^{1}

^{*}

^{1}

^{*}

The objective of this paper is to develop an efficient P wave detection method in electrocardiogram (ECG) using the local entropy criterion (EC) and wavelet transform (WT) modulus maxima. The detection of P wave relates to the diagnosis of many heart diseases and it is also a difficult point during the ECG signal detection. Determining the position of a P-wave is complicated due to the low amplitude, the ambiguous and changing form of the complex. In a first step, QRS complexes are detected using the pan-Tompkins method. Then, we look for the best position of the analysis window and the value of the most appropriate width to the P wave. Finally, the determination of P wave peaks, as well as their onsets and offsets. The method has been validated using ECG-recordings with a wide variety of P-wave morphologies from MIT-BIH Arrhythmia and QT database. The P-wave method obtains a sensitivity of 99.87% and a positive predictivity of 98.04% over the MIT-BIH Arrhythmia, while for the QT, sensitivity and predictivity over 99.8% are attained.

The analysis of the ECG is widely used for diagnosing many cardiac diseases, which are the main cause of mortality in developed countries. Since most of the clinically useful information in the ECG is found in the intervals and amplitudes defined by its significant points (characteristic wave peaks and boundaries), the development of accurate and robust methods for automatic ECG delineation is a subject of major importance, especially for the analysis of long recordings. In order to interpret the ECG and use it to diagnose abnormalities, it is important to know the normal characteristics of the ECG, and understand the mechanisms underlying the generation of each segment of the ECG.

The P wave is caused by atrial depolarization. The duration is normally not greater than 120 ms. The normal shape of the P wave does not include any notches or peaks. It can be positive, negative, or biphasic in the remaining leads. An absent P wave in the ECG may signify sinoatrial block [

We present here a new method for P wave’s location combines entropic criterion and wavelets transform (ECWT). We are interested in entropy as a measure of the amount of information contained in a signal. The amount of information can also be interpreted as the unpredictability of events in a signal. More events are unpredictable, the more information it contains is important and its entropy is high. This measure is favorable to the segmentation or an unpredictable event will be proportional to the increase in entropy. The entropic criterion can indicate the maximum of P wave’s energy.

The wavelet transform provides a description of the signal in the time-scale domain, allowing the representation of the temporal features of a signal at different resolutions; therefore, it is a suitable tool to analyze the ECG signal, which is characterized by a cyclic occurrence of patterns with different frequency content (QRS complexes, P and T waves). Moreover, the noise and artifacts affecting the ECG signal also appear at different frequency bands, thus having different contribution at the various scales.

For the singularity detection through local maxima, we rely on the approach Mallet and Hwang [

The paper is organized as follows: in Section 2, we present the theory of entropic local criterion and the notion of wavelets transforms modulus maxima. We show the detection method of P wave. Onsets and ends are performed in Section 3. Then, we interpret the results of P wave detection. The results of the validation on two databases (MIT-BIH, QT) and their comparison to other algorithms are given in Section 4. Finally, the conclusion is presented in Section 5.

This article proposes a method combining entropic criterion and Wavelet Transform Modulus Maxima to detect P wave. The two theories are introduced in the following.

In the case of the representation of the signal, entropy comes to measuring the fineness and precision of the location information, location in both time and frequency. Entropy will allow us to see where the energy signal is localized in the time-frequency plane. Specifically, it will allow us to know what time-frequency representation allows the best location information in the plan. The entropy associated with a discrete scalar random variable

with X achievements

Since entropy operator will be applied to a channel signal

lized to 1, so that:

With X[k] the Discrete Fourier Transform DFT of x[n] for

For any signal

timate with the following definition:

where

Most of the information in a signal is carried by its irregular structures and its transient phenomena, called singularities. A method that excels in finding and identifying these singularities is the Wavelet Transform; because of its capability of decomposing a signal into elementary building blocks that are well localized in both time and frequency. Because of this capability, the Wavelet Transform is capable of defining the local regularity of a signal. The local regularity of a function is often measured with the Lipschitz exponents [

・ We note a local extremum any point x_{0} such that _{0}, when x varies.

・ We note a modulus maximum; any point x_{0} such that _{0}, and _{0}.

・ We call maxima line, any connected curve in the scale space x along which all points are modulus maxima.

Based on the waveform, Laurent Clavier [

・ Class 1: the P wave has a relatively symmetrical reflecting normal patient.

・ Class 2: P wave has a normal climb, but relatively slow down.

・ Class 3: the P wave has a relatively slow rise but a normal descent.

・ Class 4: the P wave has a double hump.

・ Class 5: P wave is biphasic; usually it is a pathological wave.

The detection algorithm of P wave is generally designed according to three steps (

The first is based on a preliminary localization of the QRS complex and consists in seeking, in a window preceding the QRS, a trace of the P wave. The second step is based on the research for an analysis window well adapted to the width and to the position of the P wave by means of the local entropy criterion which measures the temporal unpredictability and disorder in the frequency signals and is an index for the segmentation. As part of the segmentation, the entropy changes reflect the presence of a change in the signal. The entropy is going to help us to adopt the width of the window to that of the P wave. In this case, our research becomes more efficient and more objective. By the way, we will gain in computing time for the next step of this method. We succeed and optimization the algorithm. The third is based on the detection of the P wave by modulus maxim’s method.

The ECG signal gotten at the time of the registration is contaminated generally by different sources of noises that can disrupt the phase and the amplitude characteristics of the useful signal, from where the necessity of a

good filtering [

In general, the normal ECG rhythm means that there is a regular rhythm and waveform. However, the ECG rhythm of the patient with arrhythmia will not be regular in certain QRS complexes. We use the different characteristics that arrhythmias exhibit to detect the abnormal ECG waveform. We first need to find the location of every QRS complex.

The locations of QRS complex have the maximum variation in the slopes. This property was used to detect the location of QRS complex. The method of Pan-Tompkins was adopted to detect the QRS complex [

The P wave is a rounded peak occurred before the QRS complex. Therefore, the P wave can be found based on the location of the QRS complex. The method consists in finding the right and better position of the analysis window (

The detection algorithm P-wave refers to the R peak backwards t_{r} (tRpeak-200 ms) so as to be in advance phase by report to the QRS complex (

t_{r} depends on the RR period in 5 seconds. Statistically, over a period for many patients and healthy on a recording time of one hour for each, P wave holds 10%, the QRS complex occupies 8% and the T wave occupies 20% of the RR period. The choice of the window is taken of such way as the energy in the side of lobes being as small as possible to minimize the errors in the entropy and increase the accuracy of positional. We perform the decomposition of the ECG signal for 4 levels (2^{1}, 2^{2}, 2^{3} and 2^{4}) in which his higher energy is. For scales larger than 2^{4}, the energy of the QRS complex is very low, only the P and T waves have significant components.

The detection of the P wave is based on the variations in local entropy rather than the entropy itself. It uses the local entropy criterion. Entropic Local Criterion is a time function, measuring variations in the energy concentration of the short-term spectrum of the signal. The measurement of concentration of energy or spectral disorder is used Shannon [

We set the first W_{i} analysis window, which Li its width, from tr. For that, we calculate the normalized and localized entropy of analysis window W_{i} (L_{i}) for 2^{4} scale according to the Equation (4). The value of entropy H_{i} calculated through the W_{i }window is saved in a table.

The search for the best width of the window operates according to an iterative algorithm. In every iteration the width of the window is multiply by 2 (L_{i} × 2) to look for the maximum of quantity of information contained in the wave P. The same steps are applied on the next analysis window W_{i+1} with L_{i+1 }= L_{i} × 2. So, we create an increasing dyadic table for the window’s various values (

By comparing the last two values of the found entropies, we start again calculating next entropy for the window W_{i+1}untilthe value of two successive entropy (H_{n+1}, H_{n}) remains constant. We retain the window W_{n }of L_{n}

Windows W_{i} | Width L_{i} |
---|---|

W_{0} | L_{0} |

W_{1} | L_{1} = L_{0} × 2 |

W_{2} | L_{2} = L_{1} × 2 |

W_{3} | L_{3} = L_{2} × 2 |

W_{4} | L_{4} = L_{3} × 2 |

W_{5} | L_{5} = L_{4} × 2 |

width.

_{0}, W_{1}, W_{2}, W_{3}) showing the growth of their width. We note that the value of the entropy increases when the width of the window widens at each iteration since the

detection of the information is more and more centralized on the P wave.

At this level of the algorithm, we have a window (W_{n}) adapted in width and in position to the P wave. Once the analysis window is fixed, we calculate the modulus maxima for the 2^{4 }scale to detect the position of the P peak applied to the W_{n}. Mainly at scales 2^{4}, P wave has its major component, whereas artifacts produce isolated maximum or minimum lines which can be easily discarded. If the signal is contaminated with high-frequency noise, the most affected scales are and, being higher scales essentially immune to this sort of noise.

Using a first derivative Gaussian function as mother wavelet due to having one zero moment. It is therefore well adapted to detect the amplitude discontinuities in the signal; it delivers more to the spread of the maximum fine scales [

The waves produce a positive maximum-negative minimum pair along the scales, with a zero crossing between them. Each sharp change in the signal is associated to a line of maxima or minima across the scales. The passages by zero or zeros-crossings of the wavelet transform correspond in maxima premises or minima of the signal smoothed in the various scales and the maximal absolute values of the transformed it waves are associated with the maximal slopes. Using the information of local maxima, minima and zero crossings at 2^{4} scales, the algorithm identifies the significant points P wave detection and delineation.

Our detection method of P wave is also responsible for detecting P_{on}/P_{off} which is the start and end of the P waves. Our algorithm comprises two steps defined as follows (_{peak}.

The onset of P wave is before the first slope of the position of P peak. In the same way, the offset of P wave is after the last slope of the position of P peak. We calculate the modulus maxima of the (first/last) slope of the P peak’s position for the 2^{4} scale. We considered the modulus maxima calculated for the analysis window W_{n} already attached to the P peak search step.

The ECG signals under test of the transformations are taken from the website of MIT-BIH Arrhythmia Database [

The ECWT based method only considers variations of the signal that has a certain resemblance with the P-wave. Therefore, it is insensitive to baseline drift (

Most methods require frequently adapted thresholds in order to detect a low amplitude P-wave. In most cases these thresholds are used to distinguish the wave from the noise. By using scales that contain most part of the energy of the P-wave it is possible to acquire a precise detection (see

Most methods require frequently adapted thresholds in order to detect a low amplitude P-wave. In most cases these thresholds are used to distinguish the wave from the noise (

As most methods, the ECWT method uses certain decision rules to distingue different kinds of P-waves. The difference with the other methods lies in the fact that the rules are applied to the transformation instead of to the pure signal. The transformation gives a clearer view of the signals information and therefore it is better suited for decision rules (

In some cases, the P-wave is absent. The figure below illustrates the case where the P wave is present with an important amplitude in start of signal but absent after (

The datasets are signals coming from the MIT-BIH Arrhythmia Database [

・ Number of True Positive detections (TP).

・ Number of False Positive detections (FP).

・ Number of False Negative detections (FN).

・ Percentage of detected P-waves (S_{e}).

・ Percentage of positive predictivity (P^{+})

Database | P-wave detection | Beats | TP | FP | FN | S_{e} (%) | P^{+} (%) | |||
---|---|---|---|---|---|---|---|---|---|---|

MITBIH | ECWT (this work) | 54,714 | 54,557 | 76 | 109 | 98.87 | 98.04 | |||

Pan et al. [ | 54,904 | 54,766 | 253 | 138 | 99.75 | 99.54 | ||||

QT | ECWT | 63,210 | 63,184 | 54 | 37 | 99.94 | 99.89 | |||

Martinez et al. [ | 64,387 | 64,395 | 57 | 41 | 99.92 | 99.88 | ||||

The detection results on the MITDB and QTDB obtained by our ECWT of P detection and other published detectors are given in _{e} = 98.87% and P^{+} = 98.04% for the MITDB. On QTDB, our algorithm presents a positive predictivity (P^{+}) about 99.89% and sensitivity (Se) about 99.94%.

We have presented in this paper a new method for P wave’s locations using the local entropic criterion, in order to focus on the maximum of P wave’s energy, and wavelet transform over some successive scales. The algorithm has been validated using two standards databases MIT-BIH Arrhythmia and QT, with different sampling rates and a wide diversity for P wave forms.

Our method achieves very good detection performance on the two studied databases. This algorithm attains S_{e} = 98.87% and P^{+} = 98.04% for the MITDB. On QTDB, it presents a positive predictivity (P^{+}) about 99.89% and sensitivity (S_{e}) about 99.94%.

SawsanRekik,NoureddinneEllouze, (2015) P-Wave Detection Combining Entropic Criterion and Wavelet Transform. Journal of Signal and Information Processing,06,217-226. doi: 10.4236/jsip.2015.63020