JP2013123495A - Respiratory sound analysis device, respiratory sound analysis method, respiratory sound analysis program, and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To classify the respiratory sound measured by a subject according to the individual difference and the state of a disease.SOLUTION: A feature value extraction part 3 obtains an envelope curve of frequency spectra calculated based on measured respiratory sound data, and extracts a feature value of a model waveform by approximating the envelope curve to at least one model waveform (a gauss waveform). A respiratory sound classification part 4 classifies the measured respiratory sound based on the feature value.

Description

  The present invention relates to a respiratory sound analyzing apparatus and a respiratory sound analyzing method for analyzing an electronically measured respiratory sound.

  A stethoscope is a diagnostic instrument that diagnoses the presence or absence of an abnormality by listening to a subject's body sound (breathing sound, heart sound, etc.). Conventionally, diagnosis using a stethoscope is based on the skill and experience of a doctor and cannot be easily performed by anyone. In particular, respiratory sounds are important information for diagnosing the presence or state of respiratory diseases, so if information from respiratory sounds can be extracted in an easy-to-understand form, specialized diagnosis based on respiratory sounds is necessary. It can be made easy regardless of skill or experience.

  In order to perform such a diagnosis automatically, it is necessary to be able to handle respiratory sounds as data. In recent years, electronic stethoscopes that digitize body sounds have been put into practical use, and therefore respiratory sound data measured by an electronic stethoscope can be used. If respiratory sounds are converted into data, it is possible to automatically diagnose respiratory system diseases by analyzing the data and classifying the respiratory sounds based on the analysis results.

  Patent Document 1 discloses an apparatus for classifying such breathing sounds. Specifically, Patent Document 1 discloses that when it is determined that the local variance obtained from the frequency spectrum of the respiratory sound exceeds a certain threshold level, the respiratory sound is determined to be a continuous rarity. Yes.

  Patent Document 2 discloses a breathing sound analyzer that can accurately detect rales. Specifically, in Patent Document 2, an analysis signal is calculated by performing Hilbert transform on respiratory sound data, an envelope is calculated from the analysis signal, and a peak that is a candidate for a ra sound is extracted from the envelope. Is disclosed.

JP 2004-357758 A (released on December 24, 2004) JP 2009-106574 A (published May 21, 2009)

  However, the peak of the frequency spectrum varies depending on individual differences and disease states. On the other hand, according to the classification method disclosed in Patent Document 1, since a local dispersion value near a preset frequency is obtained, respiratory sounds cannot be classified according to individual differences and disease states.

  Moreover, although patent document 2 has described the detection of a rarity, it has not described about the classification | category with an intermittent rarity and a continuous rarity. For this reason, even if the method disclosed in Patent Document 2 is used, respiratory sounds cannot be classified according to individual differences and disease states.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a respiratory sound analysis apparatus and a respiratory sound analysis capable of classifying respiratory sounds measured by a subject according to individual differences and disease states. It is to provide a method.

  In order to solve the above-described problem, the respiratory sound analysis apparatus of the present invention is a respiratory sound analysis apparatus that analyzes the respiratory sound of a subject, and uses the frequency spectrum of respiratory sound data in which the respiratory sound is electronically measured. A feature value extracting unit that extracts a feature value of the model waveform by approximating the model waveform, and a breathing sound classifying unit that classifies the breathing sound based on the extracted feature value; It is characterized by being.

  Further, in order to solve the above-described problem, the respiratory sound analysis method of the present invention approximates the frequency spectrum of respiratory sound data in which the respiratory sound is electronically measured to at least one model waveform. A feature value extracting step of extracting a feature value for the waveform and a breathing sound classification step of classifying the breathing sound based on the extracted feature value are provided.

  The breathing sound varies depending on the state of the trachea of the subject, and shows a different frequency spectrum depending on the presence of a stenosis or a liquid film. For example, when a plurality of stenosis occurs in the trachea, since the trachea of the stenosis portion has a different size, the breathing sound shows a frequency spectrum having different peaks accordingly. In addition, when a liquid film is generated in the trachea, the side lobe of the frequency spectrum is widened because the liquid film is ruptured by respiration.

  Therefore, in the above configuration, the feature of the respiratory sound is extracted by approximating the frequency spectrum to the model waveform by the feature value extracting means (feature value extracting step) and extracting the feature value of the model waveform. Can do. Then, by classifying the respiratory sounds based on the extracted feature values by the respiratory sound classification means (respiratory sound classification step), it is possible to perform classification that reflects the state of the trachea.

  In the respiratory sound analysis device, the feature value extracting means includes a section extracting means for extracting a predetermined section from the respiratory sound data, and a frequency spectrum calculating means for calculating a frequency spectrum of the extracted respiratory sound data in the extracted section. And an envelope calculation means for calculating an envelope of the calculated frequency spectrum, and approximating the calculated envelope to a Gaussian waveform as the model waveform, thereby obtaining the feature value based on the Gaussian waveform. It is preferable to have a feature value calculation means for calculating.

  In the above configuration, the predetermined section is extracted from the respiratory sound data by the section extracting means. Thereby, the sample of the minimum respiratory sound required for classification is extracted. A frequency spectrum is calculated by the frequency spectrum calculation means based on the extracted respiratory sound data. Then, the envelope of the calculated frequency spectrum is calculated by the envelope calculation means. The envelope is approximated to a Gaussian waveform by the feature value calculation means, and the feature value is calculated based on the Gaussian waveform.

  The Gaussian waveform is suitable as a model waveform that approximates the frequency spectrum of respiratory sounds, and a mixed Gaussian distribution can be used as an approximation method. Therefore, feature values can be extracted efficiently.

  In the respiratory sound analysis device, the feature value calculating means calculates the number of peaks of each Gaussian waveform and a variance value of each Gaussian waveform as the characteristic value based on the Gaussian waveform, and the respiratory sound classification means includes the It is preferable to classify the respiratory sounds based on the number of peaks and the maximum variance value among the variance values.

  The peak number indicates that the measured respiratory sound includes a plurality of respiratory sounds. The dispersion value represents the degree of spread of the Gaussian waveform (frequency spectrum side lobe). Therefore, it can be said that the greater the number of peaks, the more abnormal the breathing sound. Further, it is possible to determine whether or not the breathing sound is a continuous rarity depending on the degree of spread of the Gaussian waveform.

  Therefore, the respiratory sounds are classified by the respiratory sound classification means based on the number of peaks calculated by the characteristic value calculating means and the maximum variance value based on the characteristic values, thereby classifying the respiratory sounds using the minimum characteristic values. be able to.

  It is preferable that the respiratory sound analysis apparatus includes a classification result display unit that displays a classification result obtained by the respiratory sound classification unit.

  Thereby, a classification result can be checked easily.

  In the respiratory sound analyzer, the classification result display means preferably displays a waveform of the respiratory sound data corresponding to the classification result.

  Thereby, the classification result can be verified against the waveform of the respiratory sound data. For example, when it is difficult to diagnose a detailed disease based only on the classification result, it is easy to determine an abnormality by referring to the disturbance according to the waveform.

  In the respiratory sound analysis apparatus, it is preferable that the classification result display means displays a disease name corresponding to the classification result.

  Thereby, the presence or absence of a disease based on the breathing sound can be easily diagnosed.

  Moreover, it is preferable that the said classification result display means displays the timing of the expiration and inspiration in the said respiratory sound corresponding to the said classification result.

  This makes it easy to determine the type of disease or the like by combining whether the breathing sound is more disturbed in expiration or inspiration with the waveform of the breathing sound data.

  The respiratory sound analysis program of the present invention is a program for causing the feature value extraction means and the respiratory sound classification means in the respiratory sound analysis apparatus to function. The recording medium of the present invention is a computer-readable recording medium that records the respiratory sound analysis program. These respiratory sound analysis programs and recording media are also included in the technical scope of the present invention.

  The semiconductor device voltage measurement method and the respiratory sound analysis apparatus according to the present invention are configured as described above, thereby enabling classification of respiratory sounds according to individual differences and disease states. Therefore, by adopting the present invention, it is possible to improve the classification accuracy of lung noise.

It is a block diagram which shows the structure of the respiratory sound analyzer which concerns on embodiment of this invention. It is a wave form diagram which shows the waveform of the respiratory sound measured by the biological sound measurement part in the said respiratory sound analyzer. It is a figure which shows the frequency spectrum of the respiratory sound calculated by the frequency spectrum calculation part in the said respiratory sound analyzer. It is a figure which shows the frequency spectrum smooth | blunted by the frequency spectrum smoothing part in the said respiratory sound analyzer. It is a figure for demonstrating the peak detection of the frequency spectrum by the peak detection part in the said respiratory sound analyzer. It is a figure which shows the concept of the approximation to the mixed Gaussian distribution of the frequency spectrum by the feature value calculation part in the said respiratory sound analyzer. (A) And (b) is a figure which shows the state of the trachea which a continuous ra sound produces, (c) is a figure which shows the mixed Gaussian distribution of the frequency spectrum about the respiratory sound produced in this trachea. (A) is a figure which shows the state of the trachea which an intermittent ra sound produces, (b) is a figure which shows the mixed Gaussian distribution of the frequency spectrum about the respiratory sound produced in this trachea. It is a figure which shows the display screen containing the classification result of the respiratory sound displayed by the classification result display part in the said respiratory sound analyzer. It is a flowchart which shows the procedure which processes the classification | category of the respiratory sound by the said respiratory sound classification | category part. It is a distribution map at the time of classifying the respiratory sound based on the number of peaks and the maximum variance value by the respiratory sound classification unit. The figure which shows the specific example of the Gaussian waveform (Gaussian approximate waveform) in the frequency spectrum (measurement waveform) and mixed Gaussian distribution corresponding to each respiratory sound which classified the respiratory sound based on the number of peaks and the maximum dispersion value by the respiratory sound classification part. is there.

  An embodiment according to the present invention will be described below with reference to FIGS.

[Configuration of respiratory sound analyzer]
FIG. 1 is a block diagram illustrating a configuration of a respiratory sound analysis apparatus 1 according to the present embodiment.

  The respiratory sound analysis apparatus 1 shown in FIG. 1 is an apparatus that analyzes respiratory sounds, classifies the respiratory sounds based on the results, and displays them. The respiratory sound analysis device 1 includes a body sound measurement unit 2, a feature value extraction unit 3, a respiratory sound classification unit 4, and a classification result display unit 5. In addition, the respiratory sound analysis apparatus 1 includes a storage unit (not shown) including a memory or the like as a temporary storage area for data obtained as a result of processing of each unit or data generated during processing of each unit. Yes.

[Configuration of biological sound measurement unit]
The body sound measurement unit 2 is provided to measure a body sound (especially breathing sound) of a subject and output the measurement value as digital data, and is configured by a stethoscope (an electronic stethoscope, a sound collector, etc.). Has been. For example, an electronic stethoscope detects vibrations of a diaphragm generated by receiving a body sound by a vibration sensor and converts them into an electric signal, and further samples the electric signal at a predetermined sampling frequency and converts it into digital data.

[Configuration of Feature Value Extraction Unit]
The feature value extraction unit 3 (feature value extraction means) is a mixed Gaussian distribution in which a plurality of Gaussian waveforms are mixed with the frequency spectrum of the respiratory sound from the respiratory sound data (respiratory sound data) measured by the body sound measuring unit 2 To extract feature values. The feature value extraction unit 3 includes a section extraction unit 31, a frequency spectrum calculation unit 32, a frequency spectrum smoothing unit 33, a peak detection unit 34, and a feature value calculation unit 35 in order to extract feature values.

<Configuration of section extraction unit>
FIG. 2 is a waveform diagram showing the waveform of the respiratory sound measured by the body sound measurement unit 2.

  The section extraction unit 31 (section extraction means) divides a section of any one respiratory period from the respiratory sound data (breathing sound data) measured by the body sound measurement unit 2, and the respiratory sound data of the section, or Respiratory sound data of either exhaled air or inhaled air in the section is extracted and stored in the storage unit. The reason why the breath sound data of inspiration or expiration is individually extracted is to distinguish between inspiration and expiration since the frequency spectrum differs between inspiration and expiration.

  There are two methods for determining the period of one breath period: a method of determining based on respiratory sound data (respiratory sound waveform) and a method of determining based on information from the body sound measuring unit 2.

  A method of discriminating based on the respiratory sound waveform is to determine one respiratory cycle by comparing the magnitude of the amplitude of the respiratory sound waveform with a predetermined threshold. As shown in FIG. 2, at the start of expiration in one breathing period, the amplitude of the breathing sound changes greatly. Compared to this, at the start of inspiration, the amplitude does not change as much as at the start of expiration. In order to capture this difference in amplitude, a first threshold value that can determine an amplitude change at the start of expiration and a second threshold value that can determine an amplitude change at the start of inspiration are prepared. By comparing and determining the start of exhalation based on the first threshold value, the period from the start of one exhalation to the start of the next exhalation can be specified as an interval of one breath period. Furthermore, by comparing and determining the start of inspiration based on the second threshold, it is possible to specify the inspiration period in one respiration cycle.

  When this method is used, the section extraction unit 31 may perform the above determination, or another determination unit may perform the above determination and give the determination result to the section extraction unit 31.

  For example, when an electronic stethoscope is used as the biological sound measurement unit 2, an acceleration sensor is built in the chest piece of the electronic stethoscope, and breathing is performed by the acceleration sensor. By capturing the movement of the chest wall at the time, the period of expiration and inspiration in one respiratory cycle or one respiratory cycle can be specified. Specifically, since the thorax contracts during expiration, the moving direction of the chest wall that moves with the contraction is detected by an acceleration sensor. On the other hand, since the rib cage expands during inhalation, the acceleration sensor detects the moving direction of the chest wall that moves with the expansion (the direction opposite to that during contraction). Thereby, the start of exhalation and the start of inspiration can be specified.

  When this method is used, the section extraction unit 31 divides one respiratory period based on the detection result of the acceleration sensor given from the biological sound measurement unit 2 simultaneously with the respiratory sound data, and extracts the respiratory sound data of the section. To do.

<Configuration of frequency spectrum calculation unit>
FIG. 3 is a diagram showing the frequency spectrum of the breathing sound calculated by the frequency spectrum calculation unit 32.

  The frequency spectrum calculation unit 32 (frequency spectrum calculation means) calculates the frequency spectrum of the respiratory sound based on the respiratory sound data of the section extracted by the section extraction unit 31, and stores it in the storage unit. The frequency spectrum calculation unit 32 obtains a frequency spectrum as shown in FIG. 3 by performing a general time function-frequency function conversion process in which a fast Fourier transform (FFT) process is performed on the respiratory sound data. This frequency spectrum represents the frequency distribution of respiratory sound data (respiratory sound).

<Configuration of frequency spectrum smoothing section>
FIG. 4 is a diagram illustrating the frequency spectrum smoothed by the frequency spectrum smoothing unit 33.

  As shown in FIG. 3, the frequency spectrum has a large power spectrum variation, and it is difficult for the peak detector 34 to detect a peak if the frequency spectrum remains as it is. Therefore, the frequency spectrum smoothing unit 33 (envelope calculating means) smoothes the frequency spectrum calculated by the frequency spectrum calculating unit 32 as shown in FIG. 4 and stores it in the storage unit. Thereby, the envelope of a frequency spectrum can be obtained.

  As the frequency spectrum smoothing unit 33, a moving average filter, a cepstrum filter, a Savitzky-Golay filter, or the like can be used. Of these, the average moving filter can be configured most easily and is suitable for smoothing the waveform. The average moving filter is a filter that averages the latest n pieces of data and uses the average value as a representative value, and is a kind of low-pass filter.

<Configuration of peak detector>
FIG. 5 is a diagram for explaining peak detection of the frequency spectrum by the peak detector 34.

  As will be described later, in order for the feature value calculation unit 35 to perform the mixed Gaussian approximation process, the peak frequency, the peak, and the peak amplitude value of the frequency spectrum smoothed by the frequency spectrum smoothing unit 33 are necessary. Therefore, the peak detector 34 detects the peak of the smoothed frequency spectrum, and specifies the peak frequency and the number of peaks for the peak as temporary values. For this reason, the peak detector 34 differentiates the smoothed frequency spectrum and detects the zero-cross point in the differentiated waveform as the peak of the frequency spectrum. In addition, the peak detection unit 34 identifies the peak frequency, the number of peaks, and the peak amplitude value for the detected peak, and stores these in the storage unit as temporary peak parameters. For example, in the example shown in FIG. 5, the peak frequencies f1 to f3, the number of peaks 3 and the peak amplitude values m1 to m3 are specified for the detected peaks.

<Configuration of feature value calculation unit>
FIG. 6 is a diagram illustrating a concept of approximation of the frequency spectrum to the mixed Gaussian distribution by the feature value calculation unit 35.

  As will be described later, since the respiratory sound is composed of one or a plurality of respiratory sounds, the frequency spectrum also includes the frequency components of those respiratory sounds. Therefore, by approximating the frequency spectrum to one or a plurality of model waveforms, it is possible to approximately specify the respiratory sound included in the frequency spectrum. The model waveform is a waveform that can be expressed by a specified model equation. By specifying the model waveform from the frequency spectrum, the feature value of the model waveform can be obtained from the model equation.

  The feature value calculation unit 35 (feature value calculation means) shown in FIG. 6 uses the above principle, and the frequency spectrum (envelope) smoothed by the frequency spectrum smoothing unit 33 is converted into a known curve fitting method. To approximate at least one model waveform. Specifically, the feature value calculation unit 35 changes the variance value of the model formula so that the above-described peak parameter is applied to the temporary value, and the difference value between the temporary value and the estimated peak parameter of the frequency spectrum. The parameter of the model formula is changed so that the value is minimized, and the peak parameter and the dispersion value when the minimum value is obtained are specified as the feature value.

  When the Gaussian waveform is used as the model waveform, the feature value calculation unit 35 uses the temporary peak parameter specified by the peak detection unit 34 to convert the frequency spectrum into the mixed Gaussian distribution represented by the equation (1). In this way, curve fitting (approximation) is performed. Further, the feature value calculation unit 35 calculates the dispersion value and peak parameter for each peak by the equation (1) by curve fitting, and stores them in the storage unit.

  A Gaussian waveform constituting the mixed Gaussian distribution is expressed by Equation (2). In Formula (1) and Formula (2), n, μ1 to μn, and σ represent the number of peaks, average, and variance, respectively.

  For example, when the number of temporary peaks is 3, the feature value calculation unit 35 obtains Expression (3) by setting n in Expression (1) to 3, and the smoothed frequency spectrum is expressed by Expression (3). Approximate a mixed Gaussian distribution. Thereby, the feature value calculation unit 35 specifies the averages μ1 to μ3, the dispersion values σ1 to σ3, and the peak frequencies f1 to f3 in Expression (3).

  Note that the model waveform is not limited to a Gaussian waveform.

[Configuration of respiratory sound classification unit]
The respiratory sound classification unit 4 (respiratory sound classification means) includes a peak number, a peak frequency, a peak amplitude value, and a Gaussian waveform that form an approximated waveform whose frequency spectrum is approximated to a mixed Gaussian distribution by the feature value calculation unit Based on the variance value, the measured breathing sound is classified into one of normal breathing sound, continuous rarity sound, and intermittent rarity sound. Further, the respiratory sound classification unit 4 outputs the classification results of normal respiratory sounds, continuous rales and intermittent rales as classification values of “1”, “2”, and “3”, respectively.

  The classification of respiratory sounds by the respiratory sound classification unit 4 is performed according to the following principle.

<Principle of respiratory sound classification>
7 (a) and 7 (b) are diagrams showing the state of the trachea in which continuous rales are generated, and FIG. 7 (c) is a diagram showing a mixed Gaussian distribution of frequency spectra for respiratory sounds generated in the trachea. . FIG. 8A is a diagram showing a state of a trachea in which intermittent rales are generated, and FIG. 8B is a diagram showing a mixed Gaussian distribution of frequency spectra for respiratory sounds generated in the trachea.

  As shown in FIGS. 7A and 7B, when the stenosis S occurs in the trachea T, the frequency component of the respiratory sound varies depending on the inner diameter of the trachea T and the airflow velocity. As shown in FIG. 7 (c), the mixed Gaussian distribution in which the frequency spectrum is approximated is obtained when the trachea T has a large inner diameter, a medium inner diameter, and a smaller inner diameter due to stenosis S generated in different places. Depending on the inner diameter, the lower peak P1, the middle peak P2, and the higher peak P3. In such a case, the breathing sound can be discriminated (classified) as a continuous rarity due to asthma or the like.

  On the other hand, as shown in FIG. 8A, when the liquid film M is generated in the trachea T, the liquid film M is ruptured by respiration, and thus a burst sound is generated. As shown in FIG. 8B, the mixed Gaussian distribution has a shape in which the side lobes spread by the liquid film M generated in the trachea T. In such a case, it is possible to discriminate (classify) the breathing sound from the intermittent rattle due to pneumonia or the like.

[Configuration of classification result display section]
FIG. 9 is a diagram showing a display screen 101 including the result of classification of respiratory sounds displayed by the classification result display unit 5.

  The classification result display unit 5 (classification result display unit) displays the result of the classification of the respiratory sounds by the respiratory sound classification unit 4. Specifically, the classification result display unit 5 refers to a look-up table provided in the respiratory sound analyzer 1, and reads out and displays the type of respiratory sound corresponding to the classification value. In the lookup table, "normal breathing sound", "continuous rarity", and "intermittent rarity" are associated with classification value "1", classification value "2", and classification value "3", respectively. It has been.

  Further, the classification result display unit 5 displays the measurement waveform of the breathing sound given from the body sound measurement unit 2 and the timing of expiration and inspiration. For this reason, the classification result display unit 5 displays the display screen 101 on a display panel (not shown) provided in the respiratory sound analyzer 1.

  The display screen 101 includes diagnosis specifying information 102, a measurement waveform 103, a respiration timing 104, and a classification result 105.

  The diagnosis specifying information 102 includes a subject ID, a subject name, a subject age, and a diagnosis date. Thereby, a subject and a diagnosis date are specified.

  The measurement waveform 103 is a waveform used for classification of respiratory sounds. The breathing timing 104 represents a period of expiration and inspiration specified by the section extraction unit 31 with a rectangular wave.

  The classification result 105 is one of “normal breath sounds”, “continuous rales”, and “intermittent rales” which are the types of breathing sounds. Moreover, the classification result 105 may include a disease name corresponding to the classified respiratory sound. For example, asthma is mentioned as a typical disease name corresponding to continuous line rales. Moreover, pneumonia is mentioned as a typical disease name corresponding to intermittent rarity.

[Analysis of respiratory sounds by respiratory sound analyzer]
Next, the processing of breathing sound analysis (breathing sound analysis method) by the breathing sound analyzer 1 configured as described above will be described.

[Analysis of respiratory sounds by respiratory sound analyzer]
Subsequently, an analysis of a respiratory sound (a respiratory sound analysis method) by the respiratory sound analysis apparatus 1 configured as described above will be described.

  FIG. 10 is a flowchart showing a procedure for processing the classification of respiratory sounds by the respiratory sound classification unit 4. FIG. 11 is a distribution diagram when respiratory sounds are classified by the respiratory sound classification unit 4 based on the number of peaks and the maximum variance value. FIGS. 12A to 12C show frequency spectra (measured waveforms) corresponding to each breathing sound obtained by classifying the breathing sounds based on the number of peaks and the maximum dispersion value by the breathing sound classification unit 4 and the approximated mixed Gaussian. It is a figure which shows the specific example of the Gaussian waveform (Gaussian approximate waveform) in distribution.

<Extraction of respiratory sound feature values>
First, when the breathing sound of the subject measured by the biological sound measuring unit 2 is given to the feature value extracting unit 3, the feature value of the breathing sound is extracted by the feature value extracting unit 3.

  In the feature value extraction unit 3, the section extraction unit 31 divides and extracts a period of one respiratory period (data of breathing sound of expiration or inspiration as necessary) from the respiratory sound data from the body sound measurement unit 2. (Feature value extraction step).

  The extracted respiratory sound data is converted into a frequency spectrum by the frequency spectrum calculation unit 32. This frequency spectrum is smoothed by the frequency spectrum smoothing unit 33. Thus, a frequency spectrum having a large amplitude variation can be formed into a smooth shape, and an envelope of the frequency spectrum can be obtained.

  The peak of the envelope obtained in this way is detected by the peak detector 34, and the peak frequency, the number of peaks, and the peak amplitude value (peak parameter) are temporarily determined for the peak. Further, the feature value calculation unit 35 performs curve fitting of the envelope curve to the mixed Gaussian distribution based on the temporary value of the peak parameter and the above-described equation (1). Thereby, the number of peaks, the peak frequency, the peak amplitude value, and the dispersion value are obtained for the specified mixed Gaussian distribution. These values are provided to the respiratory sound classification unit 4 as feature values.

《Classification of respiratory sounds》
<Classification procedure>
In the respiratory sound classification unit 4, the measured respiratory sounds are classified based on the feature values (respiratory sound classification step). Here, the case where respiratory sounds are classified using the number of peaks and the variance as features will be described with reference to the flowchart of FIG.

  As shown in FIG. 10, first, it is determined whether or not the peak number n is equal to or less than a reference peak number N that is a determination reference (step S1). Here, when it is determined that the peak number n is equal to or less than the reference peak number N, the maximum dispersion value σmax that is the maximum among the dispersion values of at least one Gaussian waveform forming the mixed Gaussian distribution is the reference dispersion that is the determination reference. It is determined whether or not the value is smaller than Σ (step S2). Here, when it is determined that the maximum variance value σmax is smaller than the reference variance value Σ, the measured respiratory sound is determined as a normal respiratory sound (step S3).

  Further, when it is determined in step S2 that the maximum variance value σmax is equal to or greater than the reference variance value Σ, the measured respiratory sound is determined as an intermittent ra sound (step S4). Further, when it is determined in step S1 that the peak number n is larger than the reference peak number N, it is determined whether or not the maximum dispersion value σmax is smaller than the reference dispersion value Σ (step S5). Here, when it is determined that the maximum variance value σmax is smaller than the reference variance value Σ, the measured breathing sound is determined as a continuous rarity (step S6). If it is determined in step S5 that the maximum variance value σmax is smaller than the variance value Σ, there is no type of breathing sound to be discriminated, and error processing is performed (step S7).

<Distribution of classification results>
Here, an example of respiratory sound classification by the respiratory sound classification unit 4 when the reference peak number N is 2 and the maximum variance value σmax is 150 will be described with reference to FIG.

  First, if the reference peak number N is 2 or less and the maximum variance σmax is less than 150, it is determined that the breathing sound is a normal breathing sound based on the determination results of steps S1 and S2. Normal breathing sounds are distributed in a range where the number of peaks is small (n = 1) and the maximum variance value is small, as indicated by ◯ in FIG.

  If the reference peak number N is greater than 2 (3 or more) and the maximum variance σmax is less than 150, it is determined that the breathing sound is a continuous rale based on the determination results of steps S1 and S5. The The continuous rales are distributed in a range where the number of peaks is large and the maximum dispersion value is small as indicated by Δ in FIG.

  Further, if the reference peak number N is 2 or less and the maximum variance value σmax is 150 or more, it is determined that the breathing sound is an intermittent rale based on the determination results of steps S1 and S2. Intermittent rales are distributed in a range where the number of peaks is small (n = 2) and the maximum dispersion value is large, as indicated by x in FIG.

<Specific examples of classification results>
Next, referring to FIG. 12, the relationship between the respiratory sound classification result by the respiratory sound classification unit 4 and the frequency spectrum when the reference peak number N is 2 and the reference dispersion value Σ is 150 as described above. explain.

  First, the example illustrated in FIG. 12A is an example in which the respiratory sound is determined to be a normal respiratory sound. In this case, the peak number n is 1, and the peak frequency and dispersion value are the values shown in Table 1.

  When the breathing sound is a normal breathing sound, the stenosis S as shown in FIGS. 7A and 7B and the liquid film M as shown in FIG. A breathing sound corresponding to the thickness of T is generated.

  In addition, the example shown in FIG. 12B is an example in which the breathing sound is determined as a continuous rarity. In this case, the number of peaks n is 5, and the peak frequency and dispersion value of each peak are the values shown in Table 2.

  When the breathing sound is a continuous sound, since the stenosis S as shown in FIGS. 7A and 7B occurs in the trachea T, in addition to the breathing sound according to the thickness of the subject's trachea T, A breathing sound corresponding to the thickness of the trachea T narrowed by the stenosis S is generated.

  Furthermore, the example shown in FIG. 12C is an example in which the breathing sound is determined to be an intermittent rale. In this case, the number of peaks n is 2, and the peak frequency and dispersion value of each peak are the values shown in Table 3.

  When the breathing sound is an intermittent rale, the liquid film M as shown in FIG. 8A is generated in the trachea T. Therefore, since the liquid film M is ruptured by breathing, the breathing sound has a dispersion value. growing.

<Display of classification result>
The types of respiratory sounds classified as described above are displayed as the classification result 105 on the display screen 101 by the classification result display unit 5. At this time, the type of breathing sound corresponding to the aforementioned classification value is read from the lookup table and displayed on the display screen 101.

  The respiratory sound classification result 105 is displayed on the display screen 101 simultaneously with the measurement waveform 103 (respiratory sound data waveform) of the subject and the respiratory timing 104. Thereby, not only the respiratory sound classification result 105 but also the classification result 105 can be verified against the respiratory sound waveform. For example, when it is difficult to diagnose a detailed disease based only on the classification result, it is easy to determine an abnormality by referring to the disturbance according to the waveform. Further, by displaying the breathing timing 104 together, it is easy to determine the type of the disease or the like by combining whether the breathing sound is disturbed in the expiration or inspiration with the measurement waveform.

  The classification result display unit 5 displays the type of respiratory sound as the classification result 105 on the display screen 101, but also displays the diagnostic result (disease name) corresponding to the type of the classified respiratory sound. For example, when the classification result 105 is a continuous rale, “asthma” is displayed as the diagnosis result, and when the classification result 105 is an intermittent rale, “pneumonia” is displayed as the diagnosis result.

[Effect of the embodiment]
As described above, in the respiratory sound analysis device 1 of the present embodiment, the body sound measurement unit 2 electronically measures the respiratory sound of the subject, and the feature value extraction unit 3 classifies the respiratory sound from the respiratory sound data. Necessary feature values are extracted, and the respiratory sound classification unit 4 classifies the respiratory sounds measured based on the characteristic values. The feature value extraction unit 3 obtains an envelope by smoothing the frequency spectrum obtained based on the respiratory sound data, acquires each value relating to the peak of the envelope, and uses these values to at least determine the envelope A feature value is extracted by approximating one model waveform.

  Thereby, the characteristic of the respiratory sound according to the state of the subject's trachea is obtained as a result of the approximation of the envelope to the model waveform. Therefore, it is possible to classify by capturing differences in respiratory sounds according to individual differences and disease states. Therefore, the lung noise classification accuracy can be improved as compared with the conventional method.

  The Gaussian waveform is suitable as a model waveform for approximating the frequency spectrum of respiratory sounds, and a mixed Gaussian distribution can be used as an approximation method. Therefore, feature values can be efficiently extracted by using a Gaussian waveform as a model waveform.

[Realization of respiratory sound analyzer]
In the respiratory sound analyzing apparatus 1, at least each block of the feature value extracting unit 3 and the respiratory sound classifying unit 4 is realized by software (respiratory sound analysis program) using a CPU as follows. Or each said block may be comprised by a hardware logic, and may be implement | achieved by the process by the program using DSP (Digital Signal Processor).

  The program code (execution format program, intermediate code program, source program) of the above software may be recorded on a recording medium recorded so as to be readable by a computer. The object of the present invention can also be achieved by supplying the recording medium to the respiratory sound analyzer 1 and reading and executing the program code recorded on the recording medium by the CPU.

  Examples of the recording medium include magnetic tapes such as magnetic tapes and cassette tapes, magnetic disks such as floppy (registered trademark) disks / hard disks, and optical disks such as CD-ROM / MO / MD / BD / DVD / CD-R. Can be used, such as a disk system including IC, a card system such as an IC card (including a memory card) / optical card, or a semiconductor memory system such as mask ROM / EPROM / EEPROM / flash ROM.

  Further, the respiratory sound analyzing apparatus 1 may be configured to be connectable to a communication network, and the program code may be supplied via the communication network. The communication network is not particularly limited. For example, the Internet, intranet, extranet, LAN, ISDN, VAN, CATV communication network, virtual private network, telephone line network, mobile communication network, satellite communication. A net or the like is available. Also, the transmission medium constituting the communication network is not particularly limited. For example, even in the case of wired such as IEEE 1394, USB, power line carrier, cable TV line, telephone line, ADSL line, etc., infrared rays such as IrDA and remote control, Bluetooth ( (Registered trademark), 802.11 wireless, HDR, mobile phone network, satellite line, terrestrial digital network, and the like can also be used. The present invention can also be realized in the form of a computer data signal embedded in a carrier wave in which the program code is embodied by electronic transmission.

[Additional Notes]
In the above-described embodiment, the example in which the respiratory sounds are classified using the peak number and the maximum variance value has been described. However, the feature values used for the classification of the respiratory sounds are not limited thereto. As the feature value, these may be combined with a peak frequency, a peak amplitude value, etc. as appropriate, and the respiratory organs may be classified using the combination.

  For example, as such a classification method, respiratory sound classification (continuous rarity, intermittent rarson, etc.) previously associated with each feature value (number of peaks, maximum variance value, peak frequency, peak amplitude value, etc.) ) Using a database. In this method, based on the correspondence in the database, a discriminant for classifying each disease is obtained by a general machine learning algorithm, and respiratory sounds are classified by the discriminant. As the machine learning algorithm, a k-means method, a support vector machine, a neural network, or the like can be used.

  The present invention is not limited to the above-described embodiments, and various modifications can be made within the scope shown in the claims. That is, embodiments obtained by combining technical means appropriately modified within the scope of the claims are also included in the technical scope of the present invention.

  The respiratory sound analysis apparatus according to the present invention obtains a characteristic value related to a respiratory sound from the peak of the envelope of the frequency spectrum obtained from the measurement waveform data of the respiratory sound, and classifies the respiratory sound based on the characteristic value. It can be suitably used for diagnosing respiratory diseases based on respiratory sounds.

DESCRIPTION OF SYMBOLS 1 Respiratory sound analyzer 2 Body sound measurement part 3 Feature value extraction part (feature extraction means)
4 breathing sound classification part (breathing sound classification means)
5 Classification result display section 31 Section extraction section (section extraction means)
32 Frequency spectrum calculation unit 33 Frequency spectrum smoothing unit (envelope calculation means)
34 peak detector 35 feature value calculator (feature value calculator)
101 Display screen 103 Measurement waveform 104 Respiration timing 105 Classification result f1 to f3 Peak frequency m1 to m3 Peak amplitude value σ1 to σ3 Variance value

Claims (10)

A respiratory sound analysis device for analyzing a subject's respiratory sound,
A feature value extracting means for extracting a feature value of the model waveform by approximating a frequency spectrum of the respiratory sound data obtained by electronically measuring the breathing sound to at least one model waveform;
A respiratory sound analysis device comprising: respiratory sound classification means for classifying the respiratory sounds based on the extracted feature values. The feature value extracting means includes:
Section extracting means for extracting a predetermined section from the respiratory sound data;
Frequency spectrum calculating means for calculating a frequency spectrum of the respiratory sound data of the extracted section;
An envelope calculation means for calculating an envelope of the calculated frequency spectrum;
2. A feature value calculating unit that calculates the feature value based on the Gaussian waveform by approximating the calculated envelope to a Gaussian waveform as the model waveform. Respiratory sound analyzer described in 1. The feature value calculating means calculates, as the feature value, the number of peaks of each Gaussian waveform and the variance value of each Gaussian waveform based on the Gaussian waveform,
The respiratory sound analysis apparatus according to claim 2, wherein the respiratory sound classification unit classifies respiratory sounds based on a maximum variance value among the number of peaks and the variance value.   The respiratory sound analysis apparatus according to any one of claims 1 to 3, further comprising a classification result display unit configured to display a classification result obtained by the respiratory sound classification unit.   The respiratory sound analysis apparatus according to claim 4, wherein the classification result display unit displays a waveform of the respiratory sound data corresponding to the classification result.   The respiratory sound analysis apparatus according to claim 5, wherein the classification result display means displays a disease name corresponding to the classification result.   6. The respiratory sound analysis apparatus according to claim 5, wherein the classification result display means displays timing of expiration and inspiration in the respiratory sound corresponding to the classification result.   A respiratory sound analysis program for causing a computer to function as the feature value extraction means and the respiratory sound classification means in the respiratory sound analysis apparatus according to any one of claims 1 to 7.   A computer-readable recording medium on which the respiratory sound analysis program according to claim 8 is recorded. A respiratory sound analysis method for analyzing a subject's respiratory sound,
A feature value extracting step of extracting a feature value for the model waveform by approximating a frequency spectrum of the respiratory sound data in which the respiratory sound is electronically measured to at least one model waveform;
A respiratory sound analysis method comprising: a respiratory sound classification step of classifying the respiratory sounds based on the extracted feature values.

Images