METHOD AND APPARATUS FOR PROCESSING HEART SOUND SIGNALS

FIELD OF THE INVENTION

The invention relates to a method and apparatus for processing signal, particularly, relates to a method and apparatus for processing heart sound signals.

BACKGROUND OF THE INVENTION

Heart murmurs are abnormal heart sounds made by blood moving through the heart valves, the heart chambers or the blood vessels near the heart. Normally, heart valves do not open properly (such as stenosis) or do not close properly (such as regurgitation), which will cause heart murmurs. Heart murmurs are commonly classified according to murmur occurrence in different phases of the heart cycle. The systolic murmurs occur during contraction of ventricles and the diastolic murmurs occur during relaxation of ventricles.

Heart murmurs can be identified by performing an auscultation, i.e. listening to the internal sounds of a body, especially heart, lung, and abdominal organs. For performing an auscultation, usually, a stethoscope is placed on a body and moving the stethoscope left or right (up or down) until a physician can perceive murmur characteristics from the sound transmitted from the stethoscope.

However, it requires a lot of clinical experience, when a physician uses stethoscope to perform auscultation for acquiring murmurs of heart sounds. So, the current method of identifying heart murmurs, which is based on experienced cardiologists during performing auscultation, is not convenient and accurate.

SUMMARY OF THE INVENTION

An object of this invention is to provide a method of processing heart sound signals conveniently and accurately.

The method of processing heart sound signals comprises the steps of: - receiving heart sound signals, extracting cepstral features from the heart sound signals, identifying a murmur type for each cepstral feature by comparing the cepstral feature with pre-stored cepstral features, and outputting a characteristic element for representing the identified heart murmur type.

The advantage is that the method can identify heart murmurs accurately and conveniently.

The invention also proposes an apparatus for implementing the different steps of said method according to the invention.

Detailed explanations and other aspects of the invention will be given below.

DESCRIPTION OF THE DRAWINGS

The above and other objects and features of the present invention will become more apparent from the following detailed description considered in connection with the accompanying drawings, in which:

Fig.l depicts a flowchart of a method in accordance with an embodiment of the invention;

Fig.2 depicts a flowchart for an embodiment of the extracting step 12 in accordance with the method of Fig.1 ;

Fig. 3 depicts a schematic diagram of an apparatus in accordance with an embodiment of the invention;

Fig. 4 depicts a stethoscope comprising the apparatus of Fig.3; and

Fig. 5 depicts a schematic diagram of the sensors of Fig. 4 placed on a body.

The same reference numerals are used to denote similar parts throughout the figures.

DETAILED DESCRIPTION

Fig. 1 depicts a flowchart of a method in accordance with an embodiment of the invention. The method of processing heart sound signals comprises the following four steps.

Stepl l, receiving heart sound signals. The heart sound signals can be received from multiple auscultation areas, and detected by multiple acoustic sensors simultaneously. The received heart sound signals can be stored in a database.

The receiving step 11 may comprise a step of eliminating noise from the heart sound signals. The heart sound signals may comprise noise generated from the internal organs of a body or from ambient around a body. The step of eliminating noise can be implemented by band-pass filtering, smoothing filtering, adaptive filtering etc.

The receiving step 11 may comprise a step of normalizing the heart sound signals. The normalizing step is intended to normalize the heart sound signals for avoiding amplitude variation. The amplitude variations may be caused by age, physiology etc.

The receiving step 11 may also comprise a step of enhancing the amplitude of one or more heart sound signals which have lower amplitude compared to other heart sound signals. For example, four heart sound signals are received from four auscultation areas respectively: mitral area, aortic area, pulmonic area, tricuspid area. The amplitudes of heart sound signals from different areas may be different. So the amplitudes of one or more heart sound signals may be enhanced to be close to the amplitude of other heart sound signals.

The receiving step 11 is further intended to: receive an ECG (Electrocardiograph) signal. The ECG signal can be received from an ECG sensor. The ECG signal may comprise 12 leads (Einthoven, W.,

Galvanometrische registratie van het menschilijk electrocardiogram.

Leiden:Eduard Ijdo, 1902: p. 101-107). segment the heart sound signals by the ECG signal. The heart sound signals are intended to be segmented into heart cycles (heart beat). Heart sound signal consists of a plurality of heart cycles. The segmenting step may be carried out by one of the 12-leads of the ECG signal (preferably, lead II of the ECG signal).

Step 12, extracting cepstral features from the heart sound signals. The cepstral features are extracted from heart cycles of the heart sound signals. In the following, one embodiment of extracting step 12 is to be illustrated in Fig. 2.

Step 13, identifying a heart murmur type for each cepstral feature by comparing the cepstral feature with pre-stored cepstral features. The identifying step is intended to identify a heart murmur type for each cepstral feature by comparing the cepstral feature with the pre-stored

cepstral features based on a Support Vectors Machine (SVM). The pre-stored cepstral features are stored in a database together with corresponding characteristic element for each pre-stored cepstral feature. The pre-stored cepstral features may be corresponding to three heart murmur types: non-murmur type, systolic murmur type, and diastolic murmur type. If a cepstral feature matches with a systolic murmur type, which indicates that the heart sound signals comprise systolic murmurs, and the cepstral feature is systolic murmur type; if a cepstral feature matches with a diastolic murmur type, which indicates that the heart sound signals comprise diastolic murmurs, and the cepstral feature is diastolic murmur type; if a cepstral feature matches with a non-murmur type, which indicates that the heart sound signals do not comprise heart murmurs, and the cepstral feature are non-murmur type.

The Support Vector Machines (SVM) is used for classification and regression. The Support Vectors Machines rely on pre-processing data to represent patterns in a high dimension. The Support Vectors Machine (SVM) maps the cepstral features into a high dimension and constructs a separate hyperplane which can maximize a margin between two types of cepstral features, for example, cepstral features of murmur type and cepstral features of non- murmur type. In one embodiment, C-SVM is used, wherein C is a cost parameter. C, in the SVM optimization function, controls the penalty paid by the SVM for mis-classification and can be used to vary the performance of the SVM.

Step 14, outputting a characteristic element for representing the identified heart murmur type. The outputting step 14 may be also intended to output a phonocardiogram for the heart sound signals. The output content, which comprises characteristic element, phonocardiogram etc., can be shown by a display. The characteristic element can be pre-stored in the database together with the pre-stored cepstral feature type. The characteristic may be character or/and image for indicating different heart murmur type. If a cepstral feature belongs to systolic murmur type, then the display shows the characteristic element for the systolic murmur type; if a cepstral feature belongs to diastolic murmur type, then the display shows the characteristic element for the diastolic murmur type; if the heart sound does not comprise heart

murmurs (non-murmur type), then the display shows characteristic element (may be a character for indicating the heart sound is normal) for the non-murmur type.

Fig.2 depicts a flowchart for an embodiment of the extracting step 12 in accordance with the method of Fig.1.

The extracting step 12 may comprise the steps of:

Step 121, processing the heart sound signal (HSS as shown in Fig. 2) by a Short- Time Fourier Transform (STFT) to generate spectrum.

Step 122, processing the spectrum by a Triangular Filter. The processing step 122 is intended to fit a Triangular Filter to initial frequencies of the spectrum.

Step 123, processing the spectrum from the Triangular Filter by a logarithmic compression.

Step 124, processing the spectrum from the logarithmic compression by a Discrete Cosine Transform (DCT). In processing step 124, the DCT is type III. The processing step 124 is intended to use a Discrete Cosine Transform for computing cepstral features from the spectrum.

Step 125, generating the cepstral features (CF as shown in Fig. 2).

In a preferred embodiment, the accuracy of identifying heart murmur is much higher by using the Support Vector Machine to identify heart murmur type of cepstral feature.

Table_l lists some experiment data which indicates that the identifying accuracy of using the combination of cepstral features and Support Vector Machine is the highest — 95.20%. The accuracy of other combinations, like spectral features/wavelet-based features combining with kNN (k-nearest neighbor algorithm)/MLP (Multi-layer Perceptron), is lower than the accuracy of the combination of cepstral features and Support Vector Machine.

Table 1

Fig. 3 depicts a schematic diagram of an apparatus in accordance with an embodiment of the invention. The apparatus 30 for processing heart sound signals comprises four units in the following.

A receiving unit 31 is used for receiving heart sound signals. The heart sound signals can be received from multiple auscultation areas, and detected by multiple acoustic sensors simultaneously. The received heart signals can be stored in a database.

The receiving unit 31 may be intended to eliminate noise from the heart sound signals. The heart sound signals may comprise noise generated from the internal organs of a body or from ambient around a body. The receiving unit 41 may eliminate noise by band-pass filtering, smoothing filtering, adaptive filtering etc.

The receiving unit 31 may be intended to normalize the heart sound signals. The receiving unit 31 is intended to normalize the heart sound signals to avoid amplitude variation.

The amplitude variations may be caused by age, physiology etc.

The receiving unit 31 may be also intended to enhance the amplitude of one or more heart sound signals which have lower amplitude compared to other heart sound signals. For example, four heart sound signals are received from four auscultation areas respectively: mitral area, aortic area, pulmonic area, tricuspid area. The amplitudes of heart sound signals from

different areas may be different. So the amplitudes of one or more heart sound signals may be enhanced to be close to the amplitude of other heart sound signals.

The receiving unit 31 is further intended to: - receive an ECG (Electrocardiograph) signal. The ECG signal can be received from an ECG sensor. The ECG signal may comprise 12 leads (Einthoven, W., Galvanometrische registratie van het menschilijk electrocardiogram. Leiden:Eduard Ijdo, 1902: p. 101-107). segment the heart sound signals by the ECG signal. The heart sound signals are intended to be segmented into heart cycles. Heart sound signal consists of a plurality of heart cycles. The segmenting may be carried out by one of the 12- leads of the ECG signal (preferably, lead II of the ECG signal).

The heart sound signals and the ECG signal are shown as HSS&ECGS in Fig. 3.

An extracting unit 32 is used for extracting cepstral features from the heart sound signals. One embodiment of the apparatus 30 is that the extracting unit 32 may comprise the following five elements.

A first processing unit is used for processing the heart sound signal by a Short-

Time Fourier Transform (STFT) to generate spectrum.

A second processing unit is used for processing the spectrum by a Triangular Filter. The processing step 122 is intended to fit the Triangular Filter to initial frequencies of the spectrum.

A third processing unit is used for processing the spectrum from the Triangular Filter by a Logarithmic Compression.

A fourth processing unit is used for processing the spectrum from the Logarithmic Compression by a Discrete Cosine Transform (DCT). In the fourth processing unit, the DCT is type III. The fourth processing unit is intended to use a Discrete Cosine transform for computing features from the spectrum.

A generating unit is used for generating the cepstral features.

An identifying unit 33 is used for identifying a heart murmur type for each cepstral feature by comparing the cepstral feature with pre-stored cepstral features. The identifying unit 33 is intended to identify a heart murmur type for each cepstral feature by comparing the cepstral feature with the pre-stored cepstral features based on a Support Vectors Machine (SVM). The pre-stored cepstral features are stored in a database together with corresponding characteristic element for each pre-stored cepstral feature. The pre-stored cepstral features may be corresponding to three heart murmur types: non- murmur type, systolic murmur type, and diastolic murmur type. If a cepstral feature matches with a systolic murmur type, which indicates that the heart sound signals comprise systolic murmurs, and the cepstral feature is systolic murmur type; if a cepstral feature matches with a diastolic murmur type, which indicates that the heart sound signals comprise diastolic murmurs, and the cepstral feature is diastolic murmur type; if a cepstral feature matches with a non-murmur type, which indicates that the heart sound signals do not comprise heart murmurs, and the cepstral feature are non-murmur type.

The Support Vector Machines are used for classification and regression. The Support Vectors Machines rely on pre-processing data to represent patterns in a high dimension. The Support Vectors Machine (SVM) maps the cepstral features into a high dimension and constructs a separate hyperplane which can maximize a margin between two types of cepstral features, for example, cepstral features of murmur type and cepstral features of non-murmur type. In one embodiment, C-SVM is used, wherein C is a cost parameter. C, in the SVM optimization function, controls the penalty paid by the SVM for mis-classification and can be used to vary the performance of the SVM.

Outputting unit 34 is used for outputting a characteristic element (CF as shown in Fig. 3) for representing the identified heart murmur type. The outputting unit 34 may be also intended to output a phonocardiogram for the heart sound signals. The output content, which may comprise characteristic element, phonocardiogram, can be shown by a display. The characteristic element can be pre-stored in a database together with pre-stored cepstral features. The characteristic may be character and/or image for indicating different heart murmur type. If a cepstral feature belongs to systolic murmur type, then the display shows the characteristic element for the systolic murmur type; if a cepstral feature belongs to diastolic murmur type, then the display shows the characteristic element for the diastolic murmur type; if the heart sound does not comprise heart murmurs (non-murmur type or normal), then the display may show characteristic element (may be a character for indicating the heart sound is normal) for the non- murmur type.

Fig. 4 depicts a stethoscope comprising the apparatus of Fig.3.

The stethoscope 40 comprises a first acoustic sensor 41 for aortic area, a second acoustic sensor 42 for pulmonic area, and an ECG sensor 43 for left limb area. Sensors 41-42-43 are integrated into one module.

The stethoscope 40 may also comprise a third acoustic sensor 44 for mitral area.

Sensors 41-42-43 and sensor 44 are connected to an apparatus 30 as previously described.

Optionally, the stethoscope 40 may also comprise an earphone 45 connected to the apparatus 30.

Fig. 5 depicts a schematic diagram of the sensors of Fig. 4 placed on a body. The first acoustic sensor 41, the second acoustic sensor 42, the ECG sensor 43 and the third acoustic sensor 44 are placed on the aortic area, the pulmonic area, the left limb area and the mitral area of

a body respectively. The sensors can be movable on a body, attached on a body, or sucked on a body.

In another embodiment, the ECG sensor 43, the first acoustic sensor 41, and the second acoustic sensor 42 can be separate (not integrated into one module), and all these sensors can be separately moveable on a body, or attached on a body, or sucked on a body.

In a further embodiment, the stethoscope 40 comprises a plurality of ECG sensors and a plurality of acoustic sensors. Each ECG sensor may be integrated with each acoustic sensor into one module, and then the module can be sucked on a body, attached on a body, or moveable on a body.

It should be noted that the above-mentioned embodiments illustrate rather than limit the invention and that those skilled in the art will be able to design alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim or in the description. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by unit of hardware comprising several distinct elements and by unit of a programmed computer. In the system claims enumerating several units, several of these units can be embodied by one and the same item of hardware or software. The usage of the words first, second and third, et cetera, does not indicate any ordering. These words are to be interpreted as names.