The term segmentation means splitting the PCG signal into cardiac cycle and detection of main event (S1,S2,murmur) and intervals (systole, diastole) in each cycle [1-2]. Early diagnosis of abnormality of heart PCG signals segmentation is very important and any disease that may come out in connection with heart has critical value. Reason for segmentation:

• It works as tool to assist doctors in diagnosing heart malfunctioning very accurately.
• Prevent mistakes in diagnoses.
• Improve quality of care to patients.

Heart sounds are produced because of mechanical vibration due to contraction and relaxation of heart cavities. The mechanical systems consist of contraction phase of ventricle called systole and that of filling phase called diastole.

Heart sound signals are mostly comprised of four sound classes:-

• S1,S2-Two outstanding sounds named as first sound and second heart sounds.
• S3,S4-Two weak sounds named as the third and fourth heart sounds.

Wavelet analysis is the best for the analysis of highly non stationary signals that contain sudden picks and discontinuities, but the best search algorithm uses wavelet packets in which signals are expressed as combination of atoms .These atoms are extracted as a result of dilation of the analyzing function and organized into dictionaries as wavelet packet. Wavelet Decomposition which utilizes digital filters and down sampling is preferred method for the analysis of PCG signals because of time and frequency limitations.

Taking input (PCG signals), Data cleaning (using wavelet packets), features extraction (using wavelet packet), classification, Segmentation and separation of heart sound component of PCG signals are done using wavelet packet analysis.

The wavelet transform is mostly used now-a-days in digital signal processing. The advantage of wavelet is its inherent multi-resolution properties and its varying window size that are wide for low frequencies and narrow for high frequencies and are generally suitable for the application where tolerable degradation and scalability are important tasks. In the PCG signal segmentation using wavelet method, the signal to be analyzed is high pass and low pass filtered at every Decomposition level and these high and low frequency components are called Approximation (A) and Detail (D). In the wavelet analysis method, a signal is split into an approximation and detail, the approximation is then split into second level approximation and detail and the process is repeated, whereas in wavelet packet analysis, the detail as well as the approximation can be split. The wavelet packet analysis has the advantage over wavelet analysis in that the former method yield more than different ways to encode the signal and thus this method of wavelet decomposition offers a richer range of possibilities of signal analysis. Thus Discrete wavelet transform performs better for filtering of clicks and murmurs, whereas WPT gives important information for better analysis of time –frequency characteristics of the heart sound/based on synchronized Electrocardiographic (ECG) signal acquired with PCG. On the other hand, the envelope of PCG can be used to perform the segmentation. In [3] the envelope is estimated using homomorphism filtering, while in [4] it is estimated by means of Shannon normalized average energy. Moreover, Shannon energy can be used with more complex methodologies: wavelet transform in order to enhance the spectral components of S1 and S2; or with Mel-scaled filterbank and Hidden Markov Models. Another approach to segment PCG is based on time-frequency analysis, Implemented with wavelet transform or distributions belonging to the Cohen's quadratic class. That PCG signal exhibits nonlinear dynamics; this fact motivates the use of complexity measures [5], and RTS [6]. In this work, the authors combine the best of these methodologies with the aim of developing a reliable algorithm capable of segmenting the PCG signal with high accuracy.

In this paper, the authors propose a method for the boundary identification of S1 and S2, using the ECG signal as reference in order to separate each beat according to the Methodology proposed in [7]; later, the end of S1, beginning of S2 and end of S2 are detected using RTS and threshold rules. The results obtained in the previous stage are validated using biomedical features; if the fiducially points are out of certain boundaries given by physiological considerations, an alternative segmentation method is used. It is based on wavelet decomposition, Shannon energy and threshold rules. The algorithms are tested in a PCG database labeled by expert cardiologists in the beginning and end of S1 and S2. The main PCG signal processing stages for automatic detection of the heart valve disorders include noise removal from the raw PCG signals, segmentation, feature extraction and the classification. The undesired noises can be filtered out using methods as described in [3]. The segmentation of PCG signals without reference ECG is generally an important pre-processing step for the automated analysis of the PCG signals. The segmentation of PCG signals using envelope- gram has been developed in [4]. Unfortunately, this method may miss peaks of S1 heart sound or S2 heart sound if their amplitude is either too small or distorted by noise. Simplicity feature based segmentation with wavelet decomposition coefficients has been proposed in [5]. The robust method for detecting boundaries of S1 heart sound and S2 heart sound has been shown in [6]. However, the method presumes the availability of method for finding heart beat cycles. The segmentation method employing high frequency signatures based envelope is provided in [7]. The erotic hidden Markov model for classification of PCG signals into four phases has been proposed in [8] . However, the method needs a time-consuming training process involving interaction between computer and a human being during the training process. The performance of the auto correlation based segmentation methods depends largely on noise and variability of heart. Figure 1 shows the Wavelet Tree Decomposition and Figure 2 shows the Wavelet Packet Decomposition Tree.

Figure 1. Wavelet Tree Decomposition

Figure 2. Wavelet Packet Decomposition Tree

WPT is an extension of DWT whereby all nodes in the tree structure are allowed to split further at each level of decomposition. With WPT, both the approximation and detail coefficients are decomposed into approximation and detail components, in comparison to DWT that decomposes only the approximation coefficients of the signal as shown in Figure 3.

Figure 3. Wavelet Tree

In the wavelet packet framework, compression and denoising ideas are exactly the same as those developed in the wavelet framework. The only difference is that wavelet packets offer a more complex and flexible analysis, because in wavelet packet analysis, the details as well as the approximations are split.

A single wavelet packet decomposition gives a lot of bases from which you can look for the best representation with respect to a design objective. This can be done by finding the "best tree" based on an entropy criterion. De-noising and compression are interesting applications of wavelet packet analysis. The wavelet packet de-noising or compression procedure involves four steps:

For a given wavelet, compute the wavelet packet decomposition of signal x at level N.

For a given entropy, compute the optimal wavelet packet tree. Of course, this step is optional. The graphical tools provide a Best Tree button for making this computation quick and easy.

For each packet (except for the approximation), select a threshold and apply thresholding to coefficients.

The graphical tools automatically provide an initial threshold based on balancing the amount of compression and retained energy. This threshold is a reasonable first approximation for most cases. However, in general, you will have to refine your threshold by trial and error so as to optimize the results to fit your particular analysis and design criteria.

The tools facilitate experimentation with different thresholds, and make it easy to alter the trade off between amount of compression and retained signal energy.

Compute wavelet packet reconstruction based on the original approximation coefficients at level N and the modified coefficients.

• Decomposition
• Computation of the best tree
• Thresholding of wavelet packet coefficients
• Reconstruction

In this example, the authors will show how you can use one dimensional wavelet packet analysis to compress and to de-noise a signal.

2.1 Segmentation and Pre-processing of PCG signal

The average murmurs frequency range is between 100Hz to 600Hz and the spectrum range of PCG signals greater than 1khz I is not useful for diagnosis. Therefore in order to compare the result as described in [8],the sampling rate of the PCG signals has been reduced by factor of 32. Then segmentation is performed using enhanced modified p-spectrum and is computationally fast version of the original p-spectrum algorithm. This method uses the inverse of empspectrum(p) which is the angle between two vectors measured as their simarity. Larger the value of (p), greater is the similarity between the vectors. In the enhanced p-spectrum, signal is said to be periodic with a fundamental period of P0,if a relatively large peak is observed at p=p0 .

Let the discrete version of the PCG signal is {x1,x2 ,x3….xN},the steps to obtain the enhanced modified p-spectrum techniques are as follows :-

(1)

• For all P values, we will make a rectangular matrix so that 2 ≤ p ≤ N/2.
• The Ө value which gives the angle between two Vector A1p and A2p is computed.
• p-Spectrum for the enhanced modification is computed for the specified range of p which is given by below equation:-
• In step 3 Cubic interpolation can be applied to in order to increase readability.
• In the enhanced modified p-spectrum the value of p corresponding to the first peak provides p0 of the signal It should be noted that the heart rate considering normal and pathological cases lies in the range from 30bpm to 250bpm. Moreover the heart rate cannot go below the 30bpm in a alive subject.

The modified range of p-after considering the clinical information (after survey) is given 660 ≤ p ≤ 5513. For sampling frequency of 1378.125Hz

2.2 Use of Continuous Wavelet Transform (CWT) method

The above proposed method used the features of the segmented heart beat cycle of the PCG signals by CWT method with Morley wavelet as the mother wavelet because it has been shown that this wavelet is more suitable for analysis of the PCG signal, the scale range of which varies from 1 to 32.

2.3 Classification of PCG Signals based on Adaptive Feature Selection and the LS-SVM

One of the major problems is the classification of heart valve disorders which is considered as multiclass classification problem.

For a hyper plane Ω a, b, only the features which provide best classification rate are selected from the h features. Figure 4 shows the Flow chart of the PCG signal classification.

Figure 4. Flow chart of the PCG signal classification

In this paper, the authors have explored the ability of CWT method for classification of the heart valve disorders with Morlet mother wavelet. The authors used Shannon energy to pick the peak above threshold and identify S1 and S2 according to the set of rules similar to those used in segmentation with envelogram. This method proved to give best result than other denoising method.

The PCG signal enveloped conveys very useful information about heart disorder, if any. So it is very important to analyze this low intensity signal carefully. Provided signal is free from non stationary noises which becomes difficult to remove and can affect very much our original reading if present.

(2)

• We need temporal information in order to classify cardiac event.
• We used Shannon energy to emphasize the medium intensity, where Shannon energy is given by:
• The Shannon is useful to denoise the original signal by use of threshold to pick up the peak segmentation using envelogram and reject the extra peak according to the statics interval.
• Then recovery of lost peak is done by lowering threshold level which identifies S1 and S2 according to the interval between adjacent peaks.
• Finally use the frequency band which contains majority os S2 and S2 power.