capability to rapidly transmit ECG signals and efficiently store these information [2], [3]. However, the clinical ECG diagnosing always utilizes standard 12-lead ECG system to obtain heart information. A large number of ECG signals are collected over long-term monitoring and at high resolutions [4]. For example, under the sampling rate of 360 Hz and the data resolution of 11 bits per sample, a 1-h record amounts to about 1.7 Mbytes per lead. This generates heavy burdens in efficient data storage and rapid data transmission. Therefore, to overcome this problem, there has been increased interests in signal compression techniques these years. High compression ratio and minimal loss are important performance indicators of the effective compression method.

Over the past few decades, most efforts have focused on lossy compressions for higher compression ratio (CR). And there are a number of efficient lossy compression algorithms providing high-quality reconstructions. Generally, these methods can be categorized into three major classes: direct, parameter extraction and transform-based [5].

The direct methods are straight to discard the redundancy parts by analyzing the signal components. These methods usually take a low cost to realize the compression by an efficient encoding system. For instance, in [6], Cox et al. presented the amplitude zone time epoch coding (AZTEC) to provides a 20–1 reduction by using amplitude and slope to code the signals. To make a further improvement, in [7], the coordinate reduction time encoding system (CORTES) was designed. It introduced the turning-point algorithm to improve the reconstruction accuracy. Because direct methods are based on reserving important samples of the original signal. The compression ratio can be improved by discarding more samples of original signal. But the reconstruction is deteriorated due to the much information loss. Thus, it is difficult to achieve a high-quality compression with high compression ratio.

Second, parameter extraction methods generally extract the particular features in signals and encode these features to achieve the compression. In [8], Cohen et al. estimated structural features of ECG signals and these features were vector-quantized to provide the compression. In [9], Akazawaw et al. introduced multi-templates to achieve the signal matching and then the segmented signals can be compressed by Huffman coding. However, the extracted features decided the performance of compression, which arise the challenges to select suitable features. The compression performance will be degraded once the incorrect features extracted.

In recent years, transform-based methods begin to show their superiorities in high-quality compression. There are large number of researches arising. On account of the energy compaction property, transform-based methods usually decompose the original signal into coefficients associated with particular designed basis. Discarding the insignificant parameters, few parameters are encoded to achieve the signal compression. Fourier transform, wavelet transform, discrete cosine transformation (DCT), and singular value decomposition (SVD) [10] can be consider as the most prevalent transform-based compression methods. In [11], Ma et al. utilized adaptive Fourier transform decomposition to realize a signal analysis which is unconstrained by R-peak detection. Hybridized with the symbol substitution (SS), it further improved the compression performance. In [12], Peng et al. used lifting wavelet transformation to reserve the essential information in different frequency. Then Embedded Zerotree Wavelet (EZW) compression coding algorithm was introduced to achieve the compression. In [13], Mukhopadhyay et al. combined singular vector decomposition with lossless-ASCII-character-encoding (LLACE) to develop a quality-guaranteed compression algorithm. Despite their performance in data size reduction, the expansion basis is predetermined in general and the transform-based methods always combined with the independent encoding algorithm which would limit the compression performance.

To tackle aforementioned problems, this study aims to provide a high-ratio and quality-guaranteed ECG compression algorithm. Hence, a novel spindle convolutional auto-encoder (SCAE) is proposed. Convolutional auto-encoder is an attractive technique successfully used in many fields such as image super resolution, pattern recognition, and image reconstruction [14], [15]. There are a number of publications demonstrating that low-dimensional signals can produce a good quality restoration of original signal [16], [17], [18], [19]. Auto-encoder is an effective method in applications of low-dimensional representation and information retrieval [20]. Therefore, convolutional auto-encoder (CAE) can be considered as a promising method in ECG compression. Convolutional operation not only can reserve the local information, but also can also flexibly expand dimensionality for extracting high-level information and reduce dimensionality for high-ratio compression. So, a spindle convolutional auto-encoder is designed. Without parameters extraction, signal transformation and independent encoding algorithm, SCAE is an end-to-end model which directly compress the input as a low-dimensional code. The spindle structure is to expand dimensionality in first few layers and reduce dimensionality in last few layers to achieve the compression. In terms of the dimensionality expansion, the sufficient high-level features can be obtained in the hidden layers which can guarantee the quality of the signal reconstruction. As for dimensionality reduction, it can achieve to a high-ratio compression by merging these features. Compared to previous compression methods, this novel SCAE makes contributions in two aspects as:

(1)
SCAE as a deep learning technique, is an end-to-end compression method without parameter extraction, signal transformation and independent encoding algorithm.
(2)
Spindle structure can achieve a quality-guaranteed and high-ratio compression by expanding the dimensionality and reducing the dimensionality.

The remainder of this paper is organized as followings. Section 2 introduces the SCAE in detail. In Section 3 the performance evaluation criteria are listed out. Section 4 shows detailed compression performance. Discussion and comparison are presented in Section 5. Section 6 concludes this paper.

Section snippets

Methods

The proposed method can be divided into two parts as ECG signal preprocessing and spindle convolutional auto-encoder. As depicted in Fig. 1, an overall procedures of ECG compression and decompression using proposed method. Details of each part are introduced in the following contents.

Performance evaluation

For evaluation of compression performance, compression efficiency and reconstruction quality are two main aspects indicating the comprehensive performance. The evaluation criteria commonly used in the literature have been used for evaluating the compression efficiency and reconstruction quality of proposed method. These performance criteria are summarized as follows:

(1)
Compression ratio (CR)

With respect to compression efficiency, CR is widely accepted as an evaluation criterion that is defined as

Results

For the ECG compression, the training of the SCAE model on the dataset containing 48 records of MIT-BIH database was carried out. Each record randomly provides 1000 heartbeat segments used for training 200 heartbeat segments used for validation and 200 heartbeat segments used for testing. The sample ratio between training, validation and testing samples is 5:1:1. The network is trained for each record respectively. Due to all records have the similar loss graphs. Here as an example, the loss of

Discussion

To achieve an effective ECG compression, in this work, a novel spindle convolutional auto-encoder is designed. The first advantage of this work is the high-ratio and quality-guaranteed compression SCAE provided. By expanding dimensionality of SCAE, sufficient high-level features of the input can be obtained so that significant information can be merged into the compressed code by dimensionality reduction. The information capacity of the compressed code can be further enhanced to improve the

Conclusion

This paper proposed a high-ratio and quality-guaranteed ECG compression method using the novel designed SCAE. The spindle structure improves CAE to obtain the sufficient information of the original input signal by dimensionality expansion. It guarantees the compression quality. And dimensionality reduction further merges the information in the hidden layers to achieve the high-ratio compression. In addition, SCAE as an end-to-end model can directly compress the signal without parameter

Acknowledgments

This work was supported by the National Natural Science Foundation of China (61574102, 61774113 and 61331007), the Fundamental Research Funds for the Central Universities, Wuhan University (2042017gf0052 and 2042018gf0045) and the Natural Science Foundation of Hubei Province, China (2017CFB660).

References (37)

W. Einthoven
The different forms of the human electrocardiogram and their signification
Lancet
(1912)
S. Padhy
Multilead ECG data compression using SVD in multiresolution domain, Biomed
Signal Process Control
(2016)
S.K. Mukhopadhyay
An ECG compression algorithm with guaranteed reconstruction quality based on optimum truncation of singular values and ASCII character encoding
Biomed. Signal Process Control.
(2018)
B. Leng
3D object retrieval with stacked local convolutional autoencoder
Signal Process
(2015)
J. Schmidhuber
Deep learning in neural networks: an overview
Neural Netw
(2015)
B. Huang
ECG compression using the context modeling arithmetic coding with dynamic learning vector-scalar quantization
Biomed. Signal Process Control.
(2013)
O. Yildirim
An efficient compression of ECG signals using deep convolutional autoencoders
Cognit. Syst. Res.
(2018)
P. Barrett
Comparison of 24 Hour Holter monitoring versus 14 day novel adhesive patch electrocardiographic monitoring
Am. J. Med.
(2014)
M. Rosenberg
Use of a noninvasive continuous monitoring device in the management of atrial fibrillation: a pilot study
Pacing Clin. Electrophysiol.
(2013)
S.M.S. Jalaleddine
ECG data compression techniques—A unified approach
IEEE Trans. Biomed. Eng.
(1990)

J.R. Cox, AZTEC, a preprocessing program for real time ECG rhythm analysis, IEEE Trans. Biomed. Eng., 15(1968)...

J.P. Abenstein

A new data reduction algorithm for real time ECG analysis

IEEE Trans. Biomed. Eng.

(1982)

A. Cohen

Compression of ECG signals using vector quantization

K Akazawa

Adaptive data compression of ambulatory ECG using multi templates

An ECG signals compression method and its validation using NNs

IEEE Trans. Biomed. Eng.

(2008)

J. Ma

A novel ECG data compression method using adaptive Fourier decomposition with security guarantee in e-Health applications

J. Biomed. Health Informat.

(2015)

Peng Ziran

Research and improvement of ECG compression algorithm based on EZW

Comput. Meth. Prog. Bio.

(2017)

O.E. David

DeepPainter: painter classification using deep convolutional antoencoders

Int. Conf. Arti. Neural Netw.

(2016)

Cited by (44)

A new automated compression technique for 2D electrocardiogram signals using discrete wavelet transform
2024, Engineering Applications of Artificial Intelligence
The long-term electrocardiogram (ECG) signals are one of the crucial tools for the detection of severe heart diseases. However, a huge data is generated in its acquisition process, which affects the storage, transmission, and power efficiency of a system. Thus, there is a strong need for the development of new compression algorithms for effective ECG data management.
In this regard, an automated compression algorithm is proposed using 1D Cohen-Daubechies-Feauveau 9/7 wavelet transform and particle swarm optimization (PSO) for the compression of 2D ECG signals. It is applied to columns and rows of 2D ECG signals, which significantly increases sparsity in the transform domain and improves compression performance. The quantization of transform coefficients is performed using the optimized midtread quantizer. Its input parameters are also optimized using PSO to develop a more effective and user-independent compression algorithm. The quantized coefficients are then encoded by adaptive Huffman encoding to further improve the compression performance.
The experimental work is performed on MIT-BIH arrhythmia database and also compared with the existing compression techniques to validate the proposed method. Furthermore, different wavelets such as sym2, haar, db5, coif4, and beta wavelet are also used and compared with the proposed work. The performance of proposed algorithm is measured in terms of compression ratio, signal-to-noise ratio, percent root-mean-square difference, and quality score. The obtained results validate the efficacy of the proposed work and ensure the applications in telemedicine, mobile healthcare, and wearable monitoring devices for optimal utilization of storage, bandwidth, and power resources.
A novel asymmetrical autoencoder with a sparsifying discrete cosine Stockwell transform layer for gearbox sensor data compression
2024, Engineering Applications of Artificial Intelligence
The lack of an efficient compression model remains a challenge for the wireless transmission of gearbox data in non-contact gear fault diagnosis problems. In this paper, we present a signal-adaptive asymmetrical autoencoder with a transform domain layer to compress sensor signals. First, a new discrete cosine Stockwell transform (DCST) layer is introduced to replace linear layers in a multi-layer autoencoder. A trainable filter is implemented in the DCST domain by utilizing the multiplication property of the convolution. A trainable hard-thresholding layer is applied to reduce redundant data in the DCST layer to make the feature map sparse. In comparison to the linear layer, the DCST layer reduces the number of trainable parameters and improves the accuracy of data reconstruction. Second, training the autoencoder with a sparsifying DCST layer only requires a small number of datasets. The proposed method is superior to other autoencoder-based methods on the University of Connecticut (UoC) and Southeast University (SEU) gearbox datasets, as the average quality score is improved by 2.00% at the lowest and 32.35% at the highest with a limited number of training samples.
Electrocardiogram signal compression using adaptive tunable-Q wavelet transform and modified dead-zone quantizer
2023, ISA Transactions
The electrocardiogram (ECG) signals are commonly used to identify heart complications. These recordings generate large data that needed to be stored or transferred in telemedicine applications, which require more storage space and bandwidth. Therefore, a strong motivation is present to develop efficient compression algorithms for ECG signals. In the above context, this work proposes a novel compression algorithm using adaptive tunable-Q wavelet transform (TQWT) and modified dead-zone quantizer (DZQ). The parameters of TQWT and threshold values of DZQ are selected using the proposed Sparse-grey wolf optimization (Sparse-GWO) algorithm. The Sparse-GWO is proposed in this work to reduce the computation time of the original GWO. Moreover, it is also compared with some popular algorithms such as original GWO, particle swarm optimization (PSO), Hybrid PSOGWO, and Sparse-PSO. The DZQ has been utilized to perform thresholding and quantization. Then, run-length encoding (RLE) has been used to encode the quantized coefficients. The proposed work has been performed on the MIT-BIH arrhythmia database. Quality assessment performed on reconstructed signals ensure the minimal impact of compression on the morphology of reconstructed ECG signals. The compression performance of proposed algorithm is measured in terms of the following evaluation matrices: percent root-mean-square difference (PRD₁), compression ratio (CR), signal-to-noise ratio (SNR), and quality score (QS₁). The obtained average values are 3.21%, 20.56, 30.62 dB, and 7.79, respectively.
Learning an autoencoder to compress EEG signals via a neural network based approximation of DTW
2023, Procedia Computer Science
In this paper, we propose a method to compress electroencephalography (EEG) data using a differentiable neural network trained with the dynamic time warping (DTW) algorithm. DTW is a popular method for aligning time series data, but it is computationally expensive and not differentiable, which makes it difficult to optimize using gradient-based methods. To circumvent this difficulty, we use a neural network based approximation of DTW as the reconstruction loss to learn a convolutional auto-encoder to compress EEG signals. We show that using a custom DTW as reconstruction loss for the autoencoder improves the sleep staging score of the reconstructed data, suggesting that it allows to better preserve important features of EEG data. We also show that these findings can be extended to multiple datasets and compression ratios.
Using ECG signals for hypotensive episode prediction in trauma patients
2022, Computer Methods and Programs in Biomedicine
Citation Excerpt :
The results of this research show that the proposed method can learn better features than the traditional methods for the ECG arrhythmias classification. Wang et al. [34], achieved a high-ratio and quality compression of ECG signals using a convolutional AE. The encoder has 8 layers, convolution layers, max-pooling layers, and batch normalization layer.
Background and objectives:Bleeding is the leading cause of death among trauma patients both in military and civilian scenarios, and it is also the most common cause of preventable death. Identifying a casualty who suffers from an internal bleeding and may deteriorate rapidly and develop hemorrhagic shock and multiorgan failure is a profound challenge. Blood pressure and heart rate are the main vital signs used nowadays for the casualty clinical evaluation in the battlefield and in intensive care unit. However, these vitals tend to deteriorate at a relatively late stage, when the ability to prevent hazardous complications is limited. Identifying, treating, and rapidly evacuating such casualties might mitigate these complications. In this work, we try to improve a state-of-the-art method for early identification of Hypotensive Episode (HE), by adding electrocardiogram signals to several vital signs.
Methods:In this research, we propose to extend the state-of-the-art HE early detection method, In-Window Segmentation (InWise), by adding new types of features extracted from ECG signals. The new predictive features can be extracted from ECG signals both manually and automatically by a convolutional auto-encoder. In addition to InWise, we are trying to predict HE using a Transformer model. The Transformer is using the encoder output as an embedding of the ECG signal. The proposed approach is evaluated on trauma patients data from the MIMIC III database.
Results:We evaluated the InWise prediction algorithm using four different groups of features. The first feature group contains the 93 original features extracted from vital signs. The second group contains, in addition to the original features, 24 features extracted manually from ECG signal (117 features in total). The third group contains the original features and 20 ECG features extracted by the AE (113 features in total), and the last group is the union of all three previous groups containing 137 features. The results show that each model, which has used ECG data, is outperforming the original InWise model, in terms of AUC and sensitivity with $p$ -value $< 0.001$ (by 0.7% in AUC and up to 3.8% in sensitivity). The model which has used all three feature types (vital signs, manual ECG and AE ECG), outperforms the original model both in terms of accuracy and specificity with $p$ -value $< 0.001$ (by 0.3% and 0.4% respectively).
Conclusion:The results show an improvement in the prediction success rates as a result of using ECG-based features. The importance of ECG features was confirmed by the feature importance analysis.
A physiological signal compression approach using optimized Spindle Convolutional Auto-encoder in mHealth applications
2022, Biomedical Signal Processing and Control
Citation Excerpt :
These records were all sampled at 360 Hz with 11 bit/sample resolution. Signals were preprocessed for noise removal and centralisation, to achieve a higher compression ratio [4]. SCAElite compression model was applied over the dataset and the performance was evaluated in terms of compression ratio and reconstruction error so as to compare its performance with the state-of-the-art compression model SCAE [4].
The COVID-19 pandemic manifested the need of developing robust digital platforms for facilitating healthcare services such as consultancy, clinical therapies, real time remote monitoring, early diagnosis and future predictions. Innovations made using technologies such as Internet of Things (IoT), edge computing, cloud computing and artificial intelligence are helping address this crisis. The urge for remote monitoring, symptom analysis and early detection of diseases lead to tremendous increase in the deployment of wearable sensor devices. They facilitate seamless gathering of physiological data such as electrocardiogram (ECG) signals, respiration traces (RESP), galvanic skin response (GSR), pulse rate, body temperature, photoplethysmograms (PPG), oxygen saturation (SpO2) etc. For diagnosis and analysis purpose, the gathered data needs to be stored. Wearable devices operate on batteries and have a memory constraint. In mHealth application architectures, this gathered data is hence stored on cloud based servers. While transmitting data from wearable devices to cloud servers via edge devices, a lot of energy is consumed. This paper proposes a deep learning based compression model SCAElite that reduces the data volume, enabling energy efficient transmission.
Stress Recognition in Automobile Drivers dataset and MIT-BIH dataset from PhysioNet are used for validation of algorithm performance. The model achieves a compression ratio of up to 300 fold with reconstruction errors within 8% over the stress recognition dataset and 106.34-fold with reconstruction errors within 8% over the MIT-BIH dataset. The computational complexity of SCAElite is 51.65% less compared to state-of-the-art deep compressive model.
It is experimentally validated that SCAElite guarantees a high compression ratio with good quality restoration capabilities for physiological signal compression in mHealth applications. It has a compact architecture and is computationally more efficient compared to state-of-the-art deep compressive model.

View all citing articles on Scopus

View full text

A novel ECG signal compression method using spindle convolutional auto-encoder

Highlights

Abstract

Background and objectives

Methods

Results

Conclusions

Introduction

Section snippets

Methods

Performance evaluation

Results

Discussion

Conclusion

Acknowledgments

Lancet

Signal Process Control

Biomed. Signal Process Control.

Signal Process

Neural Netw

Biomed. Signal Process Control.

Cognit. Syst. Res.

Comparison of 24 Hour Holter monitoring versus 14 day novel adhesive patch electrocardiographic monitoring

Am. J. Med.

Use of a noninvasive continuous monitoring device in the management of atrial fibrillation: a pilot study

Pacing Clin. Electrophysiol.

ECG data compression techniques—A unified approach

IEEE Trans. Biomed. Eng.

A new data reduction algorithm for real time ECG analysis

IEEE Trans. Biomed. Eng.

Compression of ECG signals using vector quantization

Adaptive data compression of ambulatory ECG using multi templates

An ECG signals compression method and its validation using NNs

IEEE Trans. Biomed. Eng.

A novel ECG data compression method using adaptive Fourier decomposition with security guarantee in e-Health applications

J. Biomed. Health Informat.

Research and improvement of ECG compression algorithm based on EZW

Comput. Meth. Prog. Bio.

DeepPainter: painter classification using deep convolutional antoencoders

Int. Conf. Arti. Neural Netw.