DESC:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENT RULES, 2003

COMPLETE SPECIFICATION
(See Section 10 and Rule 13)

Title of invention:
SYSTEMS AND METHODS FOR TIME-FREQUENCY ANALYSIS OF PHONOCARDIOGRAM FOR CLASSIFYING HEART DISEASE

Applicant:
Tata Consultancy Services Limited
A company Incorporated in India under the Companies Act, 1956
Having address:
Nirmal Building, 9th Floor,
Nariman Point, Mumbai 400021,
Maharashtra, India

The following specification particularly describes the embodiments and the manner in which it is to be performed.
CROSS-REFERENCE TO RELATED APPLICATIONS AND PRIORITY
[001] The present application claims priority from Indian provisional specification no. 201621030924 filed on 09th September, 2016, the complete disclosure of which, in its entirety is herein incorporated by references.

TECHNICAL FIELD
[002] The embodiments herein generally relates to the field of classification of heart disease, and, more particularly, to a method and system for classifying heart disease in a person using phonocardiogram (PCG) signal of the person.

BACKGROUND
[003] Heart sound signals, commonly known as phonocardiogram (PCG) is typically captured using a digital stethoscope and is known to carry useful information regarding many cardiac abnormalities. Analysis of heart sound is a popular research area for non-invasive identification of several heart diseases. Various state of the art techniques have been used in the prior art for segregating the fundamental heart sounds from raw PCGs.
[004] For identifying normal and abnormal heart sounds, the prior art techniques employ a number of steps, including pre-processing of noisy data, identification of the fundamental heart sounds, followed by feature extraction and classification. Wavelet based features and spectral features, obtained using FFT was widely used in literature to identify cardiac abnormalities. More complex features like Mel frequency Cepstral Coefficients (MFCCs) were also investigated using Hidden Markov Model. However, due to the vulnerability of PCG towards ambient noise in audible range, variation in sensor quality and the location of data acquisition, automatic classification of PCG is a challenging task till date.

SUMMARY
[005] The following presents a simplified summary of some embodiments of the disclosure in order to provide a basic understanding of the embodiments. This summary is not an extensive overview of the embodiments. It is not intended to identify key/critical elements of the embodiments or to delineate the scope of the embodiments. Its sole purpose is to present some embodiments in a simplified form as a prelude to the more detailed description that is presented below.
[006] In view of the foregoing, an embodiment herein provides a system for classifying heart disease in a person using phonocardiogram (PCG) signal of the person. The system comprises a PCG sensor, a filtering module, memory and a processor. The PCG sensor captures the PCG signal from the person. The filtering module removes the noise from the captured PCG signal. The processor further comprises a segmentation module, a feature extraction module, a selection module and a classification module. The segmentation module segments the PCG signal to get a first sound, a second sound, systole and diastole out of the PCG signal. The feature extraction module for extracts a plurality of time domain features, a plurality of frequency domain features and a plurality of time-frequency domain features from the segmented PCG signal. The a selection module chooses a methodology out of a set of classification methodologies using a training data, wherein the training data comprises sample data covering various scenarios of diseased and healthy person. The classification module classifies the person as normal or abnormal using the chosen methodology and extracted plurality of features based on a predefined criteria.
[007] In another aspect, the disclosure provides a method for classifying heart disease in a person using phonocardiogram (PCG) signal of the person. Initially, phonocardiogram (PCG) signal from the person is captured using a PCG sensor. Noise is removed from the captured PCG signal using a filtering module. In the next step, the PCG signal is segmented to get a first sound, a second sound, systole and diastole out of the PCG signal. Further, a plurality of time domain features, a plurality of frequency domain features and a plurality of time-frequency domain features are extracted from the segmented PCG signal. In the next step a methodology is chosen out of a set of classification methodologies using a training data, wherein the training data comprises sample data covering various scenarios of diseased and healthy person. And finally, the person is classified as normal or abnormal using the chosen methodology and extracted plurality of features based on a predefined criteria.
[008] It should be appreciated by those skilled in the art that any block diagram herein represent conceptual views of illustrative systems embodying the principles of the present subject matter. Similarly, it will be appreciated that any flow charts, flow diagrams, state transition diagrams, pseudo code, and the like represent various processes which may be substantially represented in computer readable medium and so executed by a computing device or processor, whether or not such computing device or processor is explicitly shown.

BRIEF DESCRIPTION OF THE DRAWINGS
[009] The embodiments herein will be better understood from the following detailed description with reference to the drawings, in which:
[0010] Fig. 1 illustrates a block diagram for classifying heart disease in a person using phonocardiogram (PCG) signal of the person according to an embodiment of the present disclosure;
[0011] Fig. 2 illustrates a graphical representation of the signal quality comparison between the long data and the short data according to an embodiment of the present disclosure;
[0012] Fig. 3 illustrated the schematic block diagram of five methodologies according to an embodiment of the disclosure; and
[0013] Fig. 4a - 4b is a flowchart illustrating the steps involved for classifying heart disease in a person using phonocardiogram (PCG) signal of the person according to an embodiment of the present disclosure.

DETAILED DESCRIPTION OF EMBODIMENTS
[0014] The embodiments herein and the various features and advantageous details thereof are explained more fully with reference to the non-limiting embodiments that are illustrated in the accompanying drawings and detailed in the following description. The examples used herein are intended merely to facilitate an understanding of ways in which the embodiments herein may be practiced and to further enable those of skill in the art to practice the embodiments herein. Accordingly, the examples should not be construed as limiting the scope of the embodiments herein.
GLOSSARY – TERMS USED IN THE EMBODIMENTS
[0015] The expression “phonocardiogram” or “PCG” in the context of the present disclosure refers to the signal captured from the person using a PCG sensor. The PCG signal typically contains two prominent heart sounds, namely S1 and S2. S1 precedes the systole whereas S2 precedes the diastole region.
[0016] Referring now to the drawings, and more particularly to Fig. 1 through Fig. 4, where similar reference characters denote corresponding features consistently throughout the figures, there are shown preferred embodiments and these embodiments are described in the context of the following exemplary system and/or method.
[0017] According to an embodiment of the disclosure, a system 100 for classifying heart disease in a person using phonocardiogram (PCG) signal of the person is shown in Fig. 1. The system 100. The system 100 uses a machine learning based approach to classify the heart sounds in normal or abnormal class. Abnormal behavior corresponds to the coronary artery disease (CAD). One of the five different methodologies can be used for the classification of the heart sounds based on a training data.
[0018] According to an embodiment of the disclosure, the system 100 consists of a phonocardiogram (PCG) sensor 102, a filtering module 104, a memory 106 and a processor 108 as shown in Fig. 1. The processor 108 is in communication with the memory 106. The processor 108 configured to execute an algorithm stored in the memory 106. The processor 108 further comprises a plurality of modules for performing various functions. The processor 108 comprises a segmentation module 110, a feature extraction module 112, a selection module 114, and a classification module 116.
[0019] According to an embodiment of the disclosure, the system 100 may also include a user interface 118. The user interface 118 is configured to provide various inputs to the system 100. The inputs can be training model data or any other related data. The user interface 118 can include a variety of software and hardware interfaces, for example, a web interface, a graphical user interface, and the like and can facilitate multiple communications within a wide variety of networks N/W and protocol types, including wired networks, for example, LAN, cable, etc., and wireless networks, such as WLAN, cellular, or satellite. In an embodiment, the user interface 118 can include one or more ports for connecting a number of devices to one another or to another server.
[0020] According to an embodiment of the disclosure, the PCG signal can be captured using the PCG sensor 102. The use of any available PCG sensor is well within the scope of this disclosure. The PCG sensor 102 is a part of the data collection unit (not shown in figures). The captured PCG signal is further digitized in a fixed sampling rate and stored for further analysis and noise removal. The filtering module 104 is configured to remove the noise from the captured PCG signal. The filtering module 104 contains series of filter for removing various types of noises from the PPG signal.
[0021] According to an embodiment of the disclosure, the segmentation module 110 is configured to segment the filtered PCG signal. Typically the PCG cycle contains two prominent heart sounds, namely S1 and S2. S1 precedes the systole whereas S2 precedes the diastole region. The filtered PCG signal is down sampled at 1000 Hz, in order to segment the four cardiac states i.e., first sound S1, second sound S2, systole and diastole using the logistic regression based Hidden Semi Markov Model (HSMM) approach.
[0022] According to an embodiment of the disclosure, the system 100 further configured to extract a plurality of features from the segmented PCG signal using the feature extraction module 112. The plurality of features include a plurality of time domain features, a plurality of frequency domain features and a plurality of time-frequency domain features. In an example of this disclosure, a total of 88 features were extracted which has been explained in detail in the later section of the disclosure.
[0023] According to an embodiment of the disclosure, the system 100 further includes the selection module 114. The selection module 114 is configured to choose a methodology out of a set of classification methodologies. The classification methodology is chosen based on a training data. The training data is generated by collating all sample data covering various scenarios of diseased and healthy person. The set of classification methodologies comprises a baseline two class classifier approach, baseline method on unbalanced recordings using the plurality of features approach, models for long and short recordings, models for unbalanced long recording, and three class classifier for noisy PCG signal. The different types of classification methodologies are shown in Fig. 3. The set of classification methodologies are explained as follows:
[0024] Method 1 - Baseline Two Class Classifier: In the baseline approach, a simple two class classifier is designed for classifying normal and abnormal heart sounds. It was observed that each subset is highly unbalanced. Hence, in order to ensure a balanced training, all instances of minor class, along with equal representation of the other member class is drawn at random from each of the subset.
[0025] Method 2 - Baseline Method on Unbalanced Recordings using all Features: Due to highly unbalanced ratio of normal and abnormal classes in each of the subsets, random under-sampling of majority class leads to removal of a significant number of observations. Hence, in method 2, the baseline method was applied on the entire corpus of unbalanced recordings.
[0026] Method 3 - Separate Models for Long and Short Recordings: In another embodiment, the system 100 also configured to determine the length of the captured PCG signal which is further used by the selection module 114 to choose the methodology for classification. It was observed in method 1 and 2 that, the overall classification accuracy obtained on the long recordings (minimum duration of more than 10 seconds) is significantly higher compared to the short recordings (duration less than 10 seconds). It was also observed that, small recordings are mostly present in one of the partitions of the entire dataset. The signal quality of the small recordings are also very poor compared to others. The following metric (Psig) was used to measure the signal quality of the PCGs.

For a noisy recording, the S1 and S2 peaks are often suppressed by the noise, induced in systole and / or diastole. Due to its prolonger duration, the diastolic portion is generally more vulnerable. Thus the ratio between diastolic power (Pdia) and power of S1 or S2 becomes comparable. Resulting in a higher value of Psig. The value of Psig should ideally be lesser for a recording of good signal quality. Fig. 2, shows the values of Psig of more than 120 long and short recordings, drawn at random from the entire dataset. It is evident that the small recordings are generally nosier than the longer recordings owing to the lesser values of Psig. It is also evident that certain features, mostly related to HRV are often not captured properly in the recordings of very shorter duration. Thus the feature lists for long and short recordings for classifying abnormal heart sounds are expected to different. So, instead of a single two class classifier for normal / abnormal heart sounds, it was decided to create two separate classifiers for long and short data respectively. Equal representations of normal and abnormal cases were selected from each of them in order to create two balanced subset. Finally, separate sets of features are selected for each cases, for designing of the classifiers.
[0027] Method 4 - Separate Models but Unbalanced Long Recordings: This is a logical extension of method 3. Similar to method 2, here all possible PCG features were explored on the entire unbalanced dataset. However, it was found that, this only improves the performance of the long recording. The performance on short recordings actually gets degraded. Thus in this method, the analysis was modified on long recordings only, by utilizing the entire feature list. The model for the short recordings remains the same as used in the previous method.
[0028] Method 5 - Three class classifier for Noisy Data: A significant portion of the entire dataset is largely corrupted due to human speech, background noise and the frictional noise generated due to the motion of the stethoscope on human body. In previous methods, all the PCG recordings were categorized into one of the normal or abnormal class regardless their signal quality. Ground truth signal quality (good / bad) for each signal was also provided along the training set in a binary form. Few recordings are very poor to analyse even by expert annotators. Thus, a three class classifier was created which provides a scope of identifying the noisy data and mark them as unsure along with classifying the rest as normal or abnormal heart sounds. Six more features were derived for the noisy signals and are combined with the previous features. The new features are - standard deviation of ratio between 1) diastolic and S1 power, 2) diastolic and S2 power, 3) mean of ratio between S1 and S2 power, median of ratio between 4) diastolic and S1 power, 5) diastolic and S2 power and 6) Kurtosis of the envelop of the auto-correlated PCG signal.
[0029] According to an embodiment of the disclosure, the system 100 further includes the classification module 116. The classification module 116 classifies the person as normal or abnormal using the methodology chosen by the selection module 114 and extracted plurality of features based on a predefined criteria.
[0030] In operation, a flowchart 200 illustrating the steps involved for classifying heart disease in the person using phonocardiogram (PCG) signal of the person is shown in Fig. 4a and 4b. Initially at step 202 phonocardiogram (PCG) signal from the person is captured using a PCG sensor. Generally the captured PCG signal contains a plurality of noises. At step 204, the plurality of noises from the PCG signal is removed using the filtering module 104. The filtering module 104 comprises a series of filters.
[0031] At step 206 the filtered PCG signal is segmented to get a first sound, a second sound, systole and diastole out of the PCG signal. In the next step 208, a plurality of time domain features, a plurality of frequency domain features and a plurality of time-frequency domain features are extracted from the segmented PCG signal. In an example, a total of 88 features have been extracted which are explained below in detail. At step 210, a methodology out of a set of classification methodologies is chosen using a training data by the selection module 114, wherein the training data comprises sample data covering various scenarios of diseased and healthy person. IN the present disclosure, five classification methodologies have been used ad follows - a baseline two class classifier approach, baseline method on unbalanced recordings using the plurality of features approach, models for long and short recordings, models for unbalanced long recording, and three class classifier for noisy PCG signal. And finally at step 212, the person is classified as normal or abnormal using the chosen methodology and extracted plurality of features based on a predefined criteria.
[0032] According to an embodiment of the disclosure, the system 100 can be explained with the help of the following example. For the experimental purpose, all the data was taken from the Physionet dataset. A total of 88 features were explored. First 20 time domain features are related to the arithmetic mean and standard deviation of the intermediate distance between different cardiac states. These features contain information regarding individual heart beat as well as heart rate variability (HRV).
[0033] Feature 21 measures the standard deviation of the successive differences between adjacent NN intervals. Feature indices 22 to 37 measures the normalized spectral power within the frequency range of 0-20 Hz, 20-40 Hz, 40-60 Hz and 60-80 Hz respectively for S1, systole, S2 and diastole regions. Features 38 to 45 are the magnitude and phase angles of the first four poles of the diastolic regions, modeled using autoregressive (AR) model. The diastolic portion is sub segmented into non-overlapping windows of 50 ms and each window is modeled with a 10th order autoregressive (AR) model. Finally the median across all the sub-segments are considered as representative feature values. Features 46 to 53 represent the mean and standard deviation of the spectral centroid across all the S1, systole, S2 and diastole segments in a signal. Features 54 to 60 represent the spectral power between 0-100 Hz in five equal frequency bands of 20 Hz as well as the mean and standard deviation of the spectral centroid for the entire spectrum for all complete cardiac cycles.
[0034] The next 26 features are the mean and standard deviation of 13 dimensional MFCC coefficients. To extract these features, the entire signal is broken into 250 ms windows with 100 ms overlapping using hamming window. The signals are analysed up to 300 Hz for extracting the coefficients. The final two features are wavelet related features, extracted from the diastolic portion. The diastolic portion is decomposed up to third level using the Reverse biorthogonal 3.9 (rbio 3.9) mother wavelet. The median values of the mean and the standard deviation of the third level detailed coefficients across all the diastolic segments are included in the feature list.
[0035] As mentioned above a total of five classification methodologies have been explored for the classification of person. The optimum feature list for each of them is selected from the exhaustive lists of 88 features by ranking them based on Maximal Information Coefficients scores.
[0036] Method 1 - Baseline two class classifier: In the baseline approach, a simple two class classifier is designed for classifying normal and abnormal heart sounds. This random under-sampling, resulted in a total of 944 recordings from the entire set. Top 31 most significant features were selected for performance evaluation. Method 2 -Baseline method on unbalanced recordings using all features: Due to highly unbalanced ratio of normal and abnormal classes in each of the subsets, random under-sampling of majority class leads to removal of a significant number of observations. Further, the analysis is done using all 88 features to mitigate the effect of possible information loss occurred due to exclusion of certain feature in method 1. Method 3 - Separate models for long and short recordings: It was observed in method 1 and 2 that, the overall classification accuracy obtained on the long recordings (minimum duration of more than 10 seconds) is significantly higher compared to the short recordings (duration less than 10 seconds). It was also observed that, small recordings are mostly present in one of the partitions (set b) of the entire dataset. The signal quality of the small recordings are also very poor compared to others. A total of 684 long and 260 short recordings are available in the balanced subset of 944 recordings used in method 1. Equal representations of normal and abnormal cases were selected from each of them in order to create two balanced subset. Finally, separate sets of features are selected for each cases, for designing of the classifiers. Method 4 - Separate models but unbalanced long recordings: This is a logical extension of method 3. Similar to method 2, here all possible PCG features were explored on the entire unbalanced dataset. The model for the short recordings remains the same as used in the previous method. Method 5 - Three class classifier for noisy data: A significant portion of the entire dataset is largely corrupted due to human speech, background noise and the frictional noise generated due to the motion of the stethoscope on human body. In all the previous methods, all the PCG recordings were categorized into one of the normal or abnormal class regardless their signal quality. Ground truth signal quality (good/bad) for each signal was also provided along the training set in a binary form. This shows that 279 out of a total 3153 recordings are very poor to analyse even by expert annotators. Six more features were derived for the noisy signals and are combined with the previous 88 features. the new features are - standard deviation of ratio between 1) diastolic and S1 power, 2) diastolic and S2 power, 3) mean of ratio between S1 and S2 power, median of ratio between 4) diastolic and S1 power, 5) diastolic and S2 power and 6) Kurtosis of the envelop of the auto-correlated PCG signal. All 94 features are used to train the classifier and tested on the entire dataset of 3153 recordings for performance evaluation.
Experimental Results
[0037] The popular ensemble learning method Random Forest (RF) is used for creating the learning models and classification. The number of decision trees in the forest for generating the models is optimized during training. All the results in this disclosure are reported using 5-fold cross validation technique. The performance is evaluated in terms of three metrics 1) Sensitivity (Se) 2) Specificity (Sp) and M Acc = (Se + Sp) = 2. All unsure predictions, obtained in method 5 are marked as correct in the scoring system, if the ground truth signal quality is poor and incorrect otherwise.
[0038] Table 1 shows a comparative analysis among all the five methodologies explored in this paper. For all the cases, the performance metrics are reported in terms of their mean (mean) and standard deviation (std) values obtained across all folds of the 5-fold cross validation technique. It can be concluded that overall performance (MAcc) of the first three methodologies is quite similar. However, due to training on a balanced dataset, method 1 and 3 generates more unbiased classifiers, resulting in sensitivity and specificity scores close to each other.

[0039] Method 4 shows a significant improvement in meanMAcc over the first three methods owing to very high sensitivity. However, the overall classification score is fairly unstable as evident in high standard deviation values across all matrices.
[0040] A high value of sensitivity and specificity can simultaneously be achieved in method 5. In spite of being trained on an unbalanced dataset, addition of new features for identifying the noisy recordings is found to improve the accuracy significantly over the other methods. A possible reason may be, treating the noisy recordings as a separate class, reduces the anomaly in both normal and abnormal classes, thereby improving the overall training.
[0041] In the present disclosure, sensitivity measures the fraction of abnormal heart sounds out of all the test cases, getting correctly detected by the classifier. Specificity on the other hand, measures the fraction of normal heart sounds that are being correctly identified. Since, the present embodiment is dealing with a screening system, a high value of sensitivity is always required to ensure that most of the abnormal heart sound gets detected by the system. Thus, in spite of a lesser accuracy compared to method 5, method 4 is the expected to come out to be ta suitable method for developing a screening system due to its mean sensitivity score 0.9. However, if both sensitivity and specificity are equally important, method 5 comes out to be the most optimum approach.
[0042] The written description describes the subject matter herein to enable any person skilled in the art to make and use the embodiments. The scope of the subject matter embodiments is defined by the claims and may include other modifications that occur to those skilled in the art. Such other modifications are intended to be within the scope of the claims if they have similar elements that do not differ from the literal language of the claims or if they include equivalent elements with insubstantial differences from the literal language of the claims.
[0043] The embodiments of present disclosure herein addresses unresolved problem of noise in PCG, variation in signal quality, sensor quality etc. for identifying heart diseases using PCG signal. The embodiment, thus provides a system and method for classifying heart disease in a person using phonocardiogram (PCG) signal of the person.
[0044] It is, however to be understood that the scope of the protection is extended to such a program and in addition to a computer-readable means having a message therein; such computer-readable storage means contain program-code means for implementation of one or more steps of the method, when the program runs on a server or mobile device or any suitable programmable device. The hardware device can be any kind of device which can be programmed including e.g. any kind of computer like a server or a personal computer, or the like, or any combination thereof. The device may also include means which could be e.g. hardware means like e.g. an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of hardware and software means, e.g. an ASIC and an FPGA, or at least one microprocessor and at least one memory with software modules located therein. Thus, the means can include both hardware means and software means. The method embodiments described herein could be implemented in hardware and software. The device may also include software means. Alternatively, the embodiments may be implemented on different hardware devices, e.g. using a plurality of CPUs.
[0045] The embodiments herein can comprise hardware and software elements. The embodiments that are implemented in software include but are not limited to, firmware, resident software, microcode, etc. The functions performed by various modules described herein may be implemented in other modules or combinations of other modules. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can comprise, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
[0046] The medium can be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system (or apparatus or device) or a propagation medium. Examples of a computer-readable medium include a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a random access memory (RAM), a read-only memory (ROM), a rigid magnetic disk and an optical disk. Current examples of optical disks include compact disk-read only memory (CD-ROM), compact disk-read/write (CD-R/W) and DVD.
[0047] A data processing system suitable for storing and/or executing program code will include at least one processor coupled directly or indirectly to memory elements through a system bus. The memory elements can include local memory employed during actual execution of the program code, bulk storage, and cache memories which provide temporary storage of at least some program code in order to reduce the number of times code must be retrieved from bulk storage during execution.
[0048] Input/output (I/O) devices (including but not limited to keyboards, displays, pointing devices, etc.) can be coupled to the system either directly or through intervening I/O controllers. Network adapters may also be coupled to the system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks. Modems, cable modem and Ethernet cards are just a few of the currently available types of network adapters.
[0049] A representative hardware environment for practicing the embodiments may include a hardware configuration of an information handling/computer system in accordance with the embodiments herein. The system herein comprises at least one processor or central processing unit (CPU). The CPUs are interconnected via system bus to various devices such as a random access memory (RAM), read-only memory (ROM), and an input/output (I/O) adapter. The I/O adapter can connect to peripheral devices, such as disk units and tape drives, or other program storage devices that are readable by the system. The system can read the inventive instructions on the program storage devices and follow these instructions to execute the methodology of the embodiments herein.
[0050] The system further includes a user interface adapter that connects a keyboard, mouse, speaker, microphone, and/or other user interface devices such as a touch screen device (not shown) to the bus to gather user input. Additionally, a communication adapter connects the bus to a data processing network, and a display adapter connects the bus to a display device which may be embodied as an output device such as a monitor, printer, or transmitter, for example.
[0051] The preceding description has been presented with reference to various embodiments. Persons having ordinary skill in the art and technology to which this application pertains will appreciate that alterations and changes in the described structures and methods of operation can be practiced without meaningfully departing from the principle, spirit and scope.
,CLAIMS:1. A method for classifying heart disease in a person using phonocardiogram (PCG) signal of the person, the method comprising a processor implemented steps of:

capturing phonocardiogram (PCG) signal from the person using a PCG sensor;
removing noise from the PCG signal using a filtering module;
segmenting the PCG signal to get a first sound, a second sound, systole and diastole out of the PCG signal;
extracting a plurality of time domain features, a plurality of frequency domain features and a plurality of time-frequency domain features from the segmented PCG signal;
choosing a methodology out of a set of classification methodologies using a training data, wherein the training data comprises sample data covering various scenarios of diseased and healthy person; and
classifying the person as normal or abnormal using the chosen methodology and extracted plurality of features based on a predefined criteria.

2. The method of claim 1, wherein the set of classification methodologies comprises a baseline two class classifier approach, baseline method on unbalanced recordings using the plurality of features approach, models for long and short recordings, models for unbalanced long recording, and three class classifier for noisy PCG signal.

3. The method of claim 1 further comprising determining the length of captured PCG signal to determine the methodology for classification.

4. The method of claim 1, wherein the segmenting is performed using logistic regression based Hidden Semi Markov Model (HSMM) approach.

5. The method of claim 1 wherein the plurality of features are ranked based on the maximal informal coefficients.

6. The method of claim 1 further comprising the step of digitalizing the captured PCG signal.

7. The method of claim 1 further comprising the step of storing the PCG signal for analysis.

8. A system for classifying heart disease in a person using phonocardiogram (PCG) signal of the person, the system comprises:

a PCG sensor for capturing the PCG signal from the person;
a filtering module for removing the noise from the captured PCG signal;
a memory; and
a processor in communication with the memory, the memory further comprises:
a segmentation module for segmenting the PCG signal to get a first sound, a second sound, systole and diastole out of the PCG signal;
a feature extraction module for extracting a plurality of time domain features, a plurality of frequency domain features and a plurality of time-frequency domain features from the segmented PCG signal;
a selection module for choosing a methodology out of a set of classification methodologies using a training data, wherein the training data comprises sample data covering various scenarios of diseased and healthy person; and
a classification module for classifying the person as normal or abnormal using the chosen methodology and extracted plurality of features based on a predefined criteria.