A noval approach for characterizatiion of cardiac abnormalities using sub band coding.

Sharma, Devendra Kumar ; Sharma, R.K. ; Yadav, Suneel 等

Introduction

Automatic detection of electrocardiogram hereby referred to as ECG signals is of great importance for diagnosis of cardiac abnormalities. In recent years, computer assisted ECG interpretation is playing an increasing role in assisting medical doctors in diagnosis and treatment of heart abnormalities, so significant amount of research has focused on the development of algorithms for accurate diagnosis of heart anomalies. The variable nature of ECG has made automated detection a challenging task .Several researchers have proposed algorithms to detect the R-peaks and measure the heart rate [2-6][10-11]. Some other authors[7-9] claim an accuracy ranging between 95%-97.65% in the detection of R-peaks.

A typical ECG signal is shown in Fig. 1.

[FIGURE 1 OMITTED]

QRS complex represents the activation of ventricles. ST segment is the portion of tracing falling between the QRS wave and the T wave. During this time the ventricular is contracting, but no electricity is flowing. Re-polarization of ventricular occurs in the T wave.

In this paper, a new algorithm is presented to detect the QRS complex. The proposed algorithm is based on filtering & processing the recorded data using sub band coding and a new method is devised to calculate the heart rate by using the R-R intervals between R-peaks. As a result it is observed that the proposed algorithm gives 98% accuracy. Another advantage of this proposed algorithm is that negative R-peaks are also detectable and it is also observed that there is no change in amplitude of R-peaks if the position of V3 &V4 leads are changed.

(A). Heart Rate Measurement

System Hardware

The signal sensing device used to acquire the ECG signal is characterized by it's ability to output the part of the frequency spectrum which compound the ECG while removing the undesired frequencies.

The signal from the ECG leads is applied to the input of an Instrumentation Amplifier Schemes (IAS) with high common mode rejection ratio .The amplified signal is then filtered using a high pass butterworth filter at 50MHz to diminish the baseline wandering and slow motion interference and a lowpass butterworth filter at 150Hz to diminish the EMG interference .The scheme (Fig.2.) was proposed in order to create a portable system based on a Laptop/PC where ECG-sensing device(6108,Make,BPL) are powered using 12V,0.8AH batteries incorporated in the system.

[FIGURE 2(a) OMITTED]

[FIGURE 2(b) OMITTED]

(B) Theory

Discrete Wavelet Transform

In this work the discrete wavelet transform technique has been used for processing of ECG signal. The discrete wavelet transform is implemented using the sub band coding scheme where the wavelet is high pass filter and the low pass filter is the scaling function. Each sub band is generated by the band pass filtering of the input The level of decomposition depends upon the requirement, where H0 & L0 are the High Pass & Low Pass Filter respectively. The wavelet decomposition is shown in Fig.3.

[FIGURE 3 OMITTED]

Quadratic Spline Wavelet

The wavelet filter provides multi-resolution property and permits the inspection of characteristic waves of the ECG signals at different scale with different frequency range.

The wavelet used in the proposed algorithm is Quadratic Spline Wavelet .The expressions for Quadratic wavelet in frequency domain are as follows,

Y (w) = iw [(sin(w/4)/(w/4)).sup.4] (1)

L (w) = [e.sup.iw/2] [(cos(w/2)).sup.3] (2)

H (w) = i4 [e.sup.iw/2] (sin(w/2)) (3)

Y(w) is the expression of mother wavelet.

H(w) is the expression of high pass filter.

L(w) is the expression of high pass filter.

The response of quadratic spline wavelet is shown in Fig. 4.

[FIGURE 4 OMITTED]

Proposed Algorithm for Beat Detection Algorithm

Pan J, Tompkins [1] has presented an algorithm to detect a real time QRS complex. In this paper a new algorithm based on sub band coding is presented as per Fig. 5 to detects the R-Peaks and a new method is presented to calculate the heart rate.

The first wavelet of ECG signal shows a large change in the amplitude of R wave due to it's high frequency nature while the low frequency P and T waves have no significant effect on the first wavelet. So, for efficient QRS detection first wavelet transform is used in the presented algorithm.

The ECG signal after passing through the filter is shown in Fig.6(i). The local peaks are observed in the filtered waveform as shown in Fig 6(ii). The filter eliminates motion artifacts and also the P wave and T wave.

[FIGURE 5 OMITTED]

Mathematical Analysis

After the signal has been filtered, it is then differentiated as shown in Fig.6(iii) through a five point derivative to obtained information on slope of QRS complex and to overcome the baseline drift problem. It also helps to accentuates QRS complex relative to P and T wave. In this algorithm five point derivative is used to optimized the response.

The Transfer function & the difference equation of differentiator is given as

H(z) = 0.1 (-2[z.sup.-2] - [z.sup.-1] + [z.sup.1] + 2[z.sup.2]) (4)

Y(n) = 1/8 [2x(nT) + x(nT-T) - x(nT - 3T) - 2x(nT -4T)} (5)

Now the differentiated signal is passed through a squaring system which makes all data positive as shown in Fig.6(iv).

Y(n) = [[x(n)].sup.2] (6)

Now the detection of R-Peaks every 150ms[1] is done by moving average filter which acts as a smoother as shown in fig.6(v).The integrator is mathematically given as--

Y(nT)=((x[nT - (N - 1)T] + x[nT - (N - 1)T + ... + x[nT])/N (7)

[FIGURE 6 OMITTED]

One usual approach [1] to count the number of beats per minute is HEART RATE = r * samfreq * 60/s beats /min.

s = number of sample on which operation are performed

r = number of peaks obtained in s samples

samfreq = sampling frequency at which ECG signal is sampled.

In this paper a new formula is devised which is given in equation (8) to calculate the heart rate. The concept is drawn from the usual approach adopted by medical professionals in which the paper role in the ECG instrument moves at the rate of 25mm/sec.

HEART RATE = 1500/R - R interval beats/min (8)

On the basis of calculated heart rate using equation (8), the heart related diseases can be characterized. Matlab 7.2 is used to implement the proposed algorithm. The different cardiac abnormalities calculated so far on the basis of heart rate are shown in Table 1.

Results

The equation (9) computes the accuracy of presented algorithm.

Accuracy = 1 - (F_N + F_P/N) (9)

F_P = No. of false negative misdetection.

F_N = No. of false positive misdetection.

A comparison is done with MIT-BIH database by taking 10 test samples as in Table 2.

Conclusion

A beat detection algorithm is proposed in this paper provides the accuracy more than 98%. After detecting the R-wave, heart rate is calculated and the heart related diseases are characterized in Table 1.

In the proposed algorithm, negative R-peaks are also detectable. Further it is also observed that there is no change in amplitude of R-wave if the position of V3&V4 leads are changed.

The results from proposed algorithm are compared with other algorithms given by Lee, et al.[7] , Haque,et al[8], and A.Alexandridi [9] to achieve 95% and 97.65%, 97.6% accuracy respectively.

References

[1] PAN J, Tompkins W, "A real time QRS detection algorithm." IEEE Trans. Biom. Eng. 1985; 32(3),230-236.

[2] Y.H.HU, W. J. Tompkins, J.L. Urrusti and V. X. Afonso "Application of artificial neural network for ECG signal detection and classification", J.Electrocardiography.vol.26(suppl.) pp.66-73,1993.

[3] G. Vijaya, V. kumar and H.K. Verma , "ANN based QRS-complex analysis of ECG" J.Med.Eng Technol, vol. 22 no.4, pp.160-167,1998.

[4] R. Poli, S. Cagnoni and G. valli, "Genetic design of optimum linear and nonlinear QRS detectors", IEEE trans. Eng. Vol. 42, pp.137-1141, 1995.

[5] L. Szi Lagyi, Z.Benyo, S. M. Szi Lagyi , A. szlavecz and nagy, "On-line QRS detection using wavelet filtering" proc .of 23rd Ann Embs int. conf. oct 25,2001,Istanbul, Turkey.

[6] F. Gritzali, "Towards a generalized scheme for QRS detection in ECG waveform", signal processing vol. 15 pp.183-192,1998.

[7] Lee Ren-Guey et al., "A Novel QRS Detection algorithm applied to the analysis of heart rate variability of patients with sleep apnea", Biomedical Eng.Application, Basis & communication, vol.17 No. 5 october 2005.

[8] Haque M.A., Rahman M.E., "A fast algorithm in detecting ECG characteristic points", second international conf. on electrical & computer Engineering ICECE 2002,26-28 December 2002,Dhaka,Bangladesh.

[9] A. Alexandridi, I. Panagopoulos, G. Manis, G. Panakonstantinou, "R-Peak detection with alternative haar wavelet filter"

[10] Arzeno N.M, Poon C.S, and Deng Z.D,2006 " Quantitative analysis of QRS detection algorithms based on first derivative of the ECG", Proc. Of the 28th IEEE EMBS Ann. Int.Conf. New York City,.USA.

[11] Dinh H.A.N, Kumar D.K., Pah N.D and Burton P,2001, "Wavelets for QRS detection", Proceedings of 23th IEEE Int. Conf. of the Eng.in Med. And Biology Society, Vol.2,1883-1887.

Devendra Kumar Sharma (1), R. K. Sharma (2) and Suneel Yadav (3)

(1,3) Deptt. of Electronics & Communication, Meerut Institute of Engineering and technology, Meerut 250005, India

(2) Deptt. Of Electronics & Communication Engg., National Institute of technology, Kurukshetra, India E-mail: d_k_s1970@yahoo.co.in (1); mail2drrks@yahoo.com (2); and yadavsuneel_31@yahoo.co.in (3)

Table 1: Relation of heart rate with the cardiac abnormalities.

HEART RATE
(beats /min)   DISEASES

<60bpm         Bradycardia
70-72 bpm      Normal Heart Rate
>100 bpm       Tachycardia but at the boundary line
               between the Tachycardia & Normal Heart Rate.
110-270 bpm    Tachycardia
270-300 bpm    Severe Tachycardia
>300 bpm       Fibrillation (may be Atrial Fibrillation
               or Ventricle Fibrillation)

Table 2: Comparison of 10 test samples with MIT BIH database.

S.     Record   Total    Detected   F_P     F_N     Accuracy
No.    #        Beats    Beats                      (%)

1.     100      7        7          0       0       100%
2.     101      6        6          0       0       100%
3.     103      6        6          0       0       100%
4.     106      7        7          0       0       100%
5.     113      5        5          0       0       100%
6.     115      5        5          0       0       100%
7.     119      6        6          0       0       100%
8.     201      10       9          0       1       90%
9.     202      5        5          0       0       100%
10     203      10       9          0       1       90%
Average accuracy of all samples                     98.0%