Journal List > J Korean Soc Med Inform > v.15(1) > 1035514

TOOLS
Park, Cho, Lee, Song, Lee, Chee, Kim, and Kim: Hierarchical Classification of ECG Beat Using Higher Order Statistics and Hermite Model
Original Article

J Kor Soc Med Informatics 2009;15(1):117-131.

Published online: 13 January 2009

DOI: https://doi.org/10.4258/jksmi.2009.15.1.117

Hierarchical Classification of ECG Beat Using Higher Order Statistics and Hermite Model

Kwan Soo Park1, Baek Hwan Cho3, Do Hoon Lee1, Su Hwa Song1, Jong Shill Lee1, Young Joon Chee2, In Young Kim1, Sun I. Kim1

1Dept. of Biomedical Engineering, Hanyang Univ.

2Dept. of Biomedical Engineering, College of Engineering, University of Ulsan

3Dept. of Biomedical Engineering, Carnegie Mellon Univ.

Tel: 02-2291-1713, Fax: 02-2296-5943, E-mail: iykim@hanyang.ac.kr

Copyright © 2009 The Korean Society for Surgery of the Hand

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Objective:

The heartbeat classification of the electrocardiogram is important in cardiac disease diagnosis. For detecting QRS complex, conventional detection algorithm have been designed to detect P, QRS, T wave, first. However, the detection of the P and T wave is difficult because their amplitudes are relatively low, and occasionally they are included in noise. Furthermore the conventional multiclass classification method may have skewed results to the majority class, because of unbalanced data distribution. Methods: The Hermite model of the higher order statistics is good characterization methods for recognizing morphological QRS complex. We applied three morphological feature extraction methods for detecting QRS complex: higher-order statistics, Hermite basis functions and Hermite model of the higher order statistics. Hierarchical scheme tackle the unbalanced data distribution problem. We also employed a hierarchical classification method using support vector machines.

Results:

We compared classification methods with feature extraction methods. As a result, our mean values of sensitivity for hierarchical classification method (75.47%, 76.16% and 81.21%) give better performance than the conventional multiclass classification method (46.16%). In addition, the Hermite model of the higher order statistics gave the best results compared to the higher order statistics and the Hermite basis functions in the hierarchical classification method.

Conclusion:

This research suggests that the Hermite model of the higher order statistics is feasible for heartbeat feature extraction. The hierarchical classification is also feasible for heartbeat classification tasks that have the unbalanced data distribution.

Keywords: Electrocardiogram, Higher Order Statistics, Hermite Basis Function, Support Vector Machine, Hierarchical classification

REFERENCES

1. Hu YH, Palreddy S, Tompkins W. A patient adaptable ECG beat classifier using a mixture of experts approach. IEEE Trans Biomed Eng. 2007; 44:891–900.
2. Chazal P, O’Dwyer M, Reilly R. Automatic Classification of Heartbeats Using ECG Morphology and Heartbeat Interval Features. IEEE Trans Biomed Eng. 2004; 51:1196–1206.
3. Osowski S, Hoai LT, arkiewicz T. Support vector machine-based expert system for reliable heartbeat recognition. IEEE Trans Biomed Eng. 2004; 51:582–89.
crossref
4. Nikias C, Petropulu A. Higher Order Spectral Analysis. Prentice-Hall;1993.
5. Lagerholm M, Peterson C, Braccini G, Edenbrandt L, Sornmo L. Clustering ECG complexes using Hermite functions and self-organizing maps. IEEE Trans Biomed Eng. 2000; 47:838–47.
crossref
6. Vapnik V. Statistical Learning Theory. Wiley;1998.
7. Kreel U. Advances in Kernel Methods: Support Vector Learnings. MIT Press;1999. p. 255–68.
8. Casasent D, Wang YC. A hierarchical classifier using new support vector machines for automatic target recognition. Neural networks. 2005; 18:541–48.
crossref
9. Elkan C. The foundation of cost-sensitive learning. Proceeding of the 17th International Joint Conference on Artificial Intelligence. 2001; Aug 4-10; Seattle, USA.
10. Yang H, Carlin D. ROC surface: a generalization of ROC curve analysis. Journal of biopharmaceutical statistics. 2000; 10:183–96.
crossref
11. Mark R Mody G. MIT-BIH Arrhythmia Database. 1997. Online available at. http://www.physionet.org/. Accessed November 11, 2014.
12. American National Standard. Testing and reporting performance results of cardiac. ANSI/AAMI EC57 : 1998/(R)2003.
13. Sansone G. Orthogonal Functions. Dover;1991.
14. Golub G, Van Loan C. Matrix Computations. Johns Hopkins University;1991.
15. Kim Y, Kang S, Park I, Noh G. Population Pharmacokinetic and Pharmacodynamic Models of Propofol in Healthy Volunteers using NONMEM and Machine Learning Methods. Journal of Korean Society of Medical Informatics. 2008; 14(Suppl 2):147–159.
crossref
16. Cho B, Lee J, Chee Y, Kim K, Kim I, Kim S. Prediction of Diabetic Nephropathy from Diabetes Database Using Feature Selection Methods and SVM Learning. Korea Society of Medical & Biological Eng. 2007; 28:255–62.
17. Lin HT, Lin CJ, Weng RC. A note on Platt’s probabilistic outputs for support vector machines. Taiwan Univ.;2003.

Figure 1.
The procedure for classification of heartbeat class from electrocardiogram
jksmi-15-117f1.tif
Figure 2.
(a) The N class waveform of ECG beat and its cumulant representations of the (b) second, (c) third, and (d) fourth order
jksmi-15-117f2.tif
Figure 3.
(a) The V class waveform of ECG beat and its cumulant representations of the (b) second, (c) third, and (d) fourth order
jksmi-15-117f3.tif
Figure 4.
(a) The F class waveform of ECG beat and its cumulant representations of the (b) second, (c) third, and (d) fourth order
jksmi-15-117f4.tif
Figure 5.
For σ=0.8, (a) n=1, (b) n=2, (c) n=7, and (d) n=20 Hermite basis function
jksmi-15-117f5.tif
Figure 6.
The normal QRS complex of an ECG beat (black solid line) and its estimation using a Hermite polynomial of the 20th order(white dashed line)
jksmi-15-117f6.tif
Figure 7.
Hierarchical classification is that the ECG beat classes were combined into two classes based on the morphological similarity at first phase. The second phase classifies individual classes
jksmi-15-117f7.tif
Figure 8.
Division of validation set (DS1) into training and testing set for classifier evaluation using 11 fold cross validation. Final performance evaluation is performed to test testing set (DS2).
jksmi-15-117f8.tif
Figure 9.
ROC curve of the best classifier. The arrows indicate sensitivity 90% point (Sen_90), specificity 90% point (Spe_90), and minimum distance point (Msp) between ROC curve and (1,0) point
jksmi-15-117f9.tif
Table 1.
Merging the MIT-BIH heartbeat types to the AAMI heartbeat classes
AAMI heartbeat class Description MIT-BIT heartbeat types
N Any heartbeat not in the S, V, F or Q classes normal beat (NOR)
left bundle branch block beat (LBBB)
right bundle branch block beat (RBBB)
atrial escape beat (AE)
nodal (junctional) escape beat (NE)
S Supraventricular ectopic beat atrial premature beat (AP)
aberrated atrial premature beat (aAP)
nodal (junctional) premature beat (NP)
supraventricular premature beat (SP)
V Ventricular ectopic beat premature ventricular contraction (PVC)
ventricular escape beat (VE)
F Fusion beat fusion of ventricular and normal beat (fVN)
Q Unknown Beat paced beat (P)
fusion of paced and normal beat (fPN)
unclassifiable beat (U)
Table 2.
Organized Dataset 1(DS1) and Dataset 2(DS2) from MIT-BIH dataset
N S V F Total
DS1 45,868 943 4,259 415 51,013
(Ratio, %) (89.08) (1.83) (8.24) (0.81) (100)
DS2 44,259 1,837 3,221 388 49,705
(Ratio, %) (89.03) (3.71) (6.48) (0.78) (100)
Table 3.
Class ratio of DS1 and sampled DS1
N S V F Total
DS1 45,868 943 4,259 415 51,493
(Ratio, %) (89.08) (1.83) (8.24) (0.81) (100)
Sampled DS1 9,172 194 854 86 10,306
(Ratio, %) (89.00) (1.88) (8.29) (0.83) (100)
Table 4.
Result of DS1 using conventional multiclass classification by linear and RBF kernel
Kernel
 Linear
 Radial
Parameter HOS* HBF† HMH‡ HOS HBF HMH
N Sensitivity 99.11 91.39 98.58 98.82 87.70 98.07
S Sensitivity 0.00 1.70 0.00 0.00 0.85 0.00
V Sensitivity 58.07 27.70 59.23 65.25 70.02 64.22
F Sensitivity 0.24 0.24 0.24 1.44 0.96 1.45
Accuracy 93.43 84.26 93.04 93.71 84.08 92.96
Mean of Sensitivity 39.36 30.26 39.51 41.38 39.88 40.94
+P§ of S 0.00 0.99 0.00 0.00 0.44 0.00
+P§ of V 65.54 34.90 62.72 78.20 39.88 78.78

* HOS:

† HBF: Hermite basis function

Table 5.
Result of testing DS2 using conventional multiclass classification
Parameter Result
N Sensitivity 99.65
S Sensitivity 0.00
V Sensitivity 84.48
F Sensitivity 0.52
Accuracy 94.21
Mean of Sensitivity value 46.16
+P*ofS 0.00
+P*ofV 72.37
Table 6.
Sensitivity, specificity, accuracy and threshold of each arrow points in NS vs. VF ROC curve
Parameter Sen_90* Msp† Spe_90‡
Sensitivity 90 84 74
Specificity 78 86 90
Accuracy 79 86 88
Threshold 0.955 0.925 0.835
Table 7.
AUC value and kernel by feature extraction method
Phase Class Parameter HOS* HBF† HMH‡
1ststep NS vs. VF AUC Kernel 0.909 Radial 0.893 Radial 0.899 Radial
2ndstep N vs. S AUC Kernel Linear 0.897∮ Linear Linear
V vs. F AUC Kernel 0.770 Linear 0.656 Linear 0.815 Linear
Table 8.
Classification performance of hierarchical classification and multiclass classification on DS2
Parameter DS2_multi* DS2_HOS† DS2_HBF‡ DS2_HMH∮
N Sensitivity 99.65 81.23 86.25 82.98
S Sensitivity 0.00 57.65 82.63 75.23
V Sensitivity 84.48 83.11 80.88 84.17
F Sensitivity 0.52 79.90 54.90 82.47
Accuracy 94.21 80.47 85.56 82.80
Mean of Sensitivity 46.16 75.47 76.16 81.21
+P || of S 0.00 21.50 24.78 24.12
+P || of V 72.37 46.11 89.09 84.69

‡ HBF: Hermite basis function

Table 9.
Result of decision matrix on DS2 of Hermite model of higher order statistics
Actual class
N S V F
Predicted Class n 36,728 221 42 16
s 4,079 1,382 249 18
v 319 171 2,711 34
f 3,133 63 219 320
Sensitivity(%) 82.98 75.23 84.17 82.47
+P* 24.12 84.69
TOOLS
Similar articles