# Data Fusion of Multivariate Time Series: Application to Noisy 12-Lead ECG Signals

^{1}

^{2}

^{3}

^{4}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. The Local Weighted Linear Prediction Algorithm

_{k}is the current state point of the chaotic system, and the future state X

_{k+}

_{1}of the system needs to be predicted. In LWLPA, the neighboring states $\{{X}_{ki},i=1,2,\cdots ,n\}$ of the current state point X

_{k}need to be chosen from the reconstructed trajectories. With the neighborhood X

_{ki}and the linear prediction model ${X}_{k+1}=ae+b{X}_{k}$, where $e={[1,\cdots ,1]}_{m}^{T}$, the future state X

_{k+}

_{1}can be approximately estimated.

## 3. The Novel Data Fusion Algorithm

#### 3.1. Basic Idea of Novel Data Fusion Algorithm

_{1}and L

_{2}, shown in Figure 1. In this example, suppose that the trajectory L

_{F}is the fused result of the trajectories L

_{1}and L

_{2}. Furthermore, the state point X

_{F}on the trajectory L

_{F}should satisfy the linear prediction model ${X}_{F}(p+1)=ae+b{X}_{F}(p)$, where $e={[1,\cdots ,1]}_{m}^{T}$. Here, how to obtain the parameters a and b in the linear prediction model is a critical problem.

_{1}(p) and X

_{2}(p) can be regarded as the neighboring vectors of the current state X

_{F}(p). With the two vectors being employed, we can calculate the parameters a and b by (5). The equation is as follows:

_{F}will be employed for the original signal quality assessment. It implies that to some extent, the characteristic information of original signal ought to be fused in the trajectory L

_{F}. Here, how to effectively inherit the characteristic information by fused result is a key for NDFA.

_{1}(p − 1) on the trajectory L

_{1}and the time-domain characteristic should be well inherited in the fused trajectory L

_{F}. Thus, based on Equation (1), a greater value for weighted coefficient ${\omega}_{1}$ of the state X

_{1}(p − 1) should be chosen, which will further enhance the impact on the final result. Here, the Euclidean distance of two neighboring points and the angle between two neighboring vectors on evolutionary trajectory L are used to estimate the weighted coefficients. In Figure 1, X

_{1}(p) and X

_{2}(p) are the p-step state points on the evolutionary trajectories L

_{1}and L

_{2}, respectively. The vector can be easily calculated through two adjacent state points, e.g., for the evolutionary trajectories L

_{1}, ${\stackrel{\rightharpoonup}{V}}_{p}^{1}={X}_{1}(p)-{X}_{1}(p-1)$, and then the angle ${\theta}_{p}^{1}$ can be obtained via the two neighboring vectors ${\stackrel{\rightharpoonup}{V}}_{p}^{1}$ and ${\stackrel{\rightharpoonup}{V}}_{p-1}^{1}$. The modulus of vector ${\stackrel{\rightharpoonup}{V}}_{p}^{1}$ is the Euclidean distance between X

_{1}(p) and X

_{1}(p − 1). For the lth evolutionary trajectory L

_{l}, the modulus of vector ${\stackrel{\rightharpoonup}{V}}^{l}$ and the angle ${\theta}^{l}$ can well reflect the evolutionary trend of the trajectory. In Figure 1, for the trajectory L

_{1}, the values of the modulus of ${\stackrel{\rightharpoonup}{V}}_{p}^{1}$, ${\stackrel{\rightharpoonup}{V}}_{p-1}^{1}$ and the vector angles ${\theta}_{p}^{1}$, ${\theta}_{p-1}^{1}$ are greater than the values of the trajectory L

_{2}at the same step. Characteristics of evolutionary trajectory can be described objectively by them. According to the idea, the change of evolutionary trajectory is positively related to the values of the two parameters. Based on the relationship, the weighted coefficient of the data point can be approximately estimated.

#### 3.2. Fuzzy Inference System Design for NDFA

_{d}and $\mathrm{FI}{\mathrm{S}}_{\alpha}$, will be devised, which can be used to estimate the weighted coefficients of LWLPA by the modulus of vector and the angle, respectively.

_{d}is applied to estimate the evolutionary trend of the reconstruction trajectory by the modulus D of the vector and the change rate D

_{r}of the modulus of adjacent vectors. Thus, in FIS

_{d}, there are two input variables D, D

_{r}and one output variable O

_{d}.

_{r}can be calculated as:

_{r}(p) are the Euclidean distance and the change rate of X(p) at p-step, respectively.

_{d}, the universe of the three variables D, D

_{r}and O

_{d}are set within the interval [0,1] uniformly. The universe of the variables D, D

_{r}and O

_{d}are divided into several fuzzy sets and the numbers of the fuzzy sets are 3, 3 and 5, respectively, shown in Figure 2a–c. According to the aforementioned relationship, the inference rules of FIS

_{d}can be designed properly, which are summarized in Table 1. Based on FIS

_{d}, the evolutionary trend of trajectory is estimated quantitatively and the parameter ${\omega}_{d}$ can be calculated as:

_{d}, ${y}_{d}(q)$ is the output of the qth rule and ${\beta}_{d}(q)$ the rule activation for the qth rule.

_{l}, the weighted coefficient ${\omega}_{l}$ in the Equation (1) can be computed as:

_{n}is the number of the phase trajectories and the parameter $\gamma $ is set to 1 [26]. Here, the minimum value ${\omega}_{\mathrm{min}}(p)=\mathrm{min}\{{\tilde{\omega}}_{s}(p),s=1,2,\cdots ,{L}_{n}\}$ needs to be selected.

#### 3.3. NDFA Algorithm

_{max}, delay time ${\tau}_{\mathrm{min}}$, initial condition X

_{F}(0), and for each lead of 12-lead ECG signals, construct the vector X

_{l}as:

_{l}is the evolutionary trajectory of the lth lead ECG signal on reconstructed trajectory. X

_{F}(0) is chosen as centroid of all the neighbor neighboring vectors ${X}_{s}\left(0\right)=(x(0),x(0+{\tau}_{\mathrm{min}}),\cdots ,x(0+({m}_{\mathrm{max}}-1){\tau}_{\mathrm{min}}))$, which is the point of the reconstruction trajectory of the sth lead ECG signal at p = 0.

_{l}, calculate the parameters D(p), D

_{r}(p), $\alpha (p)$ and ${\alpha}_{r}(p)$ by Equations (3), (4), (6), and (7), respectively. With the two FISs, ${\omega}_{d}(p)$ and ${\omega}_{\alpha}(p)$ being properly estimated, the weighted coefficient ${\omega}_{l}(p)$ of the state X

_{l}(p) at the p-step can be computed by Equation (10).

_{F}(p + 1) is calculated as:

## 4. Application of NDFA in 12-Lead ECG Signals

_{s}and delay time ${\tau}_{s}$.

#### 4.1. Synthetic Signals Experiments

#### 4.1.1. Ideal Synthetic Signals Experiments

_{x}, V

_{y}and V

_{z}, respectively. Figure 4d represents the fused signal of VCG signals s. From the morphology of reconstructed trajectory perspective, it is clear that there are needle-like features (Feature 1) on the three trajectories. Meanwhile, the longer closed trajectory (Feature 3) and disordered feature of the closed trajectory within a small space (Feature 2) are shown distinctly in Figure 4a–c. In Figure 4a, the local trajectories of Features 2 and 3 essentially reflect the P wave and QRS complexes in ECG signal, respectively. Evidently, the three features of original VCG signals are well described by the fused trajectory s.

#### 4.1.2. Noise Contaminated Synthetic Signals Experiments

_{x}of VCG signals is randomly chosen, which is contaminated by the noise. The parameters of the SNR levels are summarized in Table 3 [11].

_{x}is polluted by BW and the magnitudes of SNR are 12 dB, 6 dB, 0 dB, and −6 dB, respectively. In Table 4, correlation coefficient $Co{r}_{\overline{x}}$ is the mean value of Cor(Vx,Vy) and Cor(Vx,Vz). The parameter $Co{r}_{\overline{s}}$ is the mean value of Cor(s,Vx), Cor(s,Vy), and Cor(s,Vz). The values of parameters $Co{r}_{\overline{x}}$ and $Co{r}_{\overline{s}}$ reflect the degree of correlations between the original and fused signals. From Table 4, we can find that, under the different SNR levels, the correlation coefficient $Co{r}_{\overline{s}}$ is greater than $Co{r}_{\overline{x}}$ consistently, thereby designating the effectiveness of NDFA.

_{x}signal is polluted by the noise of EM and MA with different magnitudes of SNR and the correlations coefficients are calculated in Table 5 and Table 6. In the two tables, the relations between noisy signals and fused results are highly consistent with the relation reflected in Table 4.

#### 4.2. Realistic Signals Experiments

_{inv}is given by

#### 4.3. Performance Comparison of Data Fusion Algorithms

_{DTW}, and the smaller value of D

_{DTW}implies that the morphology features of two signals were somewhat similar. Figure 8a shows the similarity between the fused signals by four algorithms and the lead MLII. The means and variances of the parameter D

_{DTW}derived by NDFA, MDF, algorithm 1, and algorithm 2 were 0.0616, 0.0011; 0.8474, 0.0110; 0.2257, 0.0645; 0.2256, and 0.0644, respectively. Analogously, for fused signals and the lead V5, the means and variances of the parameter D

_{DTW}based on NDFA and MDF were 0.0921, 0.0017; 0.8768, 0.0102; 0.1959, 0.0426; 0.1958, and 0.0425 in Figure 8b. As can be seen, via NDFA, the means and the variances of the parameter D

_{DTW}were smaller, compared with the experiment results under the others. Figure 8 demonstrates that the characteristics of two leads ECG signals, MLII and V5, can be well reserved on the fused signal yielded from NDFA, and reflects that the performance of our algorithm exceeds MDF, algorithms 1 and 2.

## 5. Conclusions

## Author Contributions

## Funding

## Acknowledgments

## Conflicts of Interest

## References

- Clifford, G.D. ECG statistics, noise, artifacts, and missing data. In Engineering in Medicine and Biology, 1st ed.; Artech House: Norwood, MA, USA, 2006. [Google Scholar]
- Behar, J.; Oster, J.; Li, Q.; Clifford, G.D. ECG signal quality during arrhythmia and its application to false alarm reduction. IEEE Trans. Biomed. Eng.
**2013**, 60, 1660–1666. [Google Scholar] [CrossRef] - Moody, G.B.; Mark, R.G. QRS morphology representation and noise estimation using the Karhumen-Loeve transform. Comput. Cardiol.
**1989**, 16, 267–272. [Google Scholar] - Allen, J.; Murray, A. Assessing ECG signal quality on a coronary care unit. Physiol. Meas.
**1996**, 17, 249–258. [Google Scholar] [CrossRef] [PubMed] - Kalkstein, N.; Kinar, Y.; Na’aman, M.; Neumark, N.; Akiva, P. Using machine learning to detect problems in ECG data collection. Comput. Cardiol.
**2011**, 38, 437–440. [Google Scholar] - Chudacek, V.; Zach, L.; Kuzilek, J.; Spilka, J.; Lhotska, L. Simple scoring system for ECG quality assessment on android platform. Comput. Cardiol.
**2011**, 38, 449–451. [Google Scholar] - Zaunseder, S.; Huhle, R.; Malberg, H. CinC Challenge—Assessing the usability of ECG by ensemble decision trees. Comput. Cardiol.
**2011**, 38, 277–280. [Google Scholar] - Jekova, I.; Krasteva, V.; Dotsinsky, I.; Christov, I.; Abcherli, R. Recognition of diagnostically useful ECG recordings: Alert for corrupted or interchanged leads. Comput. Cardiol.
**2011**, 38, 429–432. [Google Scholar] - Clifford, G.D.; Azuaje, F.; Mcsharry, P. Advanced Methods and Tools for ECG Data Analysis; Artech House: Norwood, MA, USA, 2006. [Google Scholar]
- Li, Q.; Mark, R.G.; Clifford, G.D. Robust heart rate estimation from multiple asynchronous noisy sources using signal quality indices and a Kalman filter. Physiol. Meas.
**2008**, 29, 15–32. [Google Scholar] [CrossRef] - Li, Q.; Rajagopalan, C.; Clifford, G.D. A mathine learning approach to multi-level ECG signal quality classification. Comput. Meth. Prog. Biol.
**2014**, 117, 435–447. [Google Scholar] [CrossRef] - Andreotti, F.; Grasser, F.; Malberg, H.; Zaunseder, S. Non-Invasive Fetal ECG Signal Quality Assessment for Multichannel Heart Rate Estimation. IEEE Trans. Biomed. Eng.
**2017**, 64, 2793–2802. [Google Scholar] - Liu, C.; Zhang, X.; Zhao, L.; Liu, F.; Chen, X.; Yao, Y.; Li, J. Signal Quality Assessment and Lightweight QRS Detection for Wearable ECG SmartVest System. IEEE Internet Things J.
**2018**. [Google Scholar] [CrossRef] - Satija, U.; Ramkumar, B.; Manikandan, M.S. A Review of Signal Processing Techniques for Electrocardiogram Signal Quality Assessment. IEEE Rev. Biomed. Eng.
**2018**, 11, 36–52. [Google Scholar] [CrossRef] [PubMed] - Chen, Y.; Yang, H. Self-organized neural network for the quality control of 12-lead ECG signals. Physiol. Meas.
**2012**, 33, 1399–1418. [Google Scholar] [CrossRef] [PubMed] - Shahriari, Y.; Fidler, R.; Pelter, M.; Bai, Y.; Villaroman, A.; Hu, X. Electrocardiogram Signal Quality Assessment Based on Structural Image Similarity Metric. IEEE Trans. Biomed. Eng.
**2017**, 65, 745–753. [Google Scholar] [CrossRef] [PubMed] - Agrafioti, F.; Dimitrios, H. Fusion of ECG sources for human identification. In Proceedings of the 3rd International Symposium on Communications, Control and Signal Processing, St Julians, Malta, 12–14 March 2008; pp. 1542–1547. [Google Scholar]
- Yu, Q.; Guan, Q.; Li, P.; Liu, T.; Huang, X.; Zhao, Y.; Liu, H.; Wang, Y. Fusion of detected multi-channel maternal electrocardiogram (ECG) R-wave peak locations. BioMed. Eng. OnLine
**2016**, 15, 1–16. [Google Scholar] [CrossRef] [PubMed] - Nweke, H.F.; Ten, Y.W.; Mujtaba, G.; Al-garadi, M.A. Data Fusion and Multiple Classifier Systems for Human Activity Detection and Health Monitoring: Review and Open Research Directions. Inf. Fusion
**2019**, 46, 147–170. [Google Scholar] [CrossRef] - Cong, R.; Liu, S.L.; Ma, R. An approach to phase space reconstruction from multivariate data based on data fusion. Acta Phys. Sin.
**2008**, 57, 7487–7493. (In Chinese) [Google Scholar] - Diao, C.; Wang, B.; Cai, N. Data fusion of multivariate time series based on local weighted zero-order prediction algorithm. In Proceedings of the 2017 36th Chinese Control Conference (CCC), Dalian, China, 26–28 July 2017; pp. 5663–5667. [Google Scholar]
- Diao, C.; Wang, B.; Cai, N. A Novel Data Fusion Algorithm for Multivariate Time Series. In Proceedings of the 2018 Chinese Control and Decision Conference (CCDC), Shenyang, China, 9–11 June 2018; pp. 6331–6333. [Google Scholar]
- Govindan, R.B.; Narayanan, K.; Gopinathan, M.S. On the evidence of deterministic chaos in ECG: Surrogate and predictability analysis. Chaos Interdiscip. J. Nonlinear Sci.
**1998**, 8, 495–502. [Google Scholar] [CrossRef] - Takens, F. Detecting strange attractors in turbulence. Lect. Notes Math.
**1981**, 898, 366–381. [Google Scholar] - Zadeh, L.A. Fuzzy sets. Inf. Control
**1965**, 8, 338–353. [Google Scholar] [CrossRef][Green Version] - Qu, J.; Wang, X.; Qiao, Y.; Gao, F.; Zhou, D. An improved local weighted linear prediction model for chaotic time series. Chin. Phys. Lett.
**2014**, 31, 16–20. [Google Scholar] [CrossRef] - Moody, G.B.; Muldrow, W.K.; Mark, R.G. A noise stress test for arrhythmia detectors. Comput. Cardiol.
**1984**, 11, 381–384. [Google Scholar] - Goldberger, A.L.; Amaral, L.A.; Glass, L.; Hausdorff, J.M.; Ivanov, P.C.; Mark, R.G.; Mietus, J.E.; Moody, G.B.; Peng, C.K.; Stanley, H.E. PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation
**2000**, 101, 215–220. [Google Scholar] [CrossRef] - Silva, I.; Mood, G.B.; Celi, L. Improving the quality of ECGs collected using mobile phones: The PhysioNet/Computing in Cardiology Challenge 2011. Comput. Cardiol.
**2011**, 38, 273–276. [Google Scholar] - McSharry, P.E.; Clifford, G.D.; Tarassenko, L.; Smith, L.A. A dynamical model for generating synthetic electrocardiogram signals. Biomed. Eng.
**2003**, 50, 289–294. [Google Scholar] [CrossRef] [PubMed][Green Version] - Sameni, R.; Clifford, G.D.; Jutten, C.; Shamsollahi, M.B. Multichannel ECG and noise modeling: Application to maternal and fetal ECG signals. EURASIP J. Appl. Signal Proces.
**2007**, 2007, 94. [Google Scholar] [CrossRef] - Clifford, G.D.; Nemati, S.; Sameni, R. An artificial vector model for generating abnormal electrocardiographic rhythms. Physiol. Meas.
**2010**, 31, 595–609. [Google Scholar] [CrossRef][Green Version] - Daubechies, A.A.I. The continuous Procrustes distance between two surfaces. Mathematics
**2011**, 29, 322–324. [Google Scholar] - Dahl, T.; Naes, T. Outlier and group detection in sensory panels using hierarchical cluster analysis with the Procrustes distance. Food Qual. Preference
**2004**, 15, 195–208. [Google Scholar] [CrossRef] - Ejlali, N.; Faghihi, M.R.; Sadeghi, M. Bayesian comparison of protein structures using partial Procrustes distance. Stat. Appl. Genet. Mol. Biol.
**2017**, 16, 243–257. [Google Scholar] [CrossRef] - Dower, G.E.; Yakush, A.; Nazzal, S.B.; Jutzy, R.V.; Ruiz, C.E. Deriving the 12-lead electrocardiogram from four (EASI) electrodes. Electrocardiology
**1988**, 21, S182–S187. [Google Scholar] [CrossRef] - Xie, B.; Han, M.; Zhang, P.; Zhang, W. Optimization algorithm of dynamic time warping for speech recognition of aircraft towing vehicle. J. Comput. Appl.
**2018**, 38, 1771–1776. [Google Scholar]

**Figure 1.**The reconstructed phase trajectory L

_{1}and L

_{2}, and the fused trajectory L

_{F}to illustrate the basic idea of the novel data fusion algorithm (NDFA).

**Figure 2.**Partitioning results of variables with triangular membership function. (

**a**–

**f**) are partitioning results of D, D

_{r}, O

_{d}, α, α

_{r}and ${O}_{\alpha}$ respectively.

**Figure 4.**Reconstructed phase trajectories of synthetic ECG signals. (

**a**–

**c**) are reconstructed phase trajectories of V

_{x}, V

_{y}and V

_{z}, respectively. (

**d**) Is reconstructed phase trajectory of final fused result via NDFA.

**Figure 6.**Three-lead VCG signals by realistic 12-lead ECG signals transformation and reconstructed trajectories of fused results. (

**a**–

**d**) are results of the signal of No.1027085, No.1075113, No.1027085, and No.1075113, respectively.

**Figure 7.**Reconstructed phase trajectories of ECG signals (No.106). (

**a**) and (

**b**) are reconstructed phase trajectories of lead MLII and lead V5, respectively. (

**c**) is reconstructed phase trajectory of final fused result via NDFA.

**Figure 8.**Performance comparison of NDFA, MDF, Algorithms 1 and 2 by DTW. (

**a**) Is the comparison for the fused signals by four algorithms and the lead MLII. (

**b**) Is the comparison for the fused signals by four algorithms and the lead V5.

**Table 1.**Inference rules of fuzzy inference system (FIS

_{d}) on the evolutionary trend of reconstruction trajectory.

D | ||||

S_{D} | M_{D} | B_{D} | ||

${\mathit{S}}_{{\mathit{D}}_{\mathit{r}}}$ | ${\mathit{S}}_{{\mathit{O}}_{\mathit{d}}}$ | $\mathit{S}{\mathit{R}}_{{\mathit{O}}_{\mathit{d}}}$ | ${\mathit{M}}_{{\mathit{O}}_{\mathit{d}}}$ | |

D_{r} | ${\mathit{M}}_{{\mathit{D}}_{\mathit{r}}}$ | $\mathit{S}{\mathit{R}}_{{\mathit{O}}_{\mathit{d}}}$ | ${\mathit{M}}_{{\mathit{O}}_{\mathit{d}}}$ | $\mathit{B}{\mathit{R}}_{{\mathit{O}}_{\mathit{d}}}$ |

${\mathit{B}}_{{\mathit{D}}_{\mathit{r}}}$ | ${\mathit{M}}_{{\mathit{O}}_{\mathit{d}}}$ | $\mathit{B}{\mathit{R}}_{{\mathit{O}}_{\mathit{d}}}$ | ${\mathit{B}}_{{\mathit{O}}_{\mathit{d}}}$ |

**Table 2.**Inference rules of fuzzy inference system ($\mathrm{FI}{\mathrm{S}}_{\alpha}$) on the angle of adjacent vectors.

$\mathit{\alpha}$ | ||||||

$\mathit{N}{\mathit{B}}_{\mathit{\alpha}}$ | $\mathit{N}{\mathit{M}}_{\mathit{\alpha}}$ | ${\mathit{Z}}_{\mathit{\alpha}}$ | $\mathit{P}{\mathit{M}}_{\mathit{\alpha}}$ | $\mathit{P}{\mathit{B}}_{\mathit{\alpha}}$ | ||

${\mathit{S}}_{{\mathit{\alpha}}_{\mathit{r}}}$ | $\mathit{V}{\mathit{B}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | ${\mathit{B}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | $\mathit{B}{\mathit{R}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | $\mathit{M}{\mathit{R}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | ${\mathit{M}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | |

${\mathit{\alpha}}_{\mathit{r}}$ | ${\mathit{M}}_{{\mathit{\alpha}}_{\mathit{r}}}$ | ${\mathit{B}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | $\mathit{B}{\mathit{R}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | $\mathit{M}{\mathit{R}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | ${\mathit{M}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | $\mathit{S}{\mathit{R}}_{{\mathit{o}}_{\mathit{\alpha}}}$ |

${\mathit{B}}_{{\mathit{\alpha}}_{\mathit{r}}}$ | $\mathit{B}{\mathit{R}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | $\mathit{M}{\mathit{R}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | ${\mathit{M}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | $\mathit{S}{\mathit{R}}_{{\mathit{o}}_{\mathit{\alpha}}}$ | ${\mathit{S}}_{{\mathit{o}}_{\mathit{\alpha}}}$ |

**Table 3.**SNR magnitudes for noise, baseline wander (BW), electrode movement (EM) and muscle artifact (MA).

SNR Levels (dB) | ||||
---|---|---|---|---|

BW | 12 | 6 | 0 | −6 |

EM | 6 | 0 | −6 | −12 |

MA | 12 | 6 | 0 | −6 |

SNR (dB) | 12 | 6 | 0 | −6 |
---|---|---|---|---|

$Co{r}_{\overline{x}}$ | 0.3449 | 0.3103 | 0.2340 | 0.1368 |

$Co{r}_{\overline{s}}$ | 0.5369 | 0.5329 | 0.5177 | 0.4566 |

SNR (dB) | 6 | 0 | −6 | −12 |
---|---|---|---|---|

$Co{r}_{\overline{x}}$ | 0.3281 | 0.2598 | 0.1665 | 0.0944 |

$Co{r}_{\overline{s}}$ | 0.5347 | 0.5156 | 0.4728 | 0.4080 |

SNR (dB) | 12 | 6 | 0 | −6 |
---|---|---|---|---|

$Co{r}_{\overline{x}}$ | 0.3498 | 0.3192 | 0.2476 | 0.1539 |

$Co{r}_{\overline{s}}$ | 0.5439 | 0.5221 | 0.4970 | 0.4379 |

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Diao, C.; Wang, B.; Cai, N.
Data Fusion of Multivariate Time Series: Application to Noisy 12-Lead ECG Signals. *Appl. Sci.* **2019**, *9*, 105.
https://doi.org/10.3390/app9010105

**AMA Style**

Diao C, Wang B, Cai N.
Data Fusion of Multivariate Time Series: Application to Noisy 12-Lead ECG Signals. *Applied Sciences*. 2019; 9(1):105.
https://doi.org/10.3390/app9010105

**Chicago/Turabian Style**

Diao, Chen, Bin Wang, and Ning Cai.
2019. "Data Fusion of Multivariate Time Series: Application to Noisy 12-Lead ECG Signals" *Applied Sciences* 9, no. 1: 105.
https://doi.org/10.3390/app9010105