홈으로ArticlesAll Issue
ArticlesA Real-Time Abnormal Beat Detection Method Using a Template Cluster for the ECG Diagnosis of IoT Devices
  • Seungmin Lee1 and Daejin Park2,3,*

Human-centric Computing and Information Sciences volume 11, Article number: 04 (2021)
Cite this article 7 Accesses
https://doi.org/10.22967/HCIS.2021.11.004

Abstract

Currently, the use of the Internet of Things (IoT) devices is expanding, and research on bio-signal monitoring systems is increasing. This paper proposes an abnormal beat detection algorithm for electrocardiogram signals that is suitable for embedded devices. A typical single template-based detection method requires a great deal of memory to generate a template, and abnormal beats make it difficult to generate a normal beat template. As such, this paper proposes a reliable method of generating a normal beat template using a template cluster with Pearson similarity. The proposed method uses the weighted mean to minimize memory usage in the template cluster generation step. The results of the experiment indicate that the proposed algorithm can measure P-wave deformation shapes, which are difficult to detect, using the partial template in the P-wave region. Furthermore, the average detection time is 0.39 seconds for a 30-minute signal, confirming the algorithm’s potential for real-time operation in lightweight embedded devices.


Keywords

ECG, Template, Abnormal Beat Detection, PVC, PJC, Embedded System


Introduction

Interest in, and research on, health management using the u-healthcare system are expanding because the mortality rate caused by heart disease is increasing in parallel with the increase in life expectancy. Recently, lightweight embedded equipment becomes an effective system component due to the development of software and hardware technology. In particular, research on the measurement, transmission, and analysis of bio-signals using an embedded device is ongoing [13]. Research on embedded devices is applied not only to healthcare, but also to the smart home, elderly care, security, and diverse human activities among other things [49].
Electrocardiogram (ECG) signals are representative bio-signals that measure the heart’s electrical activity and are widely used in unique bio-signals for personal authentication, as well as in arrhythmia diagnosis to prevent heart disease [10, 11]. In addition, studies are being conducted on the measurement and analysis of ECG signals in daily life using embedded devices [1215].
Conventional ECG signal analysis using a Holter monitor is shown in Fig. 1. Long periods of measurement are essential with ECG signals because the abnormal beats which it is necessary to detect in order to diagnose arrhythmia only occur rarely. Therefore, this process requires a power supply and a recording device, such as the Holter monitor shown in Fig. 1, which is problematic because it is inconvenient to use in daily life. Recently, the development of wireless technology has made it possible to measure ECG signals simply by using a smartphone or smartwatch. Fig. 2 shows an embedded device for measuring ECG signals using the Bluetooth function of a smartphone.

Fig. 1. Holter monitor.
Fig. 2. A Bluetooth-based device used to obtain an ECG signal.

The amount of data available for the analysis of ECG signals is growing exponentially because public interest in personal health is increasing and patients can now measure such signals themselves. In general, normal beats, which account for the vast majority of ECG signals, are not necessary for arrhythmia diagnosis because similar beats are repeated periodically. Therefore, the time and cost of arrhythmia diagnosis can be greatly reduced if the abnormal beats required for diagnosis are detected. The detection of an abnormal heartbeat is largely divided into feature-based detection and shape-based detection.
In feature-based detection, the waveform’s fiducial point is detected [1622], and the feature values are acquired and analyzed in order to determine whether the waveform is distorted. Various researches have used the acquired feature values to conduct studies on knowledge-based classification, machine learning-based classification, and deep learning-based classification. However, erroneous detection can occur when the exact fiducial point is not detected because the feature values are based on detection of the fiducial points. In addition, interference, such as power noise or baseline movements, makes it even more difficult to detect the fiducial point.
Shape-based detection is a method of classifying normal and abnormal heartbeats using errors or similarities with a template [2326]. In shape-based detection, individual templates are required because normal beats vary depending on the individual’s ECG signal. In addition, if a noise change, such as a baseline movement or amplitude variation, occurs even within a normal heartbeat, the normal beats are distorted, resulting in a false detection. Thus, the existing template is determined manually, or a template is created using statistical values such as the average or median values [27].
However, manual decision-making is problematic in that the results vary depending on individual learning levels and conditions. In addition, large errors can occur when abnormal heartbeats are included in the average calculation process. A relatively strong median value is problematic in that the amount of calculation required for sorting is large, and the heartbeat must be monitored and recorded for a long time. In general, the square sum difference (SSD), which is used to measure the similarities between two signals, has significant weak points in the case of an ECG signal, such as baseline movements and changes in scale of amplitude [28].
This study aimed to improve shape-based detection performance using an existing template. First, Pearson similarity, which is suitable for ECG signals, is proposed. Pearson similarity is robust to baseline movement and changes in amplitude scale because it normalizes the signal using the mean and standard deviation and measures similarity using the two signals’ correlation coefficient.
In addition, this paper proposes a method of generating a template cluster instead of a single template. Using the Pearson similarity, the template is updated when the input signal is similar to the template, thereby generating a reliable template. In addition, during the template updating process, the weighted mean used minimizes errors that occur when an abnormal heartbeat updates the template. In particular, using a counter to represent the number of updates as a weight minimizes memory usage, even in low-capacity embedded devices, thus enabling real-time operation. In addition, based on the characteristics of the ECG signals, in which normal beats occur most frequently, the template with the highest counter value is determined as the representative template of the normal heartbeat. Then, the abnormal beat template, which is dissimilar to this representative template, is removed and a reliable normal beat template cluster is generated.
For the purposes of this paper, a template for the partial signal was used for the P-wave region to generate a template cluster in order to detect premature junctional contractions (PJCs), including P-wave deformations, as well as premature ventricular contractions (PVCs), including QRS complex deformation.
Fig. 3 presents a summary of the proposed algorithm system.
Fig. 3. A summary of the proposed algorithm system.

This paper is composed as follows: Section 2 briefly introduces the composition of ECG signals, abnormal beats, and the preprocessing used in this paper; Section 3 introduces the existing similarity and template decision methods, and briefly describes the problems with these methods; Section 4 describes the proposed method for improving on the existing methods; Section 5 confirms the proposed method’s superiority through experiments; and, finally, Section 6 presents the conclusion.


ECG Signal

The Composition of ECG Signals
ECG signals are the electrical signals that are generated when the atrium and ventricles are depolarized and repolarized [29] successively. The P-wave indicates atrial depolarization, the QRS complex indicates atrial repolarization and ventricular depolarization, and the T-wave indicates ventricular repolarization [30, 31]. Fiducial points composed of an onset (starting point), an offset (ending point), and a peak represent each waveform. Various feature values are obtained from the differences in interval, segment and amplitude between these fiducial points. Arrhythmia is diagnosed by analyzing the acquired feature values to detect an abnormal heartbeat. Fig. 4 shows an ECG signal’s waveform and fiducial point, as well as the characteristic values that are typically used.
Fig. 4. Fiducial points and feature values of ECG signals.

Abnormal Heartbeat of an ECG Signal
Arrhythmia appears in the ECG signal as abnormal beats. Such abnormal beats include distortions in the waveform’s shape or period. In general, abnormal beats occur very rarely, making long-term measurement essential for an accurate diagnosis of arrhythmia. Therefore, the detection of an abnormal heartbeat can greatly reduce the time and cost required by a cardiologist to diagnose arrhythmia. There are various types of abnormal heartbeats; the Association for the Advancement of Medical Instrumentation’s ANSI/AAMI EC57:1998 [32] classified abnormal beats into the following five categories according to their location: normal beat (N), upper ventricular arrhythmia beat (S), ventricular arrhythmia beat (V), mixed arrhythmia beat (F), and unclassified beat (Q).
PhysioNet provides a variety of arrhythmia databases, and categorizes beats using 19 categories. Table 1 shows the distribution of beats occurring in 19 beat classifications and 48 measurements taken from the MIT-BIH arrhythmia database (MIT-BIH ADB; https://www.physionet.org/content/mitdb/1.0.0/) dataset used in this paper [33].

Table 1. Classification of beat and distribution in MIT-BIH ADB
Code Total number Description Class
N 75,052 Normal beat Normal
L 8,075 Left bundle branch block beat Bundle branch block
R 7,259 Right bundle branch block beat
B 0 Bundle branch block beat
A 2,546 Atrial premature beat Premature
a 150 Aberrated atrial premature beat
J 83 Junctional premature beat
S 2 Supraventricular premature beat
V 7,130 Premature ventricular contraction
r 0 R-on-T premature ventricular contraction
F 803 Fusion of ventricular and normal beat
e 16 Atrial escape beat Escape
j 229 Junctional escape beat
n 0 Supraventricular escape beat
E 106 Ventricular escape beat
/ 7,028 Paced beat Paced
f 982 Fusion of paced and normal beat
Q 33 Unclassifiable beat Else
? 0 Beat not classified during learning
Fig. 5. Example of abnormal beat: (a) PJC and (b) PVC.
The left- and right-bundle branch blocks (L, R) and the pacemaker beat (/) replace the normal beat. This paper defines them as normal beats and focuses on the detection of PJCs (J), including P-wave deformations, and PVCs (V), including QRS complex deformations, among abnormal beats. Fig. 5 presents examples of abnormal PJC and PVC beats.
In the PJC shown in Fig. 5(a), the P-wave appears as a downward wave. In the PVC shown in Fig. 5(b), the QRS complex shape is deformed. To detect an abnormal beat in which the P-wave and QRS complex deformation have occurred, this paper proposes a shape-based method of detection using a normal beat’s template.

Preprocessing
Generally, various types of pre-processing are applied to achieve effective ECG signal analysis. In this paper, the preprocessing is divided into noise suppression, R-peak detection, and beat separation.

Noise reduction Various noises occur during the measurement of ECG signals. The most typical types of noise are as follows [3437]:
Power line interference: This high-frequency noise varies by country.
Baseline movement: This low-frequency noise (0.15 up to 0.3 Hz) results from the patient’s breathing and leads to a baseline shift in the signals.
Miscellaneous noise: This includes electrode contract noise, electrode motion artifacts, muscle contractions, electrosurgical noise, instrumentation noise, and so on.
Types 1 and 2 are typical high- and low-frequency noise in ECG signals. High-frequency noise blurs the boundary between the baseline and the waveform, making it difficult to detect the fiducial points. Thus, in fiducial point detection algorithms, the use of a low-pass filter suppresses high frequency noise, whereas the use of a notch filter only suppresses power noise [38]. Low-frequency noise distorts the signal’s shape, greatly reducing the accuracy of similarity measurement. However, this can be suppressed using a high-pass filter.
In this paper, to reliably measure the similarity between signals, a band pass filter was used to suppress high- and low-frequency noises. In consideration of the occurrence of 30 Hz power noise, a 1–25 Hz Butterworth band-pass filter was used.
Fig. 6 shows a comparison between a signal containing noise and a filtered signal.

Fig. 6. Signal with noise and filtered signal: (a) baseline movement filter and (b) noise suppression.

R-peak detection Among the fiducial points of an ECG signal, the R-peak of the QRS complex has the highest amplitude. The R-peak detection results are used to measure beats per minute (bpm), because detecting R-peak is easy and accurate [39]. In addition, the R-peak separates beats, and can be used to detect other fiducial points. Pan’s method [40] is the most representative method of R-peak detection, and can detect the R-peak in real time by obtaining various auxiliary signals based on signal differentiation and by using adaptive threshold determination. Fig. 7 shows auxiliary signals with adaptive thresholds and the results of R-peak detection using Pan’s method.
Fig. 7. Example of Pan’s method [40]: (a) input signal, (b) derivative signal, (c) squared signal, and (d) mean filtered signal, with adaptive threshold and the results of R-peak detection.
Fig. 8. Example of extracting beat centered at the R-peak.


Beat separation Beats are separated based on the results of R-peak detection. The region of 275 ms before and 375 ms after R-peak includes the P-wave, QRS complex, and T-wave, which are mainly used in the analysis of ECG signals [24]. Thus, this region is generally used to extract fiducial points based on R-peak. Fig. 8 shows an example of an individual beat in a signal.
In this paper, a template for the extracted region was obtained, and its similarity with the template was compared in order to detect any abnormal beats. In addition, the P-wave’s region of interest was determined as a region from the beginning to 100 ms before the R-peak as shown in Fig. 8, and the P-wave region’s template was used to detect an abnormal beat corresponding to the PJC.


Existence Method

Square Sum Difference
SSD is one of the techniques most commonly used to compare the similarity of two signals (X,Y), as shown in (1).

(1)

where N represents the signal length.
Normal and abnormal beats are classified by measuring the similarity between a template and an input signal using SSD. SSD is suitable for real-time measurement of the similarity between two signals in lightweight embedded devices due to its low amount of computation and simple algorithm. However, SSD can generate large errors when the baseline movement distorts the signal. In addition, an error occurs when the scale of amplitude changes due to the patient’s respiration. Fig. 9 shows an example of the measurement errors that can occur when using SSD for two signals in the presence of baseline movement or changes in the scale of amplitude.
Fig. 9. Example of a problem with measuring SSD similarity.


Single-Template Determination
It is difficult to determine the template when attempting to detect an abnormal heartbeat. One of the methods of detecting an abnormal heartbeat consists in recording the representative shape of abnormal beats and measuring their similarities. Fig. 10 represents a cluster of abnormal beats [27].
However, the shape of normal beats varies depending on the individual. In particular, problems can occur when a normal beat is similar to the template of an abnormal beat. Furthermore, if an abnormal heartbeat goes unrecorded, it could be classified as a normal beat.
Fig. 10. Template cluster of abnormal heartbeats.


Accordingly, research is currently being conducted on a technique for determining a normal-beat template rather than an abnormal beat template. In this regard, it is necessary to adaptively determine a normal beat because its shape varies depending on the individual. A representative method determines the template using the average of heartbeats. Since abnormal heartbeats occur rarely, a template similar to be normal heartbeat constituting most of the heartbeat can be obtained through the average. However, a stable, normal beat template cannot be obtained due to the large signal distortion when the similarity between abnormal and normal beats is very low. Fig. 11 shows the template obtained by using the average when including multiple normal and abnormal heartbeats with large QRS complex distortions due to PVC.
As shown in Fig. 11 above, the difference between normal heartbeats and the template is considerable. To improve upon this, a method that is robust to outliers can be employed by utilizing medians. The median can be used to acquire a stable template because abnormal beats are very rare. However, the median is unsuitable for lightweight embedded devices because it requires sorting for calculation, and all signals must be stored in the memory.
Fig. 11. The problem with determining the template based on the average of heartbeats.

Proposed Method

Pearson Similarity
In general, the similarity between the two signals can be easily compared using the SSD of (1). However, as shown in Fig. 9, a large error appears even signal’s shapes are similar if the baseline or scale of amplitude changes although the signal has been filtered. The proposed method applies the Pearson similarity, which is robust to changes in the baseline and scale of amplitude.
The Pearson similarity is also called the Pearson correlation coefficient. The Pearson similarity for the two signals (X,Y) is given by (2).

(2)

where μ and σ represent the signal’s mean and standard deviation, respectively. Pearson similarity is robust to baseline movement and scale of amplitude changes because it normalizes the signals, as shown in (2).

Template Cluster Generation
The process for generating a template cluster using the Pearson similarity as proposed in this paper consists of three steps:

Step 1. Initialize the template First, determine the initial input beat as the first template in the cluster. Next, set the first template’s counter to 1.
The counter indicates the template’s importance. The higher the counter, the greater the number of times the template matches the input beat. Thus, the ideal normal beat template has the highest counter in the template cluster.

Step 2. Update the template For each input signal, detect the most similar template. If the similarity exceeds the threshold, use the counter as a weight to update the template with a weighted mean, and increase the counter of the updated template by 1. If the similarity does not exceed the threshold, add the input signal as a new template, and set the counter of the added template to 1.
The weighted mean of the ith sample of the kth template C_S (i,k) with the input signal S is calculated using the counter C_C (k) as a weight, as shown in (3).

(3)

In contrast to the average, the weighted mean can suppress signal distortion even when an abnormal beat is included. In addition, unlike the median, the sorting operation is unnecessary, and the counter is used as a weight. Therefore, even in a low-memory embedded device, real-time updates are possible using only the input beat and the template counter.

Step 3. Remove the template Designate the template with the highest counter as the representative template, then remove templates with low similarity to the representative template.
This is based on the fact that the majority of ECG signals are composed of normal beats, and abnormal beats are rare. Accordingly, when a template cluster is formed, most of the beats match the normal beat template, thus the template with the highest counter becomes the normal beat template. Therefore, this template is used as the representative template, and those templates with low similarity to the representative template are designated as abnormal beat templates and removed. By following these steps, a template cluster consisting of normal beats can be generated. Fig. 12 is a diagram of the algorithm flow of this process.
Fig. 12. Algorithm flow chart of template cluster generation.


Experiment and Results

MIT-BIH ADB
The data used in the experiment were taken from the MIT-BIH ADB provided by PhysioNet. In the experiment, a template cluster was generated for the MIT-BIH ADB data, and then its similarity with the template cluster was assessed in order to determine its performance in classifying normal and abnormal beats. The method of classification proposed in this paper is a shape-based classification method focused on detecting abnormal beats with QRS complex and P-wave deformations due to arrhythmia. Accordingly, using the MIT-BIH ADB beats, the method’s performance in classifying abnormal beats for PVCs including QRS complex deformations, and for PJCs including P-wave deformations, was confirmed.
We evaluated its classification performance by measuring sensitivity (Se) and specificity (Sp), according to true positive (TP), true negative (TN), false positive (FP), and false negative (FN). TP and TN are the correct results of the classification of normal and abnormal heartbeats, respectively. Conversely, FP and FN are the results of the misclassification of abnormal and normal beats, respectively. Se, Sp, and Ac (accuracy) measure the over-detection rate, the abnormal beat non-detection rate, and the correct total beat detection rate, respectively, as shown in (4).

(4)

When cardiologists diagnose arrhythmia, it is important to minimize Sp because FP significantly affects false diagnosis compared to FN.

Experiment
For the purposes of this paper, a template was constructed for the P-wave region for P-wave detection. To confirm this additional template’s effectiveness, an experiment was conducted on Record #234 of an MIT-BIH ADB dataset, which is a representative record that includes PJCs. Table 2 shows the results of detection using only the template for the entire region, as well as the results obtained when using the P-wave region template as well.

Table 2. Comparison of detection performance according to P-wave template
Measured template region Total beats Abnormal beats (PVCs, PJCs) TP TN FP FN Specificity (%) Accuracy (%)
Whole region 2,750 53 2,696 3 50 1 5.66 98.15
Whole region + P-wave region 2,695 53 0 2 100 99.93
It was confirmed that including the P-wave region template correctly detected existing, previously undetected abnormal heartbeats.
Table 3 shows the experimental results when the sum of Sp and Se is maximized for the 46 units of data taken from the MIT-BIH ADB dataset.
Record #208 and Record #218 showed large distortions in the normal beats, and the number of abnormal beats was also quite large. As such, they were unsuitable for applying the proposed algorithm because the normal template cluster was not generated reliably, and thus were excluded from the experiment. As a result of the experiment, Se, Sp, and Ac for the whole data exceeded 96%. In particular, in the case of Record #100 to #124, which were relatively more stable than Record #200 to #234, Sp exceeded 99%, and the non-detection rate of abnormal heartbeats was very low. In addition, no over-detection occurred even when the algorithm was applied to data in which abnormal heartbeats did not exist, such as Record #101. For the data shown in Table 3, the average execution time was 0.39 seconds for a 30-minute signal period, confirming the possibility of the algorithm’s real-time operation in embedded equipment.


Conclusion

This paper proposes a method of detecting abnormal heartbeats using a template cluster. With existing templates using the mean or median, it is difficult to generate a reliable template due to the distortion caused by abnormal heartbeats. In contrast, the proposed method minimized the distortion caused by abnormal beats by using a template cluster that generated normal and abnormal templates separately and a weighted mean that had a template counter as a weight. Moreover, unlike the SSD method, which is vulnerable to changes in the baseline and the scale of amplitude, the proposed method made possible a stable comparison of similarity using the Pearson similarity, and minimized the distortion that abnormal beats cause. In addition, a template for the P-wave region was obtained separately, and as a result of the experiment, the rate of detection of PJC, which is generally difficult to detect, was greatly increased. Therefore it was confirmed that the proposed method was suitable for use as an abnormal heartbeat detection algorithm.
However, in the case of Record #200, for which the baseline movement and power noise were considerable, a problem arose: deformed normal beats were over-detected as abnormal beats. To improve on this, additional studies on preprocessing techniques for reducing signal distortion are required. In addition, it will be necessary to improve the detection rate, which could be done by analyzing feature values based on the fiducial point of the detected abnormal beats, and by reclassifying the normal beats accordingly.
Moreover, the degree of deformation of abnormal beats differs depending on the individual, and various types of abnormal beats occur depending on the type of arrhythmia. Therefore, the threshold value used to generate the template cluster and detect abnormal heartbeats is problematic in that it is manually determined. Advanced research will be required to study a method of determining an adaptive threshold value based on the distribution of similarities between templates in the template cluster generation step.

Table 3. Experiment result in MIT-BIH ADB

Unit Record# Total beat PVC & PJC TP TN FP FN Sensitivity (%) Specificity (%) Accuracy (%)
1 100 2,237 1 2,236 1 0 0 100 100 100
2 101 1,857 0 1,857 0 0 0 100 100 100
3 102 2,029 4 2,025 4 0 0 100 100 100
4 103 2,079 0 2,079 0 0 0 100 100 100
5 104 1,379 2 1,369 2 0 8 99.4 100 99.4
6 105 2,564 41 2,138 41 0 385 84.7 100 85
7 106 2,024 520 1,502 520 0 2 99.9 100 99.9
8 107 2,134 59 2,073 50 9 2 99.9 84.8 99.5
9 108 1,753 17 1,503 17 0 233 86.6 100 86.7
10 109 2,527 38 2,476 38 0 13 99.5 100 99.5
11 111 2,121 1 2,119 1 0 1 99.9 100 99.9
12 112 2,534 0 2,534 0 0 0 100 100 100
13 113 1,786 0 1,786 0 0 0 100 100 100
14 114 1,862 45 1,811 45 0 6 99.7 100 99.7
15 115 1,950 0 1,950 0 0 0 100 100 100
16 116 2,408 109 2,277 109 0 22 99 100 99.1
17 117 1,531 0 1,531 0 0 0 100 100 100
18 118 2,179 16 2,157 16 0 6 99.7 100 99.7
19 119 1,984 443 1,541 443 0 0 100 100 100
20 121 1,859 1 1,857 1 0 1 99.9 100 99.9
21 122 2,473 0 2,473 0 0 0 100 100 100
22 123 1,515 3 1,503 3 0 9 99.4 100 99.4
23 124 1,604 76 1,444 73 3 84 94.5 96.1 94.6
  Sub-total 46,389 1,376 44,241 1,364 12 772 98.3 99.1 98.3
24 200 2,566 825 1,605 757 68 136 92.2 91.8 92.1
25 201 1,821 199 1,396 198 1 226 86.1 99.5 87.5
26 202 2,077 19 1,947 18 1 111 94.6 94.7 94.6
27 203 2,970 444 2,308 402 42 218 91.4 90.5 91.3
28 205 2,639 71 2,566 71 0 2 99.9 100 99.9
29 207 1,561 104 1,295 100 4 162 88.9 96.2 89.4
30 209 2,619 1 2,618 1 0 0 100 100 100
31 210 2,614 194 2,386 174 20 34 98.6 89.7 97.9
32 212 1,823 0 1,823 0 0 0 100 100 100
33 213 2,858 220 2,474 208 12 164 93.8 94.6 93.8
34 214 2,256 256 1,982 255 1 18 99.1 99.6 99.2
35 215 3,356 164 3,125 161 3 67 97.9 98.8 97.9
36 217 1,701 162 1,398 162 0 141 90.8 100 91.7
37 219 2,143 64 1,919 58 6 160 92.3 90.6 92.3
38 220 1,951 0 1,951 0 0 0 100 100 100
39 221 2,424 396 1,479 374 22 549 72.9 94.4 76.4
40 222 2,060 1 1,284 1 0 775 62.4 100 62.4
41 223 2,499 473 2,000 473 0 26 98.7 100 99
42 228 2,047 362 1,637 362 0 48 97.2 100 97.7
43 230 2,253 1 2,252 1 0 0 100 100 100
44 231 1,253 2 1,251 2 0 0 100 100 100
45 233 3,058 830 2,223 830 0 5 99.8 100 99.8
46 234 2,750 53 2,695 53 0 2 99.9 100 99.9
  Sub-total 53,299 4,841 45,614 4,661 180 2,844 94.1 96.3 94.3
  Total 99,688 6,217 89,855 6,025 192 3,616 96.1 96.9 96.2


Author’s Contributions

SL designed the entire core architecture and performed the hardware/software implementation and experiments; DP has his responsibility in writing entire paper as the corresponding author.


Funding

This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (No. NRF-2020R1I1A1A01072343, NRF-2018R1A6A1A03025109), the Ministry of Science and ICT (No. NRF-2019R1A2C2005099).


Competing Interests

The authors declare that they have no competing interests.


References

[1] T. H. Kim, S. Y. Kim, J. H. Kim, B. J. Yun, and K. H. Park, “Curvature based ECG signal compression for effective communication on WPAN,” Journal of Communications and Networks, vol. 14, no. 1, pp. 21-26, 2012.
[2] H. Mamaghanian, N. Khaled, D. Atienza, and P. Vandergheynst, “Compressed sensing for real-time energy-efficient ECG compression on wireless body sensor nodes,” IEEE Transactions on Biomedical Engineering, vol. 58, no. 9, pp. 2456-2466, 2011.
[3] M. Abdel-Basset, G. Manogaran, A. Gamal, and V. Chang, “A novel intelligent medical decision support model based on soft computing and IoT,” IEEE Internet of Things Journal, vol. 7, no. 5, pp. 4160-4170, 2020.
[4] A. Arshad, S. Khan, A. Z. Alam, R. Tasnim, and R. I. Boby, “Health and wellness monitoring of elderly people using intelligent sensing technique,” in Proceedings of 2016 International Conference on Computer and Communication Engineering (ICCCE), Kuala Lumpur, Malaysia, 2016, pp. 231-235.
[5] W. Lee, N. Kim, and B. D. Lee, “An adaptive transmission power control algorithm for wearable healthcare systems based on variations in the body conditions,” Journal of Information Processing Systems, vol. 15, no. 3, pp. 593-603, 2019.
[6] H. Rhim, K. Tamine, R. Abassi, D. Sauveron, and S. Guemara, “A multi-hop graph-based approach for an energy-efficient routing protocol in wireless sensor networks,” Human-centric Computing and Information Sciences, vol. 8, article no. 30, 2018. https://doi.org/10.1186/s13673-018-0153-6
[7] T. Teraoka, “Organization and exploration of heterogeneous personal data collected in daily life,” Human-Centric Computing and Information Sciences, vol. 2, article no. 1, 2012. https://doi.org/10.1186/2192-1962-2-1
[8] B. Farahani, F. Firouzi, V. Chang, M. Badaroglu, N. Constant, and K. Mankodiya, “Towards fog-driven IoT eHealth: promises and challenges of IoT in medicine and healthcare,” Future Generation Computer Systems, vol. 78, pp. 659-676, 2018.
[9] H. Kaur, M. A. Alam, R. Jameel, A. K. Mourya, and V. Chang, “A proposed solution and future direction for blockchain-based heterogeneous medicare data in cloud environment,” Journal of Medical Systems, vol. 42, article no. 156, 2018. https://doi.org/10.1007/s10916-018-1007-5
[10] S. A. Israel, J. M. Irvine, A. Cheng, M. D. Wiederhold, and B. K. Wiederhold, “ECG to identify individuals,” Pattern Recognition, vol. 38, no. 1, pp. 133-142, 2005.
[11] J. S. Arteaga-Falconi, H. Al Osman, and A. El Saddik, “ECG authentication for mobile devices,” IEEE Transactions on Instrumentation and Measurement, vol. 65, no. 3, pp. 591-600, 2016.
[12] L. Khriji and A. M. Al-Busaidi, “New adaptive thresholding-based ECG R-peak detection technique,” in Proceedings of 2018 IEEE 4th Middle East Conference on Biomedical Engineering (MECBME), Tunis, Tunisia, 2018, pp. 147-152.
[13] S. Lee, Y. Jeong, J. Kwak, D. Park, and K. H. Park, “Efficient communication overhead reduction using polygonal approximation-based ECG signal compression,” in Proceedings of 2019 International Conference on Artificial Intelligence in Information and Communication (ICAIIC), Okinawa, Japan, 2019, pp. 058-061.
[14] S. Lee, Y. Jeong, J. Kwak, D. Park, and K. H. Park, “Advanced real-time dynamic programming in the polygonal approximation of ECG signals for a lightweight embedded device,” IEEE Access, vol. 7, pp. 162850-162861, 2019.
[15] S. Lee and D. Park, “Enhanced dynamic programming for polygonal approximation of ECG signals,” in Proceedings of 2020 IEEE 2nd Global Conference on Life Sciences and Technologies (LifeTech), Kyoto, Japan, 2020, pp. 121-122.
[16] S. Lee, Y. Jeong, D. Park, B. J. Yun, and K. H. Park, “Efficient fiducial point detection of ECG QRS complex based on polygonal approximation,” Sensors, vol. 18, no. 12, article no. 4502, 2018. https://doi.org/10.3390/s18124502
[17] S. Yazdani and J. M. Vesin, “Extraction of QRS fiducial points from the ECG using adaptive mathematical morphology,” Digital Signal Processing, vol. 56, pp. 100-109, 2016.
[18] H. Y. Lin, S. Y. Liang, Y. L. Ho, Y. H. Lin, and H. P. Ma, “Discrete-wavelet-transform-based noise removal and feature extraction for ECG signals,” IRBM, vol. 35, no. 6, pp. 351-361, 2014.
[19] A. Martinez, R. Alcaraz, and J. J. Rieta, “Application of the phasor transform for automatic delineation of single-lead ECG fiducial points,” Physiological Measurement, vol. 31, no. 11, pp. 1467-1485, 2010.
[20] A. Illanes-Manriquez and Q. Zhang, “An algorithm for robust detection of QRS onset and offset in ECG signals,” in Proceedings of 2008 Computers in Cardiology, Bologna, Italy, 2008, pp. 857-860.
[21] J. P. Madeiro, P. C. Cortez, J. A. Marques, C. R. Seisdedos, and C. R. Sobrinho, “An innovative approach of QRS segmentation based on first-derivative, Hilbert and Wavelet Transforms,” Medical Engineering & Physics, vol. 34, no. 9, pp. 1236-1246, 2012.
[22] P. Laguna, R. Jane, and P. Caminal, “Automatic detection of wave boundaries in multilead ECG signals: validation with the CSE database,” Computers and Biomedical Research, vol. 27, no. 1, pp. 45-60, 1994.
[23] S. Lee, D. Park, and K. H. Park, “QRS complex detection based on primitive,” Journal of Communications and Networks, vol. 19, no. 5, pp. 442-450, 2017.
[24] A. Li, S. Wang, H. Zheng, L. Ji, and J. Wu, “A novel abnormal ECG beats detection method,” in Proceedings of the 2nd International Conference on Computer and Automation Engineering (ICCAE), Singapore, 2020, pp. 47-51.
[25] A. S. Malek, A. Elnahrawy, H. Anwar, and M. Naeem, “Automated detection of premature ventricular contraction in ECG signals using enhanced template matching algorithm,” Biomedical Physics & Engineering Express, vol. 6, no. 1, article no. 015024, 2020. https://doi.org/10.1088/2057-1976/ab6995
[26] M. Komeili, N. Armanfard, and D. Hatzinakos, “Liveness detection and automatic template updating using fusion of ECG and fingerprint,” IEEE Transactions on Information Forensics and Security, vol. 13, no. 7, pp. 1810-1822, 2018.
[27] G. Bansal, P. Gera, and D. R. Bathula, “Template based classification of cardiac Arrhythmia in ECG data,” in Proceedings of 2015 IEEE 2nd International Conference on Recent Trends in Information Systems (ReTIS), Kolkata, India, 2015, pp. 337-341.
[28] A. Rehman, N. A. Saqib, S. M. Danial, and S. H. Ahmed, “ECG based authentication for remote patient monitoring in IoT by wavelets and template matching,” in Proceedings of 2017 8th IEEE International Conference on Software Engineering and Service Science (ICSESS), Beijing, China, 2017, pp. 91-94.
[29] R. J. Huszar, Basic Dysrhythmias: Interpretation & Management. St. Louis, MO: Mosby JEMS, 2007.
[30] H. L. Chan, W. S. Chou, S. W. Chen, S. C. Fang, C. S. Liou, and Y. S. Hwang, “Continuous and online analysis of heart rate variability,” Journal of Medical Engineering & Technology, vol. 29, no. 5, pp. 227-234, 2005.
[31] G. D. Clifford, F. Azuaje, and P. McSharry, Advanced Methods and Tools for ECG Data Analysis. Boston, MA: Artech House, 2006.
[32] Testing and reporting performance results of cardiac rhythm and ST segment measurement algorithms, ANSI/AAMI EC57:1998, 1999.
[33] G. B. Moody and R. G. Mark, “The MIT-BIH arrhythmia database on CD-ROM and software for use with it,” in Proceedings of 1990 Computers in Cardiology, Chicago, IL, 1990, pp. 185-188.
[34] M. Merone, P. Soda, M. Sansone, and C. Sansone, “ECG databases for biometric systems: a systematic review,” Expert Systems with Applications, vol. 67, pp. 189-202, 2017.
[35] F. A. Elhaj, N. Salim, A. R. Harris, T. T. Swee, and T. Ahmed, “Arrhythmia recognition and classification using combined linear and nonlinear features of ECG signals,” Computer Methods and Programs in Biomedicine, vol. 127, pp. 52-63, 2016.
[36] G. M. Friesen, T. C. Jannett, M. A. Jadallah, S. L. Yates, S. R. Quint, and H. T. Nagle, “A comparison of the noise sensitivity of nine QRS detection algorithms,” IEEE Transactions on Biomedical Engineering, vol. 37, no. 1, pp. 85-98, 1990.
[37] S. K. Berkaya, A. K. Uysal, E. S. Gunal, S. Ergin, S. Gunal, and M. B. Gulmezoglu, “A survey on ECG analysis,” Biomedical Signal Processing and Control, vol. 43, pp. 216-235, 2018.
[38] R. Alcaraz, F. Hornero, and J. J. Rieta, “Dynamic time warping applied to estimate atrial fibrillation temporal organization from the surface electrocardiogram,” Medical Engineering & Physics, vol. 35, no. 9, pp. 1341-1348, 2013.
[39] A. P. James, “Heart rate monitoring using human speech spectral features,” Human-centric Computing and Information Sciences, vol. 5, article no. 33, 2015. https://doi.org/10.1186/s13673-015-0052-z
[40] J. Pan and W. J. Tompkins, “A real-time QRS detection algorithm,” IEEE Transactions on Biomedical Engineering, vol. 32, no. 3, pp. 230-236, 1985.

About this article
Cite this article

Seungmin Lee1 and Daejin Park2,3,*, A Real-Time Abnormal Beat Detection Method Using a Template Cluster for the ECG Diagnosis of IoT Devices, Article number: 11:04 (2021) Cite this article 7 Accesses

Download citation
  • Recived15 September 2020
  • Accepted21 December 2020
  • Published29 January 2021
Share this article

Anyone you share the following link with will be able to read this content:

Provided by the Springer Nature SharedIt content-sharing initiative

Keywords
  • ECG
  • Template
  • Abnormal Beat Detection
  • PVC
  • PJC
  • Embedded System