A tool for the real-time evaluation of ECG signal quality and activity: Application to submaximal treadmill test in horses

https://doi.org/10.1016/j.bspc.2019.101666Get rights and content

Highlights

  • A new electrocardiogram quality index, i.e. higher-order-statistics-SQI, is defined.

  • Our index allows the real-time evaluation of electrocardiogram recording quality.

  • The results outperformed four existing indexes, also during intense exercise.

Abstract

This work describes a novel signal quality index (SQI), i.e. higher-order-statistics-SQI (hosSQI), for the real-time evaluation of electrocardiogram (ECG) recording quality. The hosSQI formula combines two already known SQIs, kurtosis (kSQI) and skewness (sSQI), exploiting the related properties to improve their performance.

We validated hosSQI using 1000 human pre-labelled twelve-lead ECGs and compared its performance with the state-of-the-art indexes in the literature. Our index outperformed four existing indexes (kSQI, sSQI, basSQI, iorSQI), reaching an accuracy up to 90.38% in the signal quality discrimination. Afterwards, we employed these indexes to compare signal quality of ECGs acquired by two different monitoring systems (red-dot and textile electrode based), in unfavourable conditions in terms of motion artifacts, adhesion and mechanical firmness of electrodes. The existing four SQIs and hosSQI were updated each second, using equine ECGs recorded during submaximal treadmill test. Wilcoxon nonparametric statistical test showed that all the SQIs were significantly higher for textile than for red-dot electrodes. A pattern recognition algorithm was implemented to test a real-time discrimination of three activity conditions (walk, trot, and gallop) based on the SQIs. Given that hosSQI values computed for red-dot were under the acceptability threshold in more than 63% of signals, we used only the textile data. We employed a C-Support Vector Classification and we found the highest accuracy value in the discrimination of walk and gallop (84.91%). Even if these results are preliminary, we proposed a promising tool for the real-time assessment of ECG signal quality and physical activity, also during intense exercise.

Introduction

A real-time evaluation of physiological signal quality, e.g., electrocardiogram (ECG), is essential at the present time, in which wearable sensors allow continuous monitoring during daily activities. Detection of low signal quality in real-time would avoid losing data and time in acquiring unusable signals. Especially in clinical monitoring, this practice would lead to greater confidence for both patients and doctors in using telemedicine devices. The use of these wireless ECG acquisition systems has recently also extended to the physiological monitoring of the animals, in particular horses involved in competitions. In this context, the acquisition of good quality signals is compromised not only by the lack of awareness of the subject, but also by the amount of motion artifacts recorded during the strong physical activity to which the horses are subjected. Horses have always been involved in both professional or pleasure riding activities, but their speed and sheer size made the monitoring of physiological signals more difficult with respect to other species. The introduction of treadmill exercise test allowed assessing horse's response, in terms of autonomic nervous system (ANS), at different levels of physical activity, in controlled conditions (e.g., velocity, running surface, environment) [1]. This test improves the clinical assessment of horse's physical conditions, especially when the animal is not able to reach the desired performance level in conditions of high exercise workload. As a matter of fact, current research findings showed how much the functional evaluation of poor performance is of strong importance during exercise practice [2], [3], [4].

A gradual submaximal treadmill test can be also performed in order to make an efficient heart disease diagnosis. In fact, moderate or severe heart pathologies can be diagnosed in standing horses, but in case of subtle rhythm disturbances or exercise-induced problems the suggested way consists in exercise testing [5]. In some clinical cases, cardiac dysrhythmias are difficult to be detected, because even if they could be present during or after a physical exercise, they are not associated to the presence of a severe cardiac disease [6]. For this purpose and considering the athletic target of horses, between standardized exercise testing, the treadmill exercise test has become an important tool for cardiovascular diagnosis [7], [8]. However, it requires a continuous monitoring of cardiac signals, which is crucial to detect pathologies that may degrade performance [9] and/or because of unexpected collapse or death [3].

As it can be easily expected, due to the high amount of motion artifacts during hard exercise, the ECG signal becomes hardly to be acquired with a sufficient quality for further analysis. Specifically in such conditions, it is difficult to preserve the adherence between the electrodes and the horse's skin. In the last decades, wearable monitoring systems have been continuously improved in terms of stability, subject comfort, and robustness against motion artifacts [10]. The innovative smart textiles for physiological signal acquisition combine conductive yarn (made of stainless steel fibers) with elastane [11]. These systems have been successfully applied to human monitoring during hospitalization and every-day life [12], [13]. In order to address issues regarding horse performances, McGreevy et al. envisioned the use of smart textiles in equestrian contexts, offering a simple link between horse trainer practise and exercise physiology of horses [14].

We here propose a novel real-time signal quality index (SQI) and an evaluation performance comparison with the most used quality indexes for ECG in real-time. Two different acquisition systems, i.e. red-dot and textile electrodes, have been simultaneously applied to five horses during gradual treadmill test [15]. The experimental protocol consisted in four different sessions with increasing velocity, from walk to gallop. Since we have already demonstrated that textile electrodes are more reliable and robust than the Ag-AgCl ones, using ECGs acquired from horses at rest [10], [16], here we investigate in real-time if the textiles remain more accurate in ECG signal acquisition also when motion artifact is supposed to be as high as possible (e.g., during gallop). Of note, considering velocity and physical efforts reached by horses in the experiment, a reliable real-time signal quality estimation is essential to identify the limit conditions in which they can be monitored, avoiding to achieve erroneous results and acquire corrupted signals. A real-time index leads to achieve good ECG recordings in tele-health applications, making the operators able to stop the acquisition when the signal quality is insufficient to provide the expected information about the physical condition of the horse.

Therefore, we propose a novel index, the higher-order-statistics-SQI (hosSQI), which uses the third and fourth statistical moments of the ECG. This index is a synthesis of information coming from kurtosis SQI (kSQI) and skewness SQI (sSQI), allowing to categorize the ECG quality into three classes (or levels), which are derived from the statistical properties of the signal: good (G), acceptable (A), or unacceptable (U). Moreover, we evaluated whether the new index provided more information than other existing SQIs, through validation on a collection of 1000 human twelve-lead ECGs, gathered from PhysioNet/CINC 2011 Challenge [17], [18]. Revising the plethora of developed ECG SQIs [19], [20], [21], [22], [23], [24], [25], we selected four SQIs to be included in the comparison, considering their reliability in ultra-short series analysis for real-time applications. Obviously, we compared our hosSQI with the two SQIs from which it derives: kSQI and sSQI [26], [27], [28], [29], [30], [31]. Then, we selected other two indexes based on the study of ECG power spectral density (PSD), specifically the relative power of the ECG in the baseline (basSQI) and the in-band to out-of-band spectral power ratio within the QRS complex (iorSQI) [22], [26], [32]. According to the literature, these indexes are particularly sensitive to activity-related artifacts and were already employed to compare different ECG acquisition systems [33], [32].

Afterwards, we computed statistical analysis between the values of the five SQIs extracted from equine ECGs for investigating the presence of significant differences between the two acquisition systems in the different phases of the experimental protocol. Furthermore, a pattern recognition algorithm has been applied to the best acquisition system for discriminating the three physical activity conditions during the treadmill test (walk, trot, and gallop), using the real-time quality indexes. Specifically, the pattern recognition algorithm consisted a Leave-One-Subject-Out (LOSO) procedure, based on a support vector machine (SVM), specifically the type called C-SVC. A feature selection based on the study of the Jensen-Shannon (JS) divergence was also performed in order to detect the most discriminant SQIs.

Section snippets

ECG signal quality indexes (SQI)

In this study we implemented four quality indexes already used in the literature [26], [22], [27]:

  • relative power of the ECG in the baseline (basSQI)

  • in-band to out-of-band spectral power ratio within the QRS complex (iorSQI)

  • kurtosis signal-quality-index (kSQI)

  • skewness signal-quality-index (sSQI) and the novel SQI, based on kSQI and sSQI:

  • higher-order-statistics-signal-quality-index (hosSQI)

The basSQI metric quantifies the ratio between the power in the band [1, 40] Hz and in the band [0, 40] Hz,

Accuracy of hosSQI in the discrimination of acceptable/unacceptable human ECG signals

The average of the values found in the 12 leads was computed for each index, in order to provide an overall SQI for each ECG and to compare it to the label (acceptable/unacceptable) assigned in the Physionet repository. Therefore we obtained for each ECG signal five SQIs, and we used these indexes as input of the C-SVC [39], in a Leave-One-Out (LOO) procedure. In Fig. 1 we show the percentage values of accuracy reached in the recognition of human ECG signal labels, calculated as the ratio

Discussion

We proposed a novel tool for real-time monitoring of physiological signals and physical activity of horses during hard exercise, based on a novel ECG quality index. A good quality of equine ECG monitoring may improve the diagnosis of peculiar cardiac disturbances, which are evident only during hard physical activity, and evaluate the athletic performances. Furthermore, ECG signal is fundamental for estimating Heart Rate Variability (HRV) during exercise, which can provide several indexes of ANS

Conflicts of interest

The authors declare no conflicts of interest.

References (45)

  • A. Fraipont et al.

    Subclinical diseases underlying poor performance in endurance horses: diagnostic methods and predictive tests

    Veterinary Record-Engl. Ed.

    (2011)
  • N. Ryan et al.

    Survey of cardiac arrhythmias during submaximal and maximal exercise in thoroughbred racehorses

    Equine Vet. J.

    (2005)
  • H. Ohmura et al.

    Risk factors for atrial fibrillation during racing in slow-finishing horses

    J. Am. Vet. Med. Assoc.

    (2003)
  • R. Buhl et al.

    Cardiac arrhythmias in clinically healthy showjumping horses

    Equine Vet. J.

    (2010)
  • K. Kiryu et al.

    Pathologic and electrocardiographic findings in sudden cardiac death in racehorses

    J. Vet. Med. Sci.

    (1999)
  • A. Lymberis et al.

    Smart fabrics and interactive textile enabling wearable personal applications: R&d state of the art and future challenges

  • A. Lanata et al.

    Complexity index from a personalized wearable monitoring system for assessing remission in mental health

    IEEE J. Biomed. Health Inform.

    (2015)
  • M. Nardelli et al.

    Heartbeat complexity modulation in bipolar disorder during daytime and nighttime

    Sci. Rep.

    (2017)
  • M. Nardelli et al.

    Real-time evaluation of ecg acquisition systems through signal quality assessment in horses during submaximal treadmill test

  • A. Lanata et al.

    A novel algorithm for movement artifact removal in ecg signals acquired from wearable systems applied to horses

    PLOS ONE

    (2015)
  • T.H.C. Tat et al.

    Physionet challenge 2011: improving the quality of electrocardiography data collected using real time qrs-complex and t-wave detection

  • I. Silva et al.

    Improving the quality of ecgs collected using mobile phones: the physionet/computing in cardiology challenge 2011

    Comput. Cardiol.

    (2011)
  • Cited by (18)

    • An automatic arrhythmia classification model based on improved Marine Predators Algorithm and Convolutions Neural Networks

      2022, Expert Systems with Applications
      Citation Excerpt :

      Higher than Order statistics (HOS) have been widely used to obtain characteristics for effective classification (Marinho et al., 2019). Moreover, kurtosis is used to estimate the deviation of a distribution from a Gaussian distribution (Nardelli et al., 2020), which is the basis of HOS. However, kurtosis is associated with a high-frequency QRS transition.

    View all citing articles on Scopus
    View full text