JP2009106574A - Apparatus and program for respiratory sound analysis - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a respiratory sound analysis apparatus which precisely detects rhonchus. <P>SOLUTION: An analytic signal is computed by carrying out Hilbert transformation of respiratory sound data. From the analytic signal, an envelope, an instantaneous frequency, and an instantaneous band width are computed. From the envelope, a peak which becomes a candidate of the rhonchus is extracted. Corresponding to the extracted peak, at least one of the instantaneous frequency and the instantaneous band width is computed. On the basis of at least one of the computed instantaneous frequency and instantaneous band width, the rhonchus is detected. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to a respiratory sound analysis device and a program.

Listening and diagnosing respiratory sounds is a diagnostic method that has long been used as auscultation. Abnormal sounds in breathing sounds are called side noises, and various types of sounds are known. Among sub-noises, ra-tone is important, and automatic detection has been attempted so far (see, for example, Patent Document 1).
Special table 2001-505085 gazette

  The respiration signal includes various noise signals. For example, there are a wide variety of voices, coughs, heart sounds, and friction sounds that rub the microphone. How to detect this noise erroneously and accurately detect the target sound is a big problem.

  Among the rales, it is difficult to detect intermittent rales (cracks), and in particular, there is a problem that the detection accuracy of coarse crackles (water bubbles) is low.

  The present invention has been made in view of the above-described problems, and an object of the present invention is to provide a respiratory sound analysis apparatus that can accurately detect rales.

  The respiratory sound analyzer of the present invention measures a respiratory sound of a subject and outputs respiratory sound data, and analyzes the respiratory sound based on the respiratory sound data measured by the respiratory sound measuring unit. A breathing sound analyzing unit that performs Hilbert transform on the breathing sound data to calculate an analysis signal, calculates an envelope, an instantaneous frequency, and an instantaneous bandwidth from the analysis signal, A peak that is a candidate for a ra sound is extracted from a line, at least one of an instantaneous frequency or an instantaneous bandwidth is calculated corresponding to the extracted peak, and based on at least one of the calculated instantaneous frequency or instantaneous bandwidth It is characterized by detecting the noise.

  The program of the present invention includes a step of calculating an analysis signal by performing Hilbert transform on input respiratory sound data, a step of calculating an envelope, an instantaneous frequency and an instantaneous bandwidth from the analysis signal, and a candidate for a ra sound from the envelope Extracting at least one of an instantaneous frequency or an instantaneous bandwidth corresponding to the extracted peak, and a step based on at least one of the calculated instantaneous frequency or instantaneous bandwidth. And a step of detecting a sound by a computer.

  According to the present invention, by using the instantaneous frequency and the instantaneous bandwidth, it is possible to accurately detect the rale.

  Hereinafter, although an embodiment of the present invention is described based on a drawing, it is an example and is not limited to this embodiment.

(First embodiment)
[Device configuration]
FIG. 1 is a configuration diagram of a respiratory sound analysis apparatus according to the present embodiment. The respiratory sound analysis device 1 includes a respiratory sound measurement unit 10 and a respiratory sound analysis unit 30 and is connected to each other via a communication medium L. This communication medium L may be wired or wireless. When connected wirelessly, the chance of giving unnecessary vibration to the sensor used in the respiratory sound measurement unit 10 can be reduced, and the restriction on the behavior of the subject can be reduced.

  The respiratory sound measurement unit 10 is amplified by a contact microphone 11 such as a condenser microphone that converts an input respiratory sound into a respiratory sound signal, an amplifier 12 that amplifies the respiratory sound signal converted by the microphone 11, and an amplifier 12. Smoothing circuit 13 for smoothing the respiratory sound signal, A / D converter 14 for converting the respiratory sound signal processed by the smoothing circuit 13 into respiratory sound data, and respiratory sound data converted by the A / D converter 14 Is transmitted to the respiratory sound analysis unit 30. Here, the A / D converter 14 generates respiratory sound data from the respiratory sound signal at a sampling frequency Fs = 11.025 kHz, for example.

  The breathing sound analysis unit 30 receives the breathing sound data from the breathing sound measuring unit 10, the CPU 32 that processes the breathing sound data received by the I / F 31 according to the program, the program necessary for the processing by the CPU 32, ROM 33 for storing data, RAM 34 for temporarily storing programs and data necessary for processing in the CPU 32, and external storage device 35 for storing processing results in the CPU 32 in a hard disk, DVD-R, CD-R, etc. , And an input unit 36 for inputting various data, and a display unit 37 for displaying processing results in the CPU 32 and the like.

  The respiratory sound analysis unit 30 may be configured with a dedicated information processing apparatus or a general-purpose personal computer. If it is a personal computer, it is preferably a personal digital assistant (PDA) that can be easily carried.

  In the present embodiment, the A / D converter 14 is provided in the respiratory sound measurement unit 10, but an A / D converter may be provided in the respiratory sound analysis unit 30.

[Respiratory sound measurement processing]
FIG. 2 is a flowchart of the respiratory sound measurement process according to the present embodiment. This respiration sound measurement processing flow is a flow executed by the CPU 32 based on a respiration sound measurement program in the ROM 33. It is assumed that an ID or the like for specifying the subject is input in advance by the input unit 36.

  First, the CPU 32 determines whether or not the start of breathing sound measurement is instructed from the input unit 36 (step S10). If it is determined that the start of breathing sound measurement has been instructed (step S10; Yes), the CPU 32 starts an operation of storing the breathing sound data from the breathing sound measuring unit 10 in the external storage device 35 (step S41). At this time, the respiratory sound data is stored in association with the subject ID and time. If it is determined that the start of breathing sound measurement is not instructed (step S10; No), the process returns to step S10 and waits until the start of breathing sound measurement is instructed.

  Next, the CPU 32 determines whether or not the end of breathing sound measurement has been instructed (step S12). If it is determined that the end of breathing sound measurement has been instructed (step S12; Yes), the CPU 32 ends the operation of storing the breathing sound data from the breathing sound measurement unit 10 in the external storage device 35 (step S13). If it is determined that the end of breathing sound measurement is not instructed (step S12; No), the process returns to step S12 and waits until the end of breathing sound measurement is instructed.

[Analysis parameters]
First, parameters used for the analysis of the present invention will be described.

  The analytic signal z (t) is a complex signal, whose real part is equal to the original signal, and whose imaginary part is the Hilbert transform of the original signal. If the original signal is x (t) and the analytic signal of x (t) is z (t),

It is expressed. However,

A (t) and ψ (t) denote the envelope and instantaneous phase of x (t), respectively.

  The instantaneous frequency f (t) is

It is represented by

  The instantaneous bandwidth b (t) is

It is represented by

[La Sound Detection Concept]
A notable feature of ra sound is temporal localization. In the present embodiment, the sound is detected using an index called an energy concentration index ECI (Energy Concentration Index).

  The energy concentration index ECI is the energy E1 of the original waveform in the vicinity of the maximum value (maximum value) of the envelope, and the energy E2 of the original waveform in the time width of about one ra sound centered on the maximum value of the envelope. It is expressed as a ratio (E1 / E2). The energy concentration index ECI is an index of how much the energy of the waveform is concentrated near the maximum value, and can be used as an index of rales.

  In this algorithm, the reciprocal of the average value of the instantaneous frequency in the vicinity of the maximum value of the envelope is T (T is an estimated value of the oscillation period of the pulse waveform), and the value of E1 is twice as long as T. Energy (sum of squares of data), and the value of E2 is energy included in a time width of 6 times T.

  The instantaneous frequency stably indicates the (main) frequency of the ra sound in the ra sound portion. It varies in small increments where there is no signal. The instantaneous bandwidth becomes almost zero at the peak of the rale. This shows a bowl-shaped change that protrudes downward in the ra sound. It varies greatly in areas where there is no signal.

  In the present embodiment, the energy concentration index ECI is equal to or greater than a certain threshold (the waveform has a large temporal localization), the fluctuation of the instantaneous frequency is small, the instantaneous bandwidth is small, and the vicinity of the candidate point. If the integrated value of the instantaneous bandwidth (sum of data) (hereinafter referred to as BS) is greater than or equal to a certain threshold value (determining whether the change in instantaneous bandwidth is bowl-shaped) or not, it is determined to be a ra sound. .

[Respiratory sound analysis processing in rale detection]
FIG. 3 is a flowchart of the respiratory sound analysis process according to the first embodiment.

  This breathing sound analysis process is a process of analyzing the acquired breathing sound data and detecting a rale. It is executed by the CPU 32 based on a respiratory sound analysis program (program of the present invention) in the ROM 33. In addition, ID etc. which specify a subject are previously input by the input part 36.

  First, the CPU 32 determines whether or not an instruction for breathing sound analysis is input from the input unit 36 (step S20). If it is determined that an instruction for breathing sound analysis has been input (step S20; Yes), the CPU 32 reads the breathing sound data corresponding to the subject ID from the external storage device 35 and loads it into the RAM 34 (step S21). If it is determined that the instruction for breathing sound analysis is not input (step S20; No), the process returns to step S20 and waits until the instruction for breathing sound analysis is input.

  Next, the CPU 32 classifies the respiratory sound data (sampling frequency Fs = 11.025 kHz) loaded in the RAM 34 every 131072 points (step S22).

  Next, the CPU 32 applies a high-pass filter with a cutoff frequency of 50 Hz in order to remove noise such as heart sounds in each section (step S23).

  Next, the CPU 32 performs Fourier transform on the data, obtains an analytic signal z (t) by performing inverse Fourier transform with the DC component being 1 time, the positive frequency component being 2 times, and the negative frequency component being 0 ( Step S24). In addition to the method using the Fourier transform, there is also a method using a Hilbert transformer (digital filter).

  Next, the CPU 32 calculates the envelope A (t), the instantaneous frequency f (t), and the instantaneous bandwidth b (t) by the above formula. The derivative was replaced with the difference. Each result is shown in FIG. 4 (a) is the original signal x (t), FIG. 4 (b) is the envelope a (t), FIG. 4 (c) is the energy concentration index ECI, FIG. 4 (d) is the instantaneous frequency f (t), FIG. 4 (e) shows the instantaneous bandwidth b (t). Further, the squared amount v (t) is calculated by taking the time derivative (difference) of f (t) (step S25).

  Next, the CPU 32 calculates a moving average As (t) of A (t) with a triangular weight and 55 widths. Similarly, for b (t) and v (t), the moving averages bs (t) and vs (t) are calculated with a width of 35 points (step S26).

  Next, the CPU 32 searches for a position where the moving average As (t) is maximized, and sets it as a candidate for the position of the sound (step S27).

  Next, the CPU 32 calculates an energy concentration index ECI at each candidate position. The average value FA of f (t) in the data width of 32 points is obtained centering on the candidate position. ECI = E1 / E2, and the number of data N1 = 2 × Fs / FA for obtaining E1, and the number of data N2 = 6Fs / FA for obtaining E2. In the original signal x (t), the sum of squares of x (t) in the width of the N1 point with the candidate position as the center is E1, and the square sum of x (t) in the width of the N2 point with the candidate position as the center is E2. As an energy concentration index ECI (step S28).

  Next, the CPU 32 calculates an integral value BS of the instantaneous bandwidth at the candidate position. In the instantaneous bandwidth b (t), the sum of b (t) in the width of 442 points with the candidate position as the center is obtained and set as BS (step S29).

At the candidate position of the sound, the energy concentration index ECI is 0.65 or more, vs (t) is 3.04 × 10 ¹⁰ or less, the moving average bs (t) of the instantaneous bandwidth is 154 or less, and the instantaneous near the candidate point If the integrated value BS of the bandwidth is 12 or more, the waveform at that position is determined as a ra sound (step S30). In FIG. 4, a triangular mark is shown on the original signal x (t) in FIG.

  As described above, according to the present embodiment, by using the instantaneous frequency and the instantaneous bandwidth, the rale can be detected with high accuracy. For this reason, a portion without noise (noise portion) can be discriminated, and an algorithm robust to changes in the amplitude of the input signal can be realized. In addition, it is possible to easily obtain the (main) frequency of the ra sound using the instantaneous frequency.

  In the present embodiment, a determination as to whether or not the instantaneous bandwidth change is bowl-shaped is also incorporated, which is preferable for improving accuracy.

(Second Embodiment)
The apparatus configuration and the respiratory sound measurement process are the same as those in the first embodiment, and a description thereof will be omitted.

[La Sound Detection Concept]
In the first embodiment, the rale is detected using the energy concentration index ECI. However, in the second embodiment, the rae sound is detected using a second moment around the peak value of the crest index or envelope described later. Detection is performed.

  The crest index CI (wave height index) is the same as the wave height factor, and is defined as a value obtained by dividing the maximum value of the envelope by the effective value of the current waveform. The closer the waveform is to the pulse, the larger the value.

  The second moment is the same as the variance in probability and indicates the extent of the distribution spread. The ra sound envelope shows a relatively low value.

  In the present embodiment, at the candidate point of the ra sound, the crest index CI is greater than or equal to a certain threshold value or less than a certain threshold value (the temporal localization of the waveform is large), the fluctuation of the instantaneous frequency is small, and the instantaneous When the bandwidth is small, it is determined that the sound is a rale.

[Respiratory sound analysis processing in rale detection]
FIG. 5 is a flowchart of the breathing sound analysis process according to the second embodiment. As an example, a case where a rale is detected using the crest index CI will be described.

  The respiratory sound analysis process is a process of analyzing the acquired respiratory sound data and detecting a rale sound. It is executed by the CPU 32 based on the respiratory sound analysis program in the ROM 33. In addition, ID etc. which specify a subject by the input part 36 shall be previously input.

  First, the CPU 32 determines whether or not an instruction for breathing sound analysis is input from the input unit 36 (step S40). If it is determined that an instruction for breathing sound analysis has been input (step S40; Yes), the CPU 32 reads the breathing sound data corresponding to the subject ID from the external storage device 35 and loads it into the RAM 34 (step S41). If it is determined that the instruction for breathing sound analysis is not input (step S40; No), the process returns to step S40 and waits until the instruction for breathing sound analysis is input.

  Next, the CPU 32 classifies the respiratory sound data (sampling frequency Fs = 11.025 kHz) loaded in the RAM 34 every 131072 points (step S42).

  Next, the CPU 32 applies a high-pass filter with a cutoff frequency of 50 Hz in order to remove noise such as heart sounds in each section (step S43).

  Next, the CPU 32 performs Fourier transform on the data, obtains an analytic signal z (t) by performing inverse Fourier transform with the DC component being 1 time, the positive frequency component being 2 times, and the negative frequency component being 0 ( Step S44). In addition to the method using the Fourier transform, there is also a method using a Hilbert transformer (digital filter).

  Next, the CPU 32 calculates the envelope A (t), the instantaneous frequency f (t), and the instantaneous bandwidth b (t) by the above formula. The derivative was replaced with the difference. Further, the squared amount v (t) is calculated by taking the time derivative (difference) of f (t) (step S45).

  Next, the CPU 32 calculates a moving average As (t) of A (t) with a triangular weight and 55 widths. Similarly, for b (t) and v (t), the moving averages bs (t) and vs (t) are calculated with a width of 35 points (step S46).

  Next, the CPU 32 searches for a position where the moving average As (t) is maximized, and sets it as a candidate for the position of the sound. However, an average value of the moving average As (t) is obtained, and a peak of 3.2 times or less is excluded from the candidates (step S47).

  Next, the CPU 32 calculates the crest index CI of each peak. The data length for calculating the crest index CI is a width of 1024 points centering on the peak position of the waveform (ra sound candidate) (step S48).

If the crest index CI is 2.6 or more, the instantaneous bandwidth moving average bs (t) is 154 or less, and vs (t) is 3.04 × 10 ¹⁰ or less The waveform is determined to be a ra sound (step S49).

  As described above, according to the present embodiment, by using the instantaneous frequency and the instantaneous bandwidth, the rale can be detected with high accuracy. For this reason, a portion without noise (noise portion) can be discriminated, and an algorithm robust to changes in the amplitude of the input signal can be realized. In addition, it is possible to easily obtain the (main) frequency of the ra sound using the instantaneous frequency.

  The range (data length) for calculating the second moment of the crest index or envelope may be a fixed length, but the data length should be determined for each peak based on the peak of the waveform and the instantaneous frequency in the vicinity. (For example, a constant multiple of the reciprocal of the instantaneous frequency of the peak) can improve the detection capability. In addition, it is possible to perform processing such as selectively detecting a ra sound using an instantaneous frequency, or reducing false detection due to noise.

  The index related to the roughness of the waveform of the instantaneous frequency in the present invention refers to, for example, a deviation in a certain section of the signal. The smaller the deviation, the less the fluctuation of the instantaneous frequency, and the easier it is to determine that the sound is a rale.

  The waveform shape of the instantaneous bandwidth in the present invention may be a low frequency component of a signal using a low filter, a moving average of a section, or an envelope.

  In the case of a waveform with a large instantaneous frequency, a convex area may appear even in the portion that is not the envelope that makes the envelope, but in this case, the waveform of the instantaneous frequency becomes rough, so Is not detected. The detection method of the present invention may be used alone or in combination with other methods.

  Regarding the detection of fine crackle (high-tone intermittent sound), the detection accuracy can be improved by performing waveform analysis through a high-pass filter with an appropriate cut-off frequency, and then performing this breathing sound analysis processing.

  The respiratory sound signal used in the present invention can be collected by, for example, an electronic stethoscope. The collecting device does not necessarily require a function of hearing with an ear like an electronic stethoscope, and may be any device that can collect a sound signal in the body from the body surface.

It is a block diagram of the respiratory sound analysis apparatus which concerns on this embodiment. It is a flowchart of the respiratory sound measurement process which concerns on this embodiment. It is a flowchart of the respiratory sound analysis process which concerns on 1st Embodiment. It is a conceptual diagram which shows each waveform which concerns on 1st Embodiment. It is a flowchart of the respiratory sound analysis process which concerns on 2nd Embodiment.

Explanation of symbols

10 Respiratory Sound Measurement Unit 30 Respiratory Sound Analysis Unit 32 CPU
33 ROM

Claims (8)

A respiratory sound measurement unit that measures the respiratory sound of the subject and outputs respiratory sound data;
A respiratory sound analyzer for analyzing respiratory sounds based on the respiratory sound data measured by the respiratory sound measuring unit;
Have
The respiratory sound analysis unit
Analyzing the respiratory sound data by Hilbert transform to calculate the analysis signal,
Calculate the envelope, instantaneous frequency and instantaneous bandwidth from the analysis signal,
Extract the candidate peak of ra sound from the envelope,
Calculate at least one of instantaneous frequency or instantaneous bandwidth corresponding to the extracted peak,
A respiratory sound analyzing apparatus that detects rales based on at least one of the calculated instantaneous frequency or instantaneous bandwidth. The respiratory sound analysis unit
Calculate at least one of energy concentration index, crest index or second moment corresponding to the extracted peak,
The respiratory sound analysis apparatus according to claim 1, wherein a rarity is detected including at least one of the calculated energy concentration index, crest index, or second moment. The respiratory sound analysis apparatus according to claim 1, wherein a rale is detected based on an index related to a waveform roughness at the instantaneous frequency. The respiratory sound analysis apparatus according to any one of claims 1 to 3, wherein a rale is detected based on an index indicating that the waveform shape of the instantaneous bandwidth is convex downward. The respiratory sound analysis unit calculates an integrated value of an instantaneous bandwidth corresponding to a peak extracted as the ra sound candidate, and detects ra sound including the calculated integrated value of the instantaneous bandwidth. The respiratory sound analysis device according to any one of claims 1 to 4. The said breathing sound analysis part calculates the moving average of the said envelope, and extracts the peak of the calculated moving average as a peak which becomes a ra sound candidate, The said any one of Claims 1-5 characterized by the above-mentioned. The respiratory sound analysis device described. The data length when calculating the energy concentration index, the crest index, or the second moment is determined based on an instantaneous frequency at the extracted peak. Breathing sound analyzer. Calculating the analysis signal by Hilbert transforming the input respiratory sound data;
Calculating an envelope, an instantaneous frequency and an instantaneous bandwidth from the analysis signal;
Extracting a peak that is a candidate for a ra sound from an envelope; and
Calculating at least one of an instantaneous frequency or an instantaneous bandwidth corresponding to the extracted peak;
Detecting rales based on at least one of the calculated instantaneous frequency or instantaneous bandwidth;
A program that causes a computer to execute.

Images