Biosound sensor system

Abstract

To reduce the impact of internal conduction noise introduced from within the subject's body during exertion on the acquisition of respiratory sounds, and to suppress the generation of friction noise that easily mixes with biological sounds during exertion. [Solution] A biosound sensor system comprising a holding means 1 that can be attached to the head of a subject and has a main input sensor 2 and a reference input sensor 3 installed facing inward on the left ear hook portion 4, and a biosound signal processing means 10 that reconstructs the respiratory sound waveform of the subject. The biosound signal processing means 10 applies a bandpass filter to the received main input and reference input signals to separate vascular sounds and respiratory sounds, applies STFT and HPSS to the extracted high-frequency main input and high-frequency reference input, and processes the obtained main input amplitude spectrogram and reference input amplitude spectrogram, etc. with a spectral subtraction Wiener filter (SSWF) and ISTFT to reconstruct a respiratory sound waveform with internal conduction noise reduced from the main input.

Description

Technical Field

[0001] The present invention relates to a biosound sensor system having a function of reducing body-conducted noise that can measure breathing sounds using two sensor units attached to the head of a subject even in an environment that is easily affected by noise mixed in from the body.

Background Art

[0002] National medical expenses are increasing year by year in proportion to the elderly population, and cost reduction of medical expenses has become an issue. For this purpose, self-medication in daily life (taking responsibility for one's own health and treating minor physical discomfort oneself) is necessary. For the realization of self-medication, the development of wearable devices capable of constantly measuring biological information is desired. However, even for biological information, there are various types of information such as pulse, respiration, blood pressure, and electroencephalogram, and the measurement methods and corresponding devices also vary widely. And devices capable of measuring the pulse rate and respiration rate installed in medical institutions are difficult to handle for those without specialized knowledge other than medical staff and are not easily usable. Therefore, as described in Non-Patent Document 1 (the 36th Life Support Society Conference (LIFE2020-2021), September 16-18, 2021, pp. 168-171), based on the main input consisting of biological sounds and noise measured by two sensor units (condenser microphones) attached to the head of a subject and the reference input consisting of noise, the multi-channel Wiener filter using the adaptation margin m(t, ω) determined by the square difference (|S(t, ω)| 2 -|N(t, ω)| 2 ) was found to be able to improve the SN ratio, and a noise reduction system that can measure the pulse rate and respiration rate even in a noisy environment using a wearable device was proposed.

[0003] Furthermore, the adaptation margin m(t, ω) used in the multi-channel Wiener filter of Non-Patent Document 1 is (|S(t, ω)|2 -|N(t,ω)| 2 ) is only 3 or 100 as determined by (see last line of the right column on page 169 of Non-Patent Document 1), and we noticed that there is a problem in that the noise reduction capability may be inferior. To solve this problem, we developed a biosound sensor system that uses a biosound sensor that can be easily attached to the head of a subject to simultaneously measure vascular sounds and respiratory sounds to obtain the main input, and also measures noise to obtain a reference input, and uses a special Wiener filter designed from the obtained main input and reference input, as described in Patent Document 1 (Japanese Patent Application Publication No. 2024-074099), and succeeded in improving the noise reduction capability from the main input (see paragraphs 0006, 0007 and 0010 in particular).

[0004] However, recent research has shown that it may be possible to evaluate more detailed information about cardiopulmonary status by observing its dynamic changes during exertion (exercise load). For example, in patients with COPD (chronic obstructive pulmonary disease) or asthma, dynamic lung hyperinflation during exertion has been reported to reduce maximum inspiratory volume and increase end-expiratory lung volume. This is because the airways collapse during forced exhalation, causing a phenomenon called air trapping, where air remains in the lungs as inspiration begins before all air is expelled. Furthermore, while airway narrowing can occur during exercise, and changes in breath sounds can be detected, conventional cardiopulmonary exercise testing (CPX) has several drawbacks. To evaluate the respiratory system, a mask connected to an exhaled gas analyzer must be tightly sealed to the mouth, which places a significant burden on the patient. Additionally, it cannot detect adventitious sounds (important sounds that are neither external nor internal noise, but are not present in the breath sounds of healthy individuals) found in the breath sounds of asthma patients and others. Adventitious sounds include low-pitched continuous rales (snoring sounds) and high-pitched continuous rales (wheezing sounds). The former is a low-pitched continuous sound found in patients with COPD and bronchiectasis, while the latter is a high-pitched continuous sound found in patients with bronchial asthma, congestive heart failure, and COPD. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Japanese Patent Publication No. 2024-074099 [Non-patent literature]

[0006] [Non-Patent Document 1] Katsuya Fujimoto, Mikiya Tanaka, Sotaro Shimada, Seiji Nishifuji, Shota Nakajima, "A method for reducing biological sound noise using a multi-channel Wiener filter with adaptive margins," The 36th Annual Meeting of the Japan Society for Life Support (LIFE2020-2021), September 16-18, 2021, pp. 168-171. [Overview of the project] [Problems that the invention aims to solve]

[0007] The present inventors attempted to measure vascular sounds and respiratory sounds during exertion using the biosound sensor system described in Patent Document 1. However, they found that the biosound data, including vascular sounds and respiratory sounds, measured by the wearable biosound sensor described in Patent Document 1, contains noise intruding from both inside and outside the body. In particular, during exertion, internal conduction noise and friction noise introduced from within the body significantly affect the acquisition of biosounds. In view of the above problems, the first objective of the present invention is to reduce the influence of internal conduction noise introduced from within the subject's body during exertion on the acquisition of respiratory sounds. The second objective is to suppress the generation of friction noise, which is easily mixed with biological sounds during exertion. [Means for solving the problem]

[0008] The biosound sensor system of the invention according to claim 1 is A primary input sensor that mainly measures vascular sounds, respiratory sounds, and internal conduction noise to acquire the main input, A reference input sensor that primarily measures internal conduction noise and acquires a reference input, A bandpass filter for the main input extracts a high-frequency main input, which is a high-frequency component, from the main input acquired by the aforementioned main input sensor. A reference input bandpass filter extracts a high-frequency component, which is a high-frequency reference input, from the reference input acquired by the aforementioned reference input sensor. It is designed based on the high-frequency main input extracted by the aforementioned main input bandpass filter and the high-frequency reference input extracted by the aforementioned reference input bandpass filter, and includes a Wiener filter that reduces internal conduction noise from the high-frequency main input and restores respiratory sounds. The aforementioned main input sensor is placed in contact with the area near the mastoid process of the subject. The aforementioned reference input sensor is to be placed in contact with either the inner side of the auricle of the subject or a part of the temporal region opposite to it. The Wiener filter is characterized by being a spectral subtraction type Wiener filter designed from a main input amplitude spectrogram obtained by applying a short-time Fourier transform to the high-frequency main input, a reference input amplitude spectrogram obtained by applying a short-time Fourier transform to the high-frequency reference input, and parameters for adjusting the gains of the main input and the reference input.

[0009] The invention according to claim 2 is a biosound sensor system according to claim 1, The aforementioned main input amplitude spectrogram was obtained by applying a short-time Fourier transform and harmonic / percussion sound separation to the aforementioned high-frequency main input. The aforementioned reference input amplitude spectrogram is characterized by being obtained by applying a short-time Fourier transform and harmonic / percussion sound separation to the high-frequency reference input.

[0010] The invention according to claim 3 is a biosound sensor system according to claim 1 or 2, The device comprises a left ear hook and a right ear hook that can be detachably attached to the left and right ears of the subject, and a retaining means having a connecting part that connects the rear ends of the left ear hook and the right ear hook. The main input sensor is installed facing inward near the rear end of the left or right ear hook, The reference input sensor is characterized in that it is installed facing inward at either the inside of the left ear hook or the inside of the right ear hook.

[0011] The invention according to claim 4 is a biosound sensor system according to claim 3, The holding means is characterized by having an extendable and retractable mechanism that can extend and retract in the left-right direction and can be fixed at multiple positions, and the length of the connecting portion can be adjusted. [Effects of the Invention]

[0012] According to the biosound sensor system of the invention described in claim 1, the main input sensor is brought into contact with the vicinity of the subject's mastoid process, and the reference input sensor is brought into contact with either the inside of the subject's auricle or a part of the temporal region opposite to it. Therefore, the main input, which includes vascular sounds, respiratory sounds, and internal conduction noise, and the reference input, which includes internal conduction noise, can be measured without causing discomfort to the subject. Furthermore, since the Wiener filter is a spectral subtraction type Wiener filter designed from the main input amplitude spectrogram obtained by applying a short-time Fourier transform to the high-frequency main input, the reference input amplitude spectrogram obtained by applying a short-time Fourier transform to the high-frequency reference input, and parameters for adjusting the gains of the main input and reference input, it can accurately subtract the reference input amplitude spectrogram from the main input amplitude spectrogram. Furthermore, by using a spectral subtractive Wiener filter, the influence of internal conduction noise on respiratory sound acquisition can be reduced, allowing for the reconstruction of highly accurate respiratory sounds.

[0013] According to the biological sound sensor system of the invention according to claim 2, in addition to the effects of the invention according to claim 1, the main input amplitude spectrogram is obtained by performing short-time Fourier transform and separating vocal music and percussion sounds on the high-frequency main input, and the reference input amplitude spectrogram is obtained by performing short-time Fourier transform and separating vocal music and percussion sounds on the high-frequency reference input. Therefore, the influence of slight vascular sounds remaining in the high-frequency main input and the high-frequency reference input can be reduced, and more accurate breath sounds can be restored.

[0014] According to the biological sound sensor system of the invention according to claim 3, in addition to the effects of the invention according to claim 1 or 2, it is provided with holding means having a left ear hook portion, a right ear hook portion, and a connecting portion connecting the rear ends of the left ear hook portion and the right ear hook portion. The main input sensor is installed inwardly near the rear end of the left ear hook portion or the right ear hook portion, and the reference input sensor is installed inwardly at any position inside the left ear hook portion or the right ear hook portion. Therefore, the main input sensor and the reference input sensor can be easily attached to the subject's head, and moreover, the main input sensor and the reference input sensor can be surely brought into contact with any part in the mastoid process vicinity of the subject and the temporal bone on the side facing the inside of the subject's auricle.

[0015] According to the biological sound sensor system of the invention according to claim 4, in addition to the effects of the invention according to claim 3, the holding means includes a telescopic mechanism that extends and contracts in the left-right direction and can be fixed at a plurality of positions, and the length of the connecting portion can be adjusted. Therefore, regardless of individual differences, the main input sensor and the reference input sensor can be stably brought into contact with any part in the mastoid process vicinity of the subject and the temporal bone on the side facing the inside of the subject's auricle with appropriate pressure. Therefore, it becomes difficult for vibrations to occur in the main input sensor and the reference input sensor, and in particular, friction noise that is likely to be mixed into the main input and the reference input during labor can be suppressed.

Brief Description of the Drawings

[0016] [Figure 1]Conceptual diagram of a biosound sensor system according to an embodiment. [Figure 2] A diagram showing the retaining means 1 of the embodiment attached to the head of the subject. [Figure 3] A cross-sectional view illustrating the structure of the main input sensor 2 and the reference input sensor 3. [Figure 4] A diagram showing the measurement site where experiments will be conducted to select the position of the reference input sensor 3. [Modes for carrying out the invention]

[0017] Embodiments of the present invention will be described below with reference to examples. [Examples]

[0018] Figure 1 is a conceptual diagram of a biosound sensor system according to an embodiment, and Figure 2 shows the holding means 1 of the embodiment attached to the head of a subject. As shown in Figure 1, the biosound sensor system according to the embodiment comprises a holding means 1 that is attached to the head of the subject, and a biosound signal processing means 10 that processes biosound and internal conduction noise acquired by a main input sensor 2 and a reference input sensor 3 installed on the holding means 1, reduces internal conduction noise from the respiratory sound, and reconstructs the respiratory sound waveform of the subject. Furthermore, the holding means 1 includes a left ear hook 4, a right ear hook 5, a connecting part 6 that connects the rear ends of the left ear hook 4 and the right ear hook 5, and an extension / retraction mechanism 7 for adjusting the length of the connecting part 6, in order to hold the main input sensor 2 and the reference input sensor 3. The extension / retraction mechanism 7 consists of a sliding part 8 and a sliding receiving part 9, and the sliding part 8 and the sliding receiving part 9 are provided with a click mechanism (not shown), allowing it to extend and retract in multiple stages in the left and right directions. Therefore, as shown in Figure 2, after adjusting the length of the extension / retraction mechanism 7 to match the size of the subject's head, the back side of the connecting part 6 is placed on the back of the subject's head, and the left ear hook 4 and the right ear hook 5 are placed on the subject's left and right ears, allowing it to be easily attached to the subject's head.

[0019] The main input sensor 2 is installed facing inward near the rear end of the left ear hook 4, and the reference input sensor 3 is installed facing inward at the top of the left ear hook 4. Therefore, by simply attaching the holding means 1 to the subject's head and placing the left ear hook 4 over the subject's left ear, the main input sensor 2 and the reference input sensor 3 can be stably and appropriately pressed against the area near the subject's mastoid process and the temporal region opposite the inner side of the upper part of the subject's auricle, respectively. Furthermore, since the length of the connection part 6 can be adjusted to match the size of the subject, the misalignment of the main input sensor 2 and the reference input sensor 3 that occurs during biological measurements during exertion is reduced, and frictional noise introduced into the main input and reference input can be reduced. The mastoid process is a cone-shaped protrusion located in the posterior inferior part of the temporal bone. By bringing the main input sensor 2 into contact with the area around the mastoid process, vascular sounds, respiratory sounds, and internal conduction noise can be measured. Similarly, by bringing the reference input sensor 3 into contact with the temporal region opposite the medial side of the upper part of the auricle, internal conduction noise can be measured. Furthermore, bringing the main input sensor 2 or the reference input sensor 3 into contact with the area around the mastoid process or the temporal region opposite the medial side of the upper part of the auricle will not cause discomfort to most subjects.

[0020] Figure 3 is a cross-sectional view illustrating the structure of the main input sensor 2 and the reference input sensor 3. The main input sensor 2 and the reference input sensor 3 are both acoustic sensors that utilize an electret condenser microphone (hereinafter referred to as "ECM") 11, as shown in Figure 3(B). As shown in Figure 3(A), they consist of an ECM 11, an acoustic conduction section 12 made of silicone resin, and a case 13 made of polylactic acid resin. The ECM 11 is housed in a housing 14 and has a wave-receiving surface 15, which is the surface of the diaphragm that detects sound. Furthermore, the housing 14 is housed inside the case 13, and the surface of the wave-receiving surface 15 and the periphery of the housing 14 are covered with the acoustic conduction section 12. The reason for covering the surface of the wave-receiving surface 15 with the acoustic conduction section 12 is that if the ECM 11 is brought directly close to the surface of the human body to acquire vascular sounds, respiratory sounds, and internal conduction noise, etc. (hereinafter referred to as "biological sounds, etc."), the biological sounds, etc. will propagate through the air, which has different acoustic properties from the human body, and will be reflected and attenuated at the interface. Therefore, in this embodiment, the space between the surface of the human body and the surface of the wave receiving surface 15 is filled with an acoustic conducting part 12 made of silicone resin, which has acoustic properties equivalent to those of human skin. This allows biological sounds acquired from the area around the mastoid process and the temporal region opposite the inner side of the upper part of the auricle to be transmitted to the ECM 11 without attenuation, thereby obtaining accurate primary and reference inputs.

[0021] The method for processing biological sounds in the biological sound signal processing means 10 will be described below. As shown in Figure 1, first, the main input signals (signals such as vascular sounds, respiratory sounds, and internal conduction noise) acquired by the main input sensor 2 and the reference input signals (signals such as internal conduction noise) acquired by the reference input sensor 3 are transmitted to the biosound signal processing means 10 via wired or wireless transmission means. The biosound signal processing means 10 then processes the signals of the main input and the reference input according to the following procedure to reconstruct a highly accurate respiratory sound waveform. (Procedure 1) Apply a bandpass filter for the main input and a bandpass filter for the reference input to the received main input signal and reference input signal, respectively, to separate vascular sounds and respiratory sounds. Specifically, when separating vascular sounds, use a bandpass filter with a passband frequency of 20-200 Hz, and when separating respiratory sounds, use a bandpass filter with a passband frequency of 200-800 Hz. (Procedure 2) The high-frequency main input extracted in Procedure 1 is subjected to a Short-Time Fourier Transform (hereinafter referred to as "STFT") and Harmonic Percussive Sound Separation (hereinafter referred to as "HPSS") to obtain the main input amplitude spectrogram (|M(t,f)|). In addition, the high-frequency reference input extracted in Procedure 1 is subjected to STFT and HPSS to obtain the reference input amplitude spectrogram (|R(t,f)|). HPSS is a technique used to reduce slight vascular sounds remaining in the high-frequency main input and high-frequency reference input due to differences in sound pressure.

[0022] (Step 3) Reduce internal conduction noise from the main input using a noise reduction technique with a spectral subtraction Wiener filter (SSWF). The SSWF is an improved multi-channel Wiener filter for reducing internal conduction noise. Specifically, in addition to |M(t,f) and |R(t,f)| obtained in Step 2 above, the subtraction coefficient γ(t,f) is used. The equation (1) for the designed SSWF and the equation (2) for calculating γ(t,f) are as follows. H(t,f)=(|M(t,f)| 2 -γ(t,f)|R(t,f)| 2 ) / |M(t,f)| 2 ...Equation (1) γ(t,f)=|M(t,f)| / |g(t,f)||R(t,f)|·······························································································································formula However, g(t,f) is a parameter that adjusts the gains of the primary input and the reference input. In the example, it was determined from the Euclidean distance (a distance scale representing the straight-line distance between two points) based on the amplitude values ​​of the primary input and the reference input obtained in preliminary experiments. Note that (t,f) represents the components at each time t and frequency f. (Procedure 4) Using equation (1), the reference input amplitude spectrogram is subtracted from the main input amplitude spectrogram, and the respiratory sound waveform is obtained by applying an inverse short-time Fourier transform (ISTFT) to the calculation result. (Step 5) The acquired respiratory sound waveforms are transmitted to a waveform display means, depending on the purpose of their subsequent use. For example, if a doctor or other professional checks these waveforms, they are displayed as respiratory sound waveforms, or the fluctuations in respiratory sounds are calculated and displayed. If a state determination is made based on the displayed respiratory sounds, etc., and the determination result is to be displayed, the data is transmitted to a respiratory sound determination display means.

[0023] Incidentally, while previous experience has shown that the mastoid process is a good location for contacting the main input sensor 2, it was unclear where in the temporal region opposite the inner ear would be a good location for contacting the reference input sensor 3, which measures internal conduction noise, etc. Therefore, we conducted experiments to select areas where it is easier to acquire internal conduction noise. The conditions for the position where the reference input sensor 3, which is suitable for measuring internal conduction noise, is placed (hereinafter referred to as "reference input position conditions") include: (1) it is difficult to acquire vascular sounds and respiratory sounds; (2) there is little individual variation; (3) it is not easily affected by hair or clothing; and (4) it is close to the position where the main input sensor 2 is placed. Condition (4) is to ensure that the noise power acquired by the main input sensor 2 and the reference input sensor 3 are equivalent. Ten measurement sites were selected for the experiment, taking into consideration the location of arteries in the head and neck and the positional conditions of the reference input, as shown in Figure 4. Table 1 shows the results of acquiring biological sounds, etc., by contacting the reference input sensor 3 with the ten selected sites on four men in their 20s, calculating the average amplitude of vascular sounds and respiratory sounds after STFT, and calculating the average value for the four individuals. [Table 1]

[0024] As can be seen from Table 1, areas 3 and 4, located in the temporal region opposite the medial side of the upper auricle, were confirmed to be areas where both vascular sounds and respiratory sounds are difficult to obtain, regardless of individual differences. The reason for this is likely that no major arteries pass through the vicinity of areas 3 and 4, and that these areas are also far from the airways, which are the source of breath sounds. In this embodiment, the installation position of the reference input sensor 3 was set to a position corresponding to part 4.

[0025] Next, we conducted an experiment to evaluate the noise reduction effect and the breath sound distortion reduction effect of the SSWF in the example, and we will now describe the experimental procedure and results. In this experiment, the signal-to-noise ratio (SNR), defined by equation (3), and the signal-to-distortion ratio (SDR), defined by equation (4), were used to evaluate the noise reduction effect and the effect of reducing the distortion of respiratory sounds.

number

number

number

[0026] The experimental method will be explained step by step below. (Experimental Procedure 1) The holding device 1 was attached to the heads of 13 subjects as shown in Figure 2. The main input sensor 2 was brought into contact with the area near the mastoid process, and the reference input sensor 3 was brought into contact with the temporal region opposite the inner side of the upper part of the auricle. (Experimental Procedure 2) To create data for each subject consisting only of respiratory sounds and any other interfering biological sounds, the subjects were kept still for 10 seconds in a quiet environment, and biological sounds were measured using the main input sensor 2 and the reference input sensor 3. From the acquired biological sounds, the high-frequency main input and high-frequency reference input (200~800Hz) were extracted using a biological sound bandpass filter, and then HPSS processing was applied to obtain harmonics, resulting in data consisting only of respiratory sounds and any other interfering biological sounds. (Experimental Procedure 3) To create data consisting only of the primary noise and reference noise for one subject, three types of exercise stress tests were performed in a quiet environment while the subject held their breath for 10 seconds. Noise was measured using primary input sensor 2 and reference input sensor 3. High-frequency primary noise and high-frequency reference noise (200-800 Hz) were extracted from each noise sound data using a biosound bandpass filter, and then HPSS processing was applied to obtain data consisting only of the primary noise and reference noise. The three types of exercise stress tests included the sit-to-stand test, ergometer test, and step-up test, which are commonly used in cardiac rehabilitation and cardiopulmonary function tests within medical institutions. (Experimental Procedure 4) For each subject, test data was created by adding data obtained in Experimental Procedure 3, consisting of only the main noise and only the reference noise, to the data obtained in Experimental Procedure 2, consisting of only the respiratory sounds and only the biological sounds that were mixed in. Then, anticipating the mixing of noises of various magnitudes, the SNR of the added data consisting of only the main noise and only the reference noise (hereinafter referred to as "noise SNR") was set to three patterns: 0 dB, 6 dB, and 9 dB. The noise SNR is defined by the following equation (6), where 0 dB means that the maximum amplitude of the noise, max(|n[t]|), is equal to the maximum amplitude of the biological sound, max(|s[t]|); 6 dB means that max(|n[t]|) is approximately 1 / 4 of max(|s[t]|); and 9 dB means that max(|n[t]|) is approximately 1 / 8 of max(|s[t]|). Noise SNR = 10log10 {max(|s[t]|) / max(|n[t]|)}...Equation (6) Specifically, by incorporating noise obtained from the sit-to-stand test, ergometer test, and step-up test at three different levels, a total of nine test sets were created: test data A1-A3 for the sit-to-stand test, test data B1-B3 for the ergometer test, and test data C1-C3 for the step-up test. (Experimental Procedure 5) For each test data, the conventional noise reduction method using a multi-channel Wiener filter with an adaptive margin (hereinafter referred to as "MWF") and the noise reduction method of the example using SSWF were applied to extract respiratory sounds. The SNR and SDR of the output respiratory sound signals were then determined, and a comparison table was created for each test.

[0027] (Experimental result A) Table 2 is a comparison table of SNR and SDR of breath sounds extracted from test data of the sit-to-stand test, showing the SNR and SDR of breath sounds extracted by applying MWF and SSWF to test data A1-A3. [Table 2] According to Table 2, the SNR for MWF is -0.729 to 14.3 dB and the SDR is 1.31 to 5.76 dB, while the SSWF has an SNR of 0.0334 to 15.2 dB and an SDR of 2.36 to 10.7 dB. Since higher SNR and SDR values ​​indicate less noise influence, it can be seen that SSWF is more effective at extracting breath sounds with less noise influence in all test data A1 to A3. Furthermore, compared to the conventional noise reduction method using MFW, the noise reduction method using SSWF in this example shows better results in both SNR and SDR regardless of the noise level. In addition, subjectively, it appears that the loss of breath sounds increases as the noise decreases with MFW, while the loss of breath sounds is smaller with SSWF.

[0028] (Experimental result B) Table 3 is a comparison table of SNR and SDR of breath sounds extracted from ergometer test data, showing the SNR and SDR of breath sounds extracted by applying MWF and SSWF to test data B1-B3. [Table 3] According to Table 3, the SNR for MWF is 0.999 to 15.1 dB and the SDR is 1.82 to 5.87 dB, while the SSWF has an SNR of 2.00 to 15.1 dB and an SDR of 3.15 to 10.9 dB. Furthermore, in all of the test data B1 to B3, it can be seen that SSWF is able to extract breath sounds with less influence from noise. In addition, compared to the conventional noise reduction method using MFW, the noise reduction method in the SSWF example shows good results in both SNR and SDR regardless of the noise level. Moreover, subjectively, it appears that with MFW, the loss of breath sounds increases as the noise decreases, while with SSWF, the loss of breath sounds is smaller.

[0029] (Experimental result C) Table 4 is a comparison table of SNR and SDR of breath sounds extracted from test data of the step-up test, showing the SNR and SDR of breath sounds extracted by applying MWF and SSWF to test data C1-C3. [Table 4] According to Table 4, the SNR for MWF is -1.58 to 13.3 dB and the SDR is 0.854 to 5.64 dB, while the SSWF has an SNR of -1.16 to 14.5 dB and an SDR of 1.60 to 10.4 dB. Furthermore, in all test data C1 to C3, it can be seen that SSWF is able to extract breath sounds with less influence from noise. In addition, compared to the conventional noise reduction method using MFW, the noise reduction method in the SSWF example shows good results in both SNR and SDR regardless of the noise level. Moreover, subjectively, it appears that with MFW, the loss of breath sounds increases as the noise decreases, while with SSWF, the loss of breath sounds is smaller.

[0030] Tables 2-4 show that in all three types of exercise stress tests, SSWF was able to extract respiratory sounds without being affected by noise, demonstrating superior noise reduction and low degradation in respiratory sound extraction. Furthermore, in all tests, the noise reduction method using SSWF in the examples showed good results in both SNR and SDR, regardless of the noise level. As described above, the biosound sensor system using SSWF of the present invention can reconstruct respiratory sounds with less noise and waveform distortion from biosounds acquired during exertion (exercise load) when noise is easily introduced, and can also capture adrenal noises contained in the respiratory sounds of asthma patients, etc., so it has the potential to be applied to the evaluation of air trapping phenomena and airway narrowing conditions, as well as to screening tests for respiratory diseases that become apparent during exercise load tests.

[0031] The following are examples of modifications of the biosound sensor system described in the embodiment. (1) In this embodiment, a holding means 1 was used that has a main input sensor 2 and a reference input sensor 3 installed facing inward on the left ear hook portion 4 and can be detachably attached to the subject's head. However, any holding means can be used as long as the main input and reference input can be acquired by bringing the main input sensor 2 and the reference input sensor 3 into contact with the vicinity of the subject's mastoid process and the temporal region opposite the inside of the subject's auricle, respectively. (2) In this embodiment, the main input sensor 2 and the reference input sensor 3 are installed only on the left ear hook 4, but they may also be installed only on the right ear hook 5 or on both ear hooks. Furthermore, although the reference input sensor 3 is installed on the upper part of the left ear hook 4, other locations may be more effective in acquiring internal conduction noise depending on the type of exercise stress test and individual differences. In such cases, it is better to install it on either the inside of the left ear hook 4 or the right ear hook 5, depending on the other location. Furthermore, if the location where internal conduction noise can be easily acquired is far from the holding means 1, additional means may be installed. (3) In the embodiment, the telescopic mechanism 7 is composed of a sliding part 8 and a sliding receiving part 9, and the sliding part 8 and the sliding receiving part 9 are provided with a click mechanism, but any structure is acceptable as long as it can extend and retract in the left and right directions, can be fixed in multiple positions, and the length of the connecting part 6 can be adjusted. (4) In the embodiment, there was only one type of retaining means 1, but since the size of the subject's head and auricle and the positional relationship between the auricle and the mastoid process vary, several types of left ear hooks 4 and right ear hooks 5 with different sizes and shapes may be prepared.

[0032] (5) In this embodiment, ECM11 was used for the main input sensor 2 and the reference input sensor 3, but a conventional condenser type, moving coil type, piezoelectric type microphone may be used instead of the ECM11. (6) In the embodiment, the surface of the wave receiving surface 15 and the periphery of the housing 14 were covered with an acoustic conducting part 12 made of silicone resin. However, it is not limited to silicone resin, and polyurethane elastomer or medical-grade silicone may also be used. Any suitable elastic polymer material can be used as long as it is a hydrophobic resin that has acoustic impedance characteristics equivalent to those of human skin after curing. (7) In the embodiment, in step 1, a bandpass filter for the main input and a bandpass filter for the reference input were applied to the main input signal and the reference input signal, respectively, to extract the high-frequency main input and high-frequency reference input. In step 2, the main input amplitude spectrogram and reference input amplitude spectrogram were obtained by STFT, and residual vascular sounds and instantaneous noise were separated by HPSS. In step 3, noise reduction was performed using SSWF with the main input amplitude spectrogram and reference input amplitude spectrogram after HPSS. However, it is also possible to apply STFT to the high-frequency main input and high-frequency reference input in step 2, reduce internal conduction noise from the main input using SSWF in step 3, and then apply HPSS. Furthermore, the order of steps 1 to 3 may be reversed, resulting in the following procedure. (Procedure A) Apply STFT to the received main input signal and reference input signal, respectively, to obtain the main amplitude spectrogram and reference amplitude spectrogram. (Procedure B) Reduce internal conduction noise using the SSWF noise reduction method. (Procedure C) Apply a bandpass filter to separate vascular sounds from respiratory sounds. (Procedure D) Perform HPSS to reduce residual vascular sounds. (8) In the embodiment, the main input amplitude spectrogram is obtained by applying a short-time Fourier transform (STFT) and harmonic / percussion sound separation (HPSS) to the high-frequency main input, and the reference input amplitude spectrogram is obtained by applying STFT and HPSS to the high-frequency reference input. HPSS was also applied in the above modified example 7, but the HPSS processing may be omitted. [Explanation of Symbols]

[0033] 1. Holding mechanism 2. Main input sensor 3. Reference input sensor 4. Left ear hook 5. Right ear hook 6. Connection part 7. Telescopic mechanism 8. Slide section 9. Slide receiving section 10. Bio-sound signal processing means 11 Electret condenser microphone (ECM) 12 Acoustic conduction section 13 Case 14 Housing 15 Wave-receiving surface COPD (Chronic Obstructive Pulmonary Disease) CPX Cardiopulmonary Exercise Test ECM Electret Condenser Microphone HPSS Harmonic Percussive Sound Separation ISTFT Inverse Short-Time Fourier Transform MWF (Multi-channel Wiener Filter) N: Number of samples; SDR: Signal to Distortion Ratio SNR Signal to Noise Ratio Noise SNR: SNR of the noise-only data being added. SSWF (Spectral Subtraction Wiener Filter) STFT (Short-Time Fourier Transform) f frequency t time a[t] A signal obtained by combining breathing sounds and noise and applying noise reduction processing. b[t] Signal after BPF and HPSS g(t,f) is a parameter that adjusts the gain of the main input and reference input. H(t,f) spectral subtraction Wiener filter |M(t,f)| Main input amplitude spectrogram max(|n[t]|) Maximum amplitude of noise max(|s[t]|) Maximum amplitude of biological sound |R(t,f)| Reference input amplitude spectrogram s[t] Breathing sound signal after noise reduction processing z[t] A signal obtained by combining breathing sounds and noise, with noise reduction processing applied. γ(t,f) Subtraction coefficient λ Adjustment term for the volume difference before and after noise reduction processing

Claims

1. A primary input sensor that mainly measures vascular sounds, respiratory sounds, and internal conduction noise to acquire the main input, A reference input sensor that primarily measures internal conduction noise and acquires a reference input, A bandpass filter for the main input extracts a high-frequency main input, which is a high-frequency component, from the main input acquired by the aforementioned main input sensor. A reference input bandpass filter extracts a high-frequency component, which is a high-frequency reference input, from the reference input acquired by the aforementioned reference input sensor. It is designed based on the high-frequency main input extracted by the aforementioned main input bandpass filter and the high-frequency reference input extracted by the aforementioned reference input bandpass filter, and includes a Wiener filter that reduces internal conduction noise from the high-frequency main input and restores respiratory sounds. The aforementioned main input sensor is placed in contact with the area near the mastoid process of the subject. The aforementioned reference input sensor is to be placed in contact with either the inner side of the auricle of the subject or a part of the temporal region opposite to it. The Wiener filter is a spectral subtractive Wiener filter designed from a main input amplitude spectrogram obtained by applying a short-time Fourier transform to the high-frequency main input, a reference input amplitude spectrogram obtained by applying a short-time Fourier transform to the high-frequency reference input, and parameters for adjusting the gains of the main input and the reference input. A biosound sensor system characterized by the following features.

2. The aforementioned main input amplitude spectrogram was obtained by applying a short-time Fourier transform and harmonic / percussion sound separation to the aforementioned high-frequency main input. The aforementioned reference input amplitude spectrogram was obtained by applying a short-time Fourier transform and harmonic / percussion sound separation to the high-frequency reference input. The biosound sensor system according to feature 1.

3. The device comprises a left ear hook and a right ear hook that can be detachably attached to the left and right ears of the subject, and a retaining means having a connecting part that connects the rear ends of the left ear hook and the right ear hook. The main input sensor is installed facing inward near the rear end of the left or right ear hook, The reference input sensor is installed facing inward at either the inside of the left ear hook or the inside of the right ear hook. The biosound sensor system according to claim 1 or 2, characterized by the above.

4. The holding means is equipped with an extendable and retractable mechanism that can extend and retract in the left-right direction and can be fixed in multiple positions, and the length of the connecting portion is adjustable. The biosound sensor system according to feature 3.

Images