Instrumentation & Measurement Magazine 25-5 - 41

Deep Learning-Based Signal Quality
Assessment for Wearable ECGs
Xiangyu Zhang, Jianqing Li, Zhipeng Cai, Lina Zhao, and Chengyu Liu
N
owadays, use of the dynamic electrocardiogram
(ECG) has developed rapidly because of the wide
application of wearable devices [1]-[3]. Most ECGbased
diagnostic algorithms require that the ECG signal have a
clear waveform and accurate feature points. However, the collected
wearable ECG signal usually contains a certain amount of
noise and causes many false alarms in the ECG analysis system
[4], [5]. Thus, signal quality assessment (SQA) plays a prominent
role in ruling out the ECG segments with poor signal quality [6].
Compared with traditional static ECG signals, dynamic wearable
ECGs contain more noise, which brings greater challenges
to disease detection algorithms [7]-[9]. These artifacts and
noises in dynamic ECG signals can seriously affect the R-peaks
detection, ECG beat extraction, ECG morphological feature extraction
and the detection of noise peaks, resulting in frequent
false alarms [10]. In 2008, Li et al. [11] proposed the bSQI signal
quality indexes: comparison of two beat detectors on a single
ECG lead. Liu et al. [12] generalized the two QRS wave complex
(QRS) detectors-based bSQI to multiple QRS detectors-based
bSQI (GbSQI) to improve the SQA performance. Liu et al. [8] proposed
an efficient real-time SQA method for healthy subjects.
However, special arrhythmia can lead to significant
changes in the waveform of ECG segments [13], which brings
greater challenges to the signal quality evaluation algorithm. If
noisy ECG segments can be accurately picked out and the ECG
segments with abnormal beats can be retained, the accuracy of
abnormal heartbeat monitoring algorithms can be greatly improved.
Satija [10] proved that a quality-aware ECG analysis
system is most essential to ensure the diagnosis accuracy and
reliability of ECG with different arrhythmias types. Therefore,
accurate signal quality assessment algorithms for dynamic
ECGs, especially for arrhythmia ECGs, is of great significance
in the diagnosis of arrhythmia ECG and worthy of further research
and development.
Existing Signal Quality Classification
Methods
Researchers have conducted a lot of research on quality evaluation
methods of ECG signals. With the development of
August 2022
computer analysis and processing capabilities, the ECG signal
quality assessment algorithms can be roughly divided into
three categories: Rule-based SQA methods, Machine learningbased
SQA methods, and Deep learning-based SQA methods.
Early designed signal algorithms were limited in processing
capacity. Thus, low-complexity algorithms were usually
designed with a rule-based SQA method combined with
waveform and interval features [9], [14]. However, some methods
were developed based on the morphological features of
QRS complex [15], [16], and interval-based features [17], [18]
may misjudge the ECG segment with arrhythmia. The QRS
feature-based SQA methods demand a reliable QRS complex
detector which is still one challenging task under dynamic
recording environments [1]. The arrhythmia heartbeats can
bring changes in the waveform of QRS complex and also affect
the QRS features. Moreover, R-R interval feature-based SQA
methods may only work for ECG segments with normal heartbeats,
because of the R-R interval changes in the ECG segments
with arrhythmia.
Machine learning-based signal quality classification methods
are usually divided into three steps: signal preprocessing,
feature extraction and feature classification [1], [19]. During
signal preprocessing, researchers usually remove the baseline
drift and high-frequency noise in the ECG segments through
bandpass filter-based methods and other transform-based
methods such as wavelet transform [20], Fourier transform
[21] and so on. Chen et al. [20] used a sub-band dependent
threshold wavelet-based denoising technique to reduce various
noises. Chandrakar et al. [21] used Fast Fourier transform
to reduce the low frequency noise in the ECG segments. In
addition, some researchers also detected and removed macroscopic
error signal such as missing lead, device saturation, etc.
Instrumentation & Measurement Magazine 25-5

Table of Contents for the Digital Edition of Instrumentation & Measurement Magazine 25-5
