JP7637408B2 - Biosignal analyzer, computer program, and recording medium - Google Patents

Description

The present invention relates to a technique for estimating a biological state, and more particularly to a biological signal analyzer, a computer program , and a recording medium using an index related to the apex beat.

The apex beat reflects the movement of the left ventricular apex and is used to evaluate cardiac, primarily left ventricular, function. The apex beat is generally objectively measured by the apex cardiogram (ACG) recorded using the mechanocardiogram (MCG) method, but since the apex cardiogram captures the movement of the chest wall via various tissues between the heart and the chest wall, it is easily affected by recording skill. In addition, most mechanocardiogram recording devices are relatively large, and there are restrictions such as the need to perform measurements in a soundproof room.

In consideration of this point, Patent Document 1 discloses a bioinformation measuring device for an apexcardiogram that is equipped with an ultrasound transmitter that transmits ultrasound signals from the body surface toward the vicinity of the apex of the heart and a receiving sensor that receives reflected ultrasound, and claims that the device can easily create an apexcardiogram even in special environments such as soundproof rooms.

Meanwhile, in Patent Documents 2 to 5, the present inventors have proposed a technology for capturing vibrations occurring on the surface of a person's back in an unconstrained manner and estimating the person's condition by analyzing the vibrations. The vibrations occurring on the surface of a person's back are the result of the propagation of vibrations within the body, such as the heart and aorta, and contain information on the systole and diastole of the atrium and ventricle, as well as elasticity information and reflected wave information on the blood vessel walls that act as auxiliary pumps for circulation.

In Patent Document 2, a predetermined time width is applied to the time series waveform of the back surface pulse wave near 1 Hz extracted from the vibration (biological signal) propagating through the body surface, and a sliding calculation is performed to obtain a time series waveform of the frequency gradient, and the biological condition is estimated from the tendency of the change, for example, whether the amplitude is increasing or decreasing. It also discloses that the biological signal is frequency analyzed to obtain the power spectrum of each frequency corresponding to the function adjustment signal, fatigue acceptance signal, and activity adjustment signal belonging to a predetermined ULF band (very low frequency band) to VLF band (very low frequency band), and the human condition is judged from the time series change of each power spectrum.

Patent Documents 3 and 4 disclose means for determining the homeostatic function level, while Patent Document 5 discloses a sound/vibration information collecting mechanism having a resonance layer equipped with a natural oscillator having a natural frequency corresponding to sound/vibration information of a biosignal.
In addition, Non-Patent Document 1 shows that heart sounds are composed of irregular vibrations. Non-Patent Document 2 discloses that the waveform of the above apexcardiogram is classified and used for diagnosis. Non-Patent Document 3 explains the fluctuation of the cardiac cycle.

WO2012/165660 publication JP 2011-167362 A JP 2014-117425 A JP 2014-223271 A JP 2016-26516 A

Nobuhiro, Y. et al. Development of Physical Condition Fluctuation Prediction Model Using Trunk Biosignals, Proceeding of Asia-Pacific Vibration Conference 2019 (2019) Yoshikawa, J. et al. Physical Examination in Cardiology, Bunkodo Shoten Co.,Ltd.,79-95 (2005) Kobayashi M., Musha T.: 1/f fluctuation of heart beat period. IEEE Transactions on biomedical engineering. BME, 29, 45-457 (1982)

The device disclosed in Patent Document 1 uses 192 ultrasonic transducers, and 12 transducers are grouped together and scanned by shifting them sequentially to create an apexcardiogram. Patent Document 1 also discloses a device that uses multiple rows of 192 ultrasonic transducers in order to create an apexcardiogram with higher accuracy. Patent Document 1 states that it is difficult to obtain a desired apexcardiogram without using such a large number of ultrasonic transducers. Therefore, the product is inevitably very expensive and not practical. In addition, in order to create a clear apexcardiogram, it is necessary to consider the detection of the rib position, the transmission frequency is limited, and the position of the receiving element is also limited, so the device must satisfy these conditions, which also leads to a complex structure and increased manufacturing costs.

Patent Document 1 also mentions that an advantage of this device is that it can be used by just one person, whereas conventional electrocardiogram recording devices require multiple people to operate. However, the ultrasound transducer needs to be operated by a doctor or technician to scan, which makes the measurement time-consuming and only provides data for the period of scanning. In other words, this device is not suitable for capturing the apical pulsation of the subject over a long period of time.

Meanwhile, the sensors disclosed in Patent Documents 2 to 5 that capture vibrations occurring on the body surface in an unconstrained manner consist of a three-dimensional knitted fabric, a film surrounding the three-dimensional knitted fabric, a microphone, etc., and can acquire biosignals via the body surface simply by placing them in contact with the human body. In other words, when attached to the human body, biosignal data can be obtained without any manipulation by a doctor or other such person. However, Patent Documents 2 to 5 mainly analyze autonomic nerve function and heart rate variability, and do not analyze apical beat, which is the movement of the left ventricular apex.

In addition, the sensors used in Patent Documents 2 to 5 are suitable for passively capturing acoustic vibrations of 20 to 100 Hz propagating from the surface of the human body on the back, but are not suitable for extracting apical pulsation, which is an internal vibration of 10 Hz or less.
In addition, in order to make a diagnosis using the apexcardiogram of Non-Patent Document 3, professional knowledge is required to read the waveform of the apexcardiogram, and the judgment results may differ depending on the judge, so there is room for improvement in terms of judgment accuracy and speed. Therefore, diagnosis using the apexcardiogram is only used as a supplementary tool. Furthermore, as information related to the apexcardiogram, such an apexcardiogram is used for diagnosis, albeit in a supplementary manner, but indices related to the apexcardiogram other than the apexcardiogram are not used to estimate various health conditions of a person (presence or absence of diseases such as heart disease, identification of diseases, physical condition, etc.).

The present invention has been made in consideration of the above, and aims to provide a biosignal analysis device, computer program, and recording medium that can be easily and inexpensively manufactured, is capable of long-term measurements, analyzes biosignals collected by a biosignal detection sensor, and extracts information related to the apical pulsation of the heart, and can also estimate various health conditions of a person.

To solve the above problems, the inventors developed a biosignal detection sensor that uses a three-dimensional knitted fabric and a microphone, etc., and can acquire biosignal data of 0.5 to 80 Hz. However, biosignal detection sensors only passively capture acoustic vibrations propagating from the body surface, and the captured biosignal data includes various biosounds and internal vibrations. As a result of expanding the collectable frequency band to the low frequency band, data related to the apex beat was included, but a new problem was faced in that the data was hidden by the heart sounds, making it difficult to extract only the apex beat data. Therefore, the inventors conducted extensive research to solve this problem, and have completed the present invention.

That is, the biosignal analyzer of the present invention is
a frequency analysis means for performing frequency analysis on biosignal data obtained by a biosignal detection sensor via a body surface;
The present invention is characterized in that it has a boundary frequency identifying means for determining, from a result of frequency analysis of the biological signal data obtained from the frequency analysis means, a boundary frequency between vibrations caused by apex beat and vibrations caused by heart sounds in the biological signal data.

The boundary frequency identification means
It is preferable that the frequency analysis method further comprises means for determining a sudden change portion of a power spectrum that is a boundary between harmonic vibration and irregular vibration in the frequency analysis result, and for specifying the boundary frequency based on the sudden change portion.
The boundary frequency identification means
It is preferable to include a means for determining a sudden change part of the power spectrum by adding a result of frequency analysis of heart sound data measured at the same time to the result of the frequency analysis.

The boundary frequency identification means
It is preferable to have a double logarithmic axis display means for displaying on a double logarithmic axis, using a logarithmic difference method, waveforms obtained by averaging the results of frequency analysis of the biological signal data and the heart sound data, and a sudden change part identification means for determining a fluctuation change point from the waveform displayed on the double logarithmic axis and identifying this fluctuation change point as a sudden change part of the power spectrum.

A short-time Fourier transform is applied as the frequency analysis means,
The boundary frequency identifying means preferably includes means for determining abrupt change portions of the power spectrum from the analysis results of the short-time Fourier transform.
the frequency analysis means includes means for outputting an analysis result of the short-time Fourier transform into image data indicating the degree of fluctuation in time, frequency, and power spectrum;
The boundary frequency identifying means preferably includes means for determining an abrupt change portion of the power spectrum from the image data.

a correlation data storage unit in which correlation data relating to the boundary frequency is stored;
It is preferable that the apparatus further comprises a measurement time condition estimating means for estimating the health condition of the subject at the time of measurement by comparing with the correlation data.

the correlation data is correlation data between the boundary frequency and a heart rate,
It is preferable that the measurement-time state estimation means includes means for estimating the boundary frequency of the subject at the time of measurement by comparing the subject's heart rate at the time of measurement with correlation data between the boundary frequency and the heart rate.
the correlation data is correlation data between the boundary frequency and a fluctuation characteristic of heart rate variability,
It is preferable that the measurement-time state estimation means includes means for estimating the fluctuation characteristics of the heart rate fluctuation of the subject at the time of measurement by comparing the boundary frequency at the time of measurement of the subject with correlation data between the boundary frequency and the fluctuation characteristics of the heart rate fluctuation.

an apex pulsation waveform extraction means for filtering the biological signal data by setting the boundary frequency identified by the boundary frequency identification means as an upper limit value, and obtaining a waveform of vibration caused by the apex pulsation;
A waveform classification means for classifying the waveform of the vibration generated by the apex beat of the heart,
It is preferable that the measurement-time condition estimation means includes means for classifying, by the waveform classification means, the waveform of vibration generated by the apex beat of the subject at the time of measurement, which is obtained by the apex beat waveform extraction means, and estimating the health condition of the subject at the time of measurement based on data of the waveform classification result at the time of measurement.

The correlation data storage unit stores correlation data between the waveform classification result and a health condition in advance,
It is preferable that the measurement-time condition estimation means includes means for classifying, by the waveform classification means, the waveform of vibration generated by the apex beat of the subject at the time of measurement, which is obtained by the apex beat waveform extraction means, and for estimating the health condition of the subject at the time of measurement by comparing the waveform classification result at the time of measurement with pre-stored correlation data between the waveform classification result and health condition.

a model creation means for generating an estimation model by machine learning to estimate the health condition from the waveform information by using the waveform information output by the apex beat waveform extraction means and the health condition as teacher data,
It is preferable that the measurement condition estimation means includes means for inputting waveform information of vibrations generated by the apex beat of the subject at the time of measurement, using an estimation model created by the model creation means, and outputting a value of the health condition at the time of measurement from the acquired waveform information at the time of measurement.

The waveform classification means includes the following means (A) and (B):
(A) a means for classifying the waveform by a mathematical approach using a Fourier series expansion;
(B) A means for classifying the waveform by combining one or more pieces of data obtained by the following engineering approaches (1) to (4):
(1) data indicating a shape of a time waveform of a waveform of vibration caused by the apex beat in a predetermined time range;
(2) data of a graph in which the frequency and power spectrum obtained by the frequency analysis means of the time waveform are plotted on the horizontal and vertical axes, and
(3) image data in a predetermined time range showing the degree of fluctuation in time, frequency, and power spectrum, which are the analysis results of the short-time Fourier transform of the time waveform;
(4) Data for a predetermined time range regarding the type and period of a waveform extracted by a correlogram;
Using
The waveform of the vibration generated by the apex pulsation of a healthy person is classified into five types;
It is preferable to estimate whether the subject's health condition at the time of measurement corresponds to any of the five waveform classification results.

The biological signal detection sensor includes:
Three-dimensional knitting and
A containing film that hermetically covers the periphery of the three-dimensional knitted fabric;
It is preferable that the microphone is provided on the outside of the container film, and the case covers the microphone. The case includes a disturbance intrusion suppression member that suppresses disturbances from entering the microphone.
The disturbance-mixing suppression member is preferably a gel.

The present invention also provides a computer program for processing biosignal data obtained by a biosignal detection sensor via a body surface and causing a computer to function as a biosignal analysis device, comprising:
A step of subjecting the biological signal data to frequency analysis;
The present invention provides a computer program that causes the computer to execute a procedure for identifying, from the frequency analysis results, a boundary frequency in the biosignal data between vibrations caused by apex beats and vibrations caused by heart sounds.

In the step of identifying the boundary frequency,
It is preferable to find a sudden change portion of the power spectrum that is a boundary between harmonic vibration and irregular vibration in the frequency analysis result, and to specify the boundary frequency based on this sudden change portion.
In the step of identifying the boundary frequency,
It is preferable to determine the sudden change part of the power spectrum by adding the result of frequency analysis of the heart sound data measured at the same time to the result of the frequency analysis.

In the step of identifying the boundary frequency,
It is preferable to display waveforms obtained by averaging the results of frequency analysis of the biosignal data and the heart sound data on a double logarithmic axis using a logarithmic difference method, to obtain a fluctuation change point from the waveform displayed on the double logarithmic axis, and to identify this fluctuation change point as a sudden change part in the power spectrum.

In the step of performing the frequency analysis, a short-time Fourier transform is applied,
In the step of identifying the boundary frequency, it is preferable to determine a sudden change portion of the power spectrum from the analysis result of the short-time Fourier transform.
In the step of performing the frequency analysis, the analysis result of the short-time Fourier transform is outputted as image data indicating the degree of fluctuation in time, frequency and power spectrum;
In the step of identifying the boundary frequency, it is preferable to find a sudden change portion of the power spectrum from the image data.
It is preferable to execute a procedure of estimating the health condition of the subject at the time of measurement by comparing the result with correlation data related to the boundary frequency stored in a correlation data storage unit.

the correlation data is correlation data between the boundary frequency and a heart rate,
In the step of estimating the state at the time of measurement, it is preferable to compare the heart rate of the subject at the time of measurement with correlation data between the boundary frequency and the heart rate to estimate the boundary frequency of the subject at the time of measurement.
the correlation data is correlation data between the boundary frequency and a fluctuation characteristic of heart rate variability,
In the step of estimating the state during measurement, it is preferable to compare the boundary frequency during measurement of the subject with correlation data between the boundary frequency and the fluctuation characteristics of heart rate fluctuations to estimate the fluctuation characteristics of the heart rate fluctuations of the subject during measurement.

Further, a step of filtering the biological signal data by setting the boundary frequency identified by the boundary frequency identification means as an upper limit value, and obtaining a waveform of vibration caused by the apex beat;
and classifying the waveform of the vibration caused by the apex beat of the heart.
In the procedure for estimating the condition at the time of measurement, it is preferable that the waveform of the vibration generated by the apex beat of the subject at the time of measurement is classified by executing a procedure for classifying the waveform, and the health condition of the subject at the time of measurement is estimated based on the waveform classification result.
the correlation data storage unit stores correlation data between the waveform classification result and a health condition,
In the step of estimating the condition at the time of measurement, it is preferable to classify the waveform of the vibration caused by the apex beat of the subject at the time of measurement by executing a step of classifying the waveform of the vibration caused by the apex beat of the subject, and to compare the waveform classification result at the time of measurement with correlation data between the waveform classification result and health condition to estimate the health condition of the subject at the time of measurement.

generating an estimation model by machine learning that estimates the health condition from the waveform information, using waveform information that is output by executing the procedure of extracting the apex beat waveform and is to be classified in the procedure of classifying the waveform and the health condition as teacher data;
It is preferable that the procedure for estimating the condition at the time of measurement uses waveform information of the vibration generated by the apex beat of the subject at the time of measurement as input, and uses the estimation model to output a value of the health condition at the time of measurement from the acquired waveform information at the time of measurement.

The procedure for classifying the waveform includes the following procedures (A) and (B):
(A) classifying the waveform using a mathematical approach using Fourier series expansion;
(B) classifying the waveform by combining one or more pieces of data obtained by the following engineering approaches (1) to (4):
(1) data indicating a shape of a time waveform of a waveform of vibration caused by the apex beat in a predetermined time range;
(2) data of a graph in which the frequency and power spectrum obtained by the frequency analysis means of the time waveform are plotted on the horizontal and vertical axes, and
(3) image data in a predetermined time range showing the degree of fluctuation in time, frequency, and power spectrum, which are the analysis results of the short-time Fourier transform of the time waveform;
(4) Data for a predetermined time range regarding the type and period of a waveform extracted by a correlogram;
Using
The waveform of the vibration generated by the apex pulsation of a healthy person is classified into five types;
It is preferable to estimate whether the subject's health condition at the time of measurement corresponds to any of the five waveform classification results.

The present invention also provides a recording medium on which the above computer program is recorded.

According to the present invention, it is possible to obtain the boundary frequency between the apex beat and heart sounds. This boundary frequency, which is a new index related to the apex beat, correlates with the fluctuation characteristics of heart rate variability, and it is possible to estimate health conditions (presence or absence of diseases such as heart disease, identification of diseases, physical condition, etc.) using the boundary frequency. Furthermore, according to the present invention, this boundary frequency can be expressed as a function with the heart rate, so that correlation data based on this can be used to know the boundary frequency in real time.

In addition, in the present invention, the waveform of the vibration generated by the apical pulsation obtained by using the above-mentioned boundary frequency is classified, and each waveform is associated with the person's health condition (presence or absence of diseases such as heart disease, identification of diseases, physical condition, etc.), thereby making it possible to estimate the health condition. This makes it possible to automatically perform evaluation of the apical cardiogram waveform, which was previously performed by a specialist such as a doctor. In particular, by using the waveform information of the apical pulsation and the health condition as training data and configuring an estimation model that estimates the health condition from the waveform information to be generated by machine learning, the accuracy of estimating the health condition can be improved.

FIG. 1(a) is an external perspective view of a biosignal detection sensor (4SR) according to a first embodiment of the present invention, FIG. 1(b) is an external perspective view showing an air pack and a gel pack separated, and FIG. 1(c) is a cross-sectional view. FIG. 2 is a diagram showing the structure of the biosignal detection sensor (4SR) according to the first embodiment of the present invention in comparison with a conventional biosignal detection sensor (3SR). Figure 3(a) is a diagram showing the mechanical replica sound (PCG replica input) of a phonocardiogram (PCG) output from a speaker and input to the embodiment of the biosignal detection sensor (4SR) and the conventional biosignal detection sensor (3SR), and the acoustic vibration data transmitted by the three-dimensional knitted fabric; Figure 3(b) is a diagram showing the resonant waveform of the PCG replica input and the frequency components constituting the modulated PCG replica input; Figure 3(c) is a Lissajous figure showing the spring characteristics and damping characteristics of the embodiment of the biosignal detection sensor (4SR) and the conventional biosignal detection sensor (3SR); Figure 3(d) is a Bode plot; and Figure 3(e) is a diagram showing the spring constant of a structure (sealed air pack) in which the three-dimensional knitted fabric in the embodiment of the biosignal detection sensor (4SR) is sealed within a containing film. Figure 4 shows the amplification effect when a PCG waveform is input, where (a) shows the output waveforms of the embodiment's biosignal detection sensor (4SR) and a conventional biosignal detection sensor (3SR), (b) shows the frequency analysis results for each, and (c) shows the gain of 4SR/3SR. Figure 5 shows the amplification effect of APW (Acoustic Pulse Wave), where (a) shows the output waveforms of the embodiment's biosignal detection sensor (4SR) and a conventional biosignal detection sensor (3SR), (b) shows the frequency analysis results for each, and (c) shows the gain of 4SR/3SR. FIG. 6 is a diagram for explaining a comparative experiment between the biosignal detection sensor (4SR) of the first embodiment and a structure including only an air pack without a gel pack. Figures 7(a)-(c) show the experimental results for 4SR, and Figures 7(d)-(f) show the experimental results for a structure with only air pack 1A (4SR without gel pack). Figure 7(g) is an enlarged view of (d), and Figure 7(h) is an enlarged view of (e). Figure 7(i) shows a Bode plot, Figure 7(j) shows the load-deflection characteristics of 4SR and a structure with only air pack 1A, and Figure 7(k) shows the amplitude ratio between the two (4SR/4SR without gel pack). 8(a) to (d) show the results of 3-second time-series waveform measurements of F-APW, R-APW, L-APW and PCG of a subject with an average heart rate of 58/min. 9(a) to (d) show the results of 3-second time-series waveform measurements of F-APW, R-APW, L-APW and PCG of a subject with an average heart rate of 71/min. 10(a) to (d) show the results of 3-second time-series waveform measurements of F-APW, R-APW, L-APW and PCG of a subject with an average heart rate of 79/min. 11(a) to (d) show the results of 3-second time-series waveform measurements of F-APW, R-APW, L-APW and PCG of a subject with an average heart rate of 93/min. FIG. 12 is a diagram showing a schematic structure of the biosignal analyzer according to the first embodiment of the present invention. FIG. 13 is a flowchart for explaining the process of extracting the boundary frequency. 14(a) to (d) show the frequency analysis results of CAB (Cardiac Apex Beat) and CAS (Cardiac Acoustic Sound) for subjects with average heart rates of 58/min, 71/min, 79/min, and 93/min. FIG. 15 shows the processed waveforms of Front CAB(0.5-BF), Front CAB(5-BF), Rear CAB(0.5-BF), Lumbar CAB(0.5-BF), PCG, Front CAS(BF-50), Rear CAS(BF-50), and Lumbar CAS(BF-50) obtained from a subject with an average heart rate of 58/min. FIG. 16 shows the processed waveforms of Front CAB(0.5-BF), Front CAB(5-BF), Rear CAB(0.5-BF), Lumbar CAB(0.5-BF), PCG, Front CAS(BF-50), Rear CAS(BF-50), and Lumbar CAS(BF-50) obtained from a subject with an average heart rate of 71/min. FIG. 17 shows the processed waveforms of Front CAB(0.5-BF), Front CAB(5-BF), Rear CAB(0.5-BF), Lumbar CAB(0.5-BF), PCG, Front CAS(BF-50), Rear CAS(BF-50), and Lumbar CAS(BF-50) obtained from a subject with an average heart rate of 79/min. FIG. 18 shows the processed waveforms of Front CAB(0.5-BF), Front CAB(5-BF), Rear CAB(0.5-BF), Lumbar CAB(0.5-BF), PCG, Front CAS(BF-50), Rear CAS(BF-50), and Lumbar CAS(BF-50) obtained from a subject with an average heart rate of 93/min. Figures 19(a) to (d) show the ECG, F-APW, and PCG for four subjects with average heart rates of 58/min, 71/min, 79/min, and 93/min. 20(a) to (d) show ECG, Front CAB, and Front CAS for four subjects with average heart rates of 58/min, 71/min, 79/min, and 93/min. Figures 21(a) to (d) show the power spectral densities (PSDs) of Front CAB and Front CAS for four subjects with average heart rates of 58/min, 71/min, 79/min, and 93/min. FIG. 22 shows the results of STFT analysis of a subject with an average heart rate of 93/min over a 10-second measurement period, where (a) shows the results of F-APW analysis and (b) shows the results of PCG analysis. FIG. 23(a) shows the results of STFT analysis of F-APW×PCG ⁻¹ of a subject with an average heart rate of 93/min over a measurement time of 10 seconds, and FIG. 23(b) is an enlarged view of FIG. 23(a). Figure 24(a) shows the results of STFT analysis of F-APW×PCG of a subject with an average heart rate of 93/min over a 10-second measurement time, and Figure 24(b) shows the power spectra of F-APW×PCG ^-1 , F-APW, PCG, and F-APW×PCG. Figure 25(a) shows the power spectra of F-APW×PCG ^-1 , F-APW, PCG, and F-APW×PCG of a subject with an average heart rate of 58/min, and Figure 25(b) shows the results of STFT analysis of F-APW×PCG ^-1 . Figure 26(a) shows the power spectra of F-APW×PCG ^-1 , F-APW, PCG, and F-APW×PCG of a subject with an average heart rate of 71/min, and Figure 26(b) shows the results of STFT analysis of F-APW×PCG ^-1 . Figure 27(a) shows the power spectra of F-APW×PCG ^-1 , F-APW, PCG, and F-APW×PCG of a subject with an average heart rate of 79/min, and Figure 27(b) shows the results of STFT analysis of F-APW×PCG ^-1 . FIG. 28(a) is a BF-HR correlation diagram for 50 subjects, and FIG. 28(b) is a correlation diagram between the harmonic disappearance order and HR. Figure 29(a) is a graph showing the correlation between the fluctuation characteristics of heart rate variability and heart rate, Figure 29(b) is a graph showing the correlation between the fluctuation characteristics of heart rate variability and BF, Figure 29(c) is a graph showing the correlation between the heart rate and BF of a certain subject, Figure 29(d) is a graph showing the correlation between the heart rate and BF of another subject, and Figure 29(e) is a graph showing the correlation between the heart rate and BF of yet another subject. FIG. 30 is a schematic diagram showing the configuration of a biosignal analyzer according to the second embodiment. FIG. 31 is a schematic diagram for explaining another aspect of the biosignal analyzer according to the second embodiment. 32(a) to (e) are diagrams showing typical waveform analysis results of Front CAB (0.5-BF) and illustrating examples of pseudo-sine waves. 33(a) to (e) are diagrams illustrating typical waveform analysis results of Front CAB (0.5-BF) and are used to explain an example of a triangular wave. 34(a) to (e) are diagrams illustrating typical waveform analysis results of Front CAB (0.5-BF) and are used to explain an example of a cosine wave. 35(a) to (e) are diagrams illustrating typical waveform analysis results of Front CAB (0.5-BF) and are used to explain examples of Gaussian waves. 36(a) to (e) are diagrams showing typical waveform analysis results of the Front CAB (0.5-BF) and are used to explain an example of a pseudo full-wave rectified wave. FIG. 37 shows the results of waveform analysis using two types of STFT, power spectrum, and correlogram. 38(a) to (e) are diagrams showing the results of frequency analysis for each measurement site of five subjects. FIG. 39 is a diagram showing the amplitude ratio for each measurement site. FIG. 40 is a BF-HR correlation diagram when heart rate fluctuation occurs under respiratory inhibition at rest. FIG. 41(a) is a BF-HR correlation diagram in the supine position, FIG. 41(b) is a BF-HR correlation diagram in the left lateral position, and FIG. 41(c) is a BF-HR correlation diagram in the right lateral position.

Hereinafter, the present invention will be described in more detail based on the embodiments of the present invention shown in the drawings.
(First embodiment)
1(a) to 1(c), the configuration of the biosignal detection sensor 1 used in this embodiment will be described. The biosignal detection sensor 1 of this embodiment has a laminated structure of an air pack 1A and a gel pack 1B. The air pack 1A has a three-dimensional knitted fabric (3D net) 10 and a housing film 20 that hermetically houses the three-dimensional knitted fabric (3D net) 10. The gel pack 1B has a microphone 30 fixedly disposed within a case 40, and gel 50 is filled around the microphone 30.

The three-dimensional knitted fabric 10 is formed by connecting a pair of ground knitted fabrics arranged at a distance from each other with a connecting thread. Each ground knitted fabric can be formed, for example, from a yarn made of twisted fibers into a flat knitted fabric structure (fine stitches) that is continuous in both the wale direction and the course direction, or into a knitted fabric structure having a honeycomb-shaped (hexagonal) mesh. The connecting thread imparts a predetermined rigidity to the three-dimensional knitted fabric so that one ground knitted fabric and the other ground knitted fabric maintain a predetermined distance. Therefore, by applying tension in the surface direction, it is possible to string-vibrate the yarns of the opposing ground knitted fabrics that make up the three-dimensional knitted fabric, or the connecting threads that connect the opposing ground knitted fabrics. As a result, string vibrations are generated by the sound and vibrations of the heart and blood vessels, which are biological signals, and are propagated in the surface direction of the three-dimensional knitted fabric.

Various materials can be used as the material for the yarn or connecting yarn forming the ground knit of the three-dimensional knitted fabric, including synthetic fibers such as polypropylene, polyester, polyamide, polyacrylonitrile, and rayon, regenerated fibers, and natural fibers such as wool, silk, and cotton. The above materials may be used alone or in combination. Preferably, polyester-based fibers such as polyethylene terephthalate (PET) and polybutylene terephthalate (PBT), polyamide-based fibers such as nylon 6 and nylon 66, polyolefin-based fibers such as polyethylene and polypropylene, or a combination of two or more of these fibers. The shape of the ground yarn or connecting yarn is not limited, and may be any of round cross-section yarn, irregular cross-section yarn, hollow yarn, etc. Furthermore, carbon yarn, metal yarn, etc. can also be used.

Examples of the three-dimensional knitted fabric that can be used include the following:
(a) Product number: 49013D (manufactured by Suminoe Textile Co., Ltd.), thickness 10 mm
Material:
Front ground knit: 2 strands of 450 dtex/108f polyethylene terephthalate false twist yarn Back ground knit: 2 strands of 450 dtex/108f polyethylene terephthalate false twist yarn Connecting yarn: 350 dtex/1f polytrimethylene terephthalate monofilament (b) Product number: AKE70042 (manufactured by Asahi Kasei Corporation), thickness 7 mm
(c) Product number: T28019C8G (manufactured by Asahi Kasei Corporation), thickness 7 mm

The three-dimensional knitted fabric 10 is covered with a storage film 20. In this embodiment, the storage film 20 is made of two synthetic resin films 21, 22, which are arranged to cover the front and back surfaces of the three-dimensional knitted fabric 10, and the peripheral edges of both films are fixed by welding or the like. As a result, the three-dimensional knitted fabric 10 is hermetically stored within the storage film 20. When fixing the peripheral edges of the films 21, 22, it is preferable to fix the three-dimensional knitted fabric 10 so that the films 21, 22 slightly press the three-dimensional knitted fabric 10 in the thickness direction. The tension of the three-dimensional knitted fabric 10 increases, making it easier for string vibration to occur in the yarn that constitutes the three-dimensional knitted fabric 10.

A case 40 is attached to the outside of the containing film 20, and a microphone 30 is disposed inside the case 40. Inside the case 40, the microphone 30 is filled with gel 50 as a disturbance suppression member. The case 40 is made of synthetic resin and has the function of preventing the acoustic vibrations propagated to the microphone 30 from diffusing to the outside, and the gel 50 suppresses the capture of external vibrations by the microphone 30. A cord 30a that carries detected acoustic vibration data is connected to the microphone 30.

The biosignal detection sensor 1 is used by contacting various parts of a person, such as the back, chest, or waist. Vibrations on the body surface are transmitted to the containing film 20 and the three-dimensional knitted fabric 10 and picked up by the microphone 30, but it is not limited to being attached directly to the skin surface, and can also be attached to the surface of clothing.

Here, the detection performance of the new biosignal detection sensor 1 employed in this embodiment will be explained in comparison with the conventional biosignal detection sensor disclosed by the present inventors in Patent Document 2 and the like. In the following, the biosignal detection sensor 1 of this embodiment will be referred to as 4SR (Sound Sensing System using Stochastic Resonance), and the conventional biosignal detection sensor 1000 will be referred to as 3SR (Sound Sensing System using Resonance).

(Evaluation of 4SR sensing performance in the heart sound frequency band)
Figure 2 shows the outline and component configuration of the experimental equipment for comparing the sensing performance of 4SR (biological signal detection sensor 1) and 3SR (conventional biological signal detection sensor 1000). The three photographs on the left and right of the experimental equipment in the center show the acceleration sensor (a-1) and the components (a-2 to a-6) of 3SR and 4SR, respectively, and overview photographs of 3SR (a-7) and 4SR (a-8) are shown at the bottom of the experimental equipment. In addition, photograph (a-4) shows the cross section of the 10 mm thick three-dimensional knitted fabric (hereinafter sometimes called 3D net) used in 3SR and 4SR. Photographs (Section: aa, Section: bb) are cut cross sections of the 3D net for 3SR and 4SR.

First, the basic principle of the 4SR's amplitude amplification will be explained using a cross-sectional photograph (a-4) of the 3D net. The top and bottom surfaces of the 3D net used in this experiment are fabric layers made of multifilament yarn, and the pile in the middle made of monofilament yarn is woven into the top and bottom fabric layers and is frictionally bonded to the multifilament yarn. When the subject's weight is applied to the top surface of the 3D net, the pile woven into an X-shape of the 3D net built into the air pack (hereafter referred to as the X-shaped pile) bends, and the pile rubs against each other due to body movement, generating sound and vibration. This sound and vibration causes the stochastic resonance phenomenon. The X-shaped pile generates tension as it bends, forming a natural oscillator. The natural frequency of the X-shaped pile under tension is around 20 Hz. Therefore, it is thought that the 4SR, which is a sealed air pack, increases vibrations in the 0.5 to 80 Hz range through two resonance mechanisms: stochastic resonance and string vibration.

Next, the amplification ability for heart sounds will be examined. As shown in Figure 3 (a), the heart sound used in the study was the first sound on the PCG with a split interval of 0.03 seconds. This heart sound was measured from a subject with a heart rate of 56 beats per minute, and after filtering, it was mechanically reproduced using a speaker. The speaker used was a FOSTEX P1000K without an enclosure. The P1000K is a cone-type full-range unit with a diameter of 10 cm. A 600g weight was placed on the diaphragm of the P1000K, making the weight of the diaphragm 1005g, and the lowest resonance frequency was set lower from 82Hz to 52Hz. This resulted in a playback frequency band of 52Hz to 16kHz. The lowest resonance frequency f0 was:

で求められる。s0はばねの強さを示し、m0は振動板の重さを示す。

where s0 is the strength of the spring and m0 is the weight of the diaphragm.

The mechanically reproduced sound (hereafter referred to as Reproduction PCG) output from the speaker passes through a 3D net placed on the top surface of the acceleration sensor and is then measured by the microphones of the 3SR and 4SR. The microphone 1010 of the 3SR is wrapped in an elastomer film 1030 and beads 1040 along with the 3D net 1020. Attenuation occurs when the air around the microphone flows due to the spring force of the built-in 3D net caused by the air flowing through the slit. The magnitude of attenuation is determined by the width of the slit around the cord 1010a attached to the microphone, as shown in photo (a-7). Since the microphone 1010 of the 3SR is placed inside the elastomer film 1030, the cord 1010a must be pulled out of the elastomer film 1030, and a gap is inevitably created at that location. On the other hand, the 4SR is configured such that the air pack 1A, in which the three-dimensional knit (3D net) 10 is housed in the housing film 20, is separated from the gel pack 1B in which the microphone 30 is located, and the cord 30a of the microphone 30 is pulled out from the gel pack 1B, so that the airtightness of the air pack 1A is not impaired by it. In other words, the 4SR is sealed all around the three-dimensional knit (3D net) 10 with an elastomer film (housing film 20), and when a load is applied to the 4SR, a uniform air pressure is applied to the elastomer film, which generates tension in the elastomer film and creates an environment in which acoustic vibrations can easily propagate (see FIG. 3(c)).

Figure 3(a) shows the PCG replica input and the acoustic vibration data transmitted by the 3D net as described above, but the acoustic vibration data transmitted by the 3D net consists of a waveform due to the resonance effect of the 3D net's natural oscillator, followed by a modulated PCG replica input. The area of the Lissajous figure in Figure 3(c) indicates the energy consumed during one cycle of vibration, and the damping ratio of the 3SR calculated from the damping capacity was 0.53 and that of the 4SR was 0.20.

(Evaluation of 4SR sensing performance in the apex beat frequency band)
In healthy subjects, the ACG shows a short-duration pulse observed only in the early systole (the systolic wave consists of a short-duration sharp positive wave followed by a dicrotic limb, and the A wave derived from the contraction of the left atrium is observed as a trace), which is described as tapping, and it is difficult to obtain a good record with the ACG. Therefore, we selected a subject in a seated position (heart rate: 62/min) who could palpate the apical pulse, which could be described as tapping, and measured the F-APW (biological information measured from the front of the chest: Front Acoustic Pulse Wave) using 3SR and 4SR, recorded it on a data logger, and performed the following analysis after filtering.

The input waveform recorded in the data logger is frequency analyzed, and the ratio of the power spectrum of 4SR and 3SR to the frequency (PSD _4SR /PSD _3SR ) is taken as the gain, and the sensing performance is evaluated by the gain. The 4SR is placed on the front of the chest 6 cm to the left of the 5th intercostal space and the midline of the chest, and the 3SR is placed on the front of the chest 10 cm to the left (equivalent to the V4 lead of the electrocardiogram). The microphones for the PCG and 4SR sensing systems are placed in areas where alveolar respiratory sounds are likely to be mixed in, so F-APW is measured for 10 seconds with breathing stopped. The procedure for the measurement experiment was to continue breathing at rest for 5 minutes, and then stop breathing after 2 seconds of inhalation. The subjects were healthy men in their 40s with no underlying cardiovascular diseases.

(APW measurement)
Next, the APW was measured. All measurements were performed in a seated position. The 4SR for capturing the F-APW, which is thought to include the apical pulsation, was placed on the front of the chest, 10 cm to the left of the chest midline, at the left 5th intercostal space and on the left midclavicular line (near V4). The 4SR for the R-APW (biological information measured from the posterior chest: Rear Acoustic Pulse Wave) was placed on the posterior chest, 6 cm to the left of the back midline, at the same height as the 4SR for the F-APW on the front of the chest. The 4SR for capturing the lumbar L-APW (biological information measured from the lumbar: Lumber Acoustic Pulse Wave) was placed in the midline between the 3rd and 4th lumbar vertebrae, directly behind the navel, and the microphone for PCG was placed at the apex of the heart. In addition, an ECG (Electrocardiogram) was obtained using lead II.

The placement of the microphone sensor at the apex of the heart on the anterior chest is the optimal location for recording the first, third, and fourth hearts, but it is difficult to obtain a good recording of the second heart sound. The second heart sound is said to be generated when blood in the arteries tries to flow back into the ventricle at the end of systole, causing excessive stretching of the semilunar valves and then recoil on the arterial wall (cardiohemic system), and is attenuated at the apex of the heart.

The microphones for the PCG and 4SR sensing systems are installed in locations where alveolar breathing sounds are likely to be mixed in, so calculations are performed using the results of measurements taken for 10 seconds while breathing is stopped. The APW measurement experiment at rest is performed under four conditions: a) natural breathing continued for 5 minutes, b) breathing stopped for 30 seconds after 2 seconds of inhalation, c) natural breathing for 30 seconds, and d) breathing stopped for 30 seconds after 2 seconds of inhalation.
The subjects were 70 healthy men and women with different characteristics, from a group with relatively thin chest walls and a group with relatively thick chest walls.
The 50 subjects were healthy individuals with relatively thin chest walls, ranging in age from 20 to 60 years old, and 78% were male. A thin chest wall makes it easier to observe the sine wave or triangular wave that is characteristic of the apical pulsation. Using these subjects, we defined the five waveform types described below and determined the boundary frequency (BF) between the vibration caused by the apical pulsation and the vibration caused by the heart sound.
Using 20 muscular subjects in their 20s with relatively thick chest walls, we evaluated the suitability and scalability of the analysis method based on the five waveform types previously determined.
The clinical profile of the 50 healthy subjects with relatively thin chest walls was as follows: mean age 38.2 (±10.3) years, 39 men, 11 women, mean BMI 23 (±3.4), of which 4 had hypertension and 10 were smokers.
On the other hand, the clinical profile of the 20 patients with relatively thick chest walls was as follows: mean age 23.4 (±3.6) years, 20 men, mean BMI 20.8 (±2.4), one patient with hypertension, and five smokers.
The data to be analyzed were 10 seconds from the first 30 seconds after breathing was stopped during which RR fluctuation was within 15%, and the heart rate was calculated as the average value of the 10 seconds of data.

(4SR sensing system performance evaluation results and APW measurement results)
In Fig. 3(a), the waveform shown by the thin black line is the Reproduction PCG, and the waveform shown by the thick gray line is the Acceleration PCG measured by the acceleration sensor. The Reproduction PCG is composed of the I sound (A and B waves), the waveform around 30 Hz (C wave), and the II sound (D, E, F waves), while the Acceleration PCG is composed of the Reproduction PCG resonance waveform (a and b waves, and d and e waves) and the subsequent waveforms with changed amplitude and frequency (c and f waves). As shown in the frequency analysis results in Fig. 3(b), the A and B waves and the D and E waves in the frequency band above 60 Hz have increased in amplitude and become a and b waves, and d and e waves, and the C and F waves in the vicinity of 30 Hz have become c and f waves whose amplitude and frequency are modulated by heterodyning. As a result, the power spectrum of the Acceleration PCG is smaller than that of the Reproduction PCG in the vicinity of 30 Hz.

Next, we will consider the damping performance of the mechanical vibration system that constitutes the sensing system. Figure 3 (c) is a Lissajous figure showing the damping characteristics of the 3SR and 4SR sensors. The Lissajous figure was drawn using a servo pulsar with a pressure plate of 110 x 110 mm, an amplitude of ±1.0 mm, and an input of a vibration frequency of 1.34 Hz. The precompression force is 652 N. The area enclosed by the Lissajous figure is the energy consumed during one cycle of vibration, and is the damping capacity. If this is expressed as W, then W is:

となる。4SR と3SRの動的特性を比較すると、3SRは4SRとほぼ同等のばね特性に加えて、エアの流入流出による減衰の影響が顕著であることが分かる。

Comparing the dynamic characteristics of the 4SR and 3SR, we can see that the 3SR has roughly the same spring characteristics as the 4SR, but also has a pronounced damping effect due to the inflow and outflow of air.

The amplitude magnification Z/Y used to evaluate vibration displacement is, where Z is the relative displacement and Y is the apex pulsation displacement.

で与えられる。ここに4SRと3SRの各センサの機械振動系の固有角振動数は、

であり、減衰比は、

である。

Here, the natural angular frequency of the mechanical vibration system of the 4SR and 3SR sensors is given by

and the damping ratio is

It is.

Figure 3 (d) shows the amplitude magnification curves of 4SR and 3SR calculated using equation (3). It compares the spring constants of each 4SR sensor.

の値を代入した。

The value of was substituted.

Figure 3 (e) shows the load-deflection characteristics obtained by measuring the sensor (4SR) shown in Figure 2 (a-8) with a digital force gauge (RZ-20) using a φ15 mm pressure plate. The spring constant was calculated assuming measurement of the apex pulsation. When measuring from the front of the chest, the pressing force of the sensor is small, so the spring constant when the deflection is 1 mm was used. When measuring from the back of the chest, the body weight is applied to the sensor, so the spring constant when the deflection is 4 to 5 mm, which is the design value of the 3SR, was used.

Because the natural frequency of the sensing system of the 4SR is set low, when measurements are taken in the range of Z/Y ≒ 1, the 4SR can measure frequencies above 0.45 Hz, while the 3SR can measure frequencies above 2.10 Hz. If the damping ratio is set to ζ = 0.7, the 4SR can measure frequencies above 0.99 Hz, while the 3SR can measure frequencies above 3.15 Hz. In this way, by changing the damping ratio it is possible to adjust the characteristics of the mechanical filter.

Cardiac auscultation is to grasp, interpret, and confirm the characteristics of heart sounds in the high frequency range of around 50 to 100 Hz, so it is important to record high frequency acoustic vibrations. On the other hand, the apical beat is a short-lasting beat that is only observed in the early systole and is in a lower frequency range than the heart sound, so if the emphasis is placed on measuring high frequencies, it becomes difficult to record a good apical cardiogram. As the required specifications for the sensing system for heart sounds and apical beats are different, the quantitative evaluation of the sensing systems was evaluated separately. Figure 4 (a) compares the acoustic vibration information captured by the 3SR and 4SR sensing systems (hereinafter referred to as 3SR data and 4SR data) using the Acceleration PCG as input. Figure 4 (b) shows the frequency analysis results. The 3SR data and 4SR data are acoustic vibration information affected by stochastic resonance and resonance due to the natural oscillator, and when comparing the two, increases and decreases can be seen in the power spectrum in the same frequency band. Figure 4(c) shows the gain of 4SR/3SR (PSD _4SR /PSD _3SR ). The effect of stochastic resonance near 10 Hz, the effect of resonance of the natural oscillator near 20 Hz, and the difference in the mechanical vibration characteristics of the sensor from 20 to 80 Hz are manifested as changes in gain (maximum 11.6 dB and minimum 8.0 dB), and the maximum amplitude difference of the waveform shown in Figure 4(a) is 4.3 times. The sudden drop in gain above 80 Hz is thought to be due to the damping effect of the gel attached to the 4SR sensing system.

Figure 5(a) shows the time series waveforms of 3SR data and 4SR data when a person was used as the sound source instead of a speaker, and Figure 5(b) shows the frequency analysis results. Figure 5(c) shows the gain of 4SR/3SR (PSD _4SR /PSD _3SR ). The attenuation inherent in the 3SR sensing system functions above 1.34Hz, and while the gain near 1Hz is 13.4dB, the gain near 1.34 to 10Hz drops to 4.9dB. The effect of attenuation is noticeable when viewed from the waveform, and as shown in Figure 5(a), the maximum amplitude of the systolic wave is 6.8 times larger.

Although the air damping of the 3SR only had a small effect on the irregular vibrations related to heart sounds, it did affect the power spectrum of the apical pulsation wave below 13 Hz, reducing its value, as shown in Figure 5(b). The damping effect shown in Figure 3(c) was particularly noticeable in the fundamental and second harmonics. This is consistent with the simulation results in Figure 3(d), and it was found that the apical pulsation wave cannot be recorded with the 3SR.

Considering the vibration transmission mechanism, tension is generated in the elastomer film (the housing film with reference number 20 in 4SR, and the elastomer film with reference number 1030 in 3SR) itself, so that in 4SR, low-frequency vibrations around 1 Hz are transmitted to the internal 3D net through tension fluctuations. As a result, vibrations of 13 Hz or more were measured in 3SR and 4SR as almost the same power spectrum between 13 and 30 Hz as shown in Figure 5 (b) due to the effect of the natural oscillator occurring around 20 Hz. It is considered that the vibration components are transmitted through the piles of the 3D net. In addition, in Figure 5 (b), it was found that the minimum frequency of the heart sound and the frequency band where the harmonic components of the apex pulsation wave are mixed into the heart sound are around 10 Hz. Therefore, it was found that 4SR has higher sensitivity than 3SR over the entire frequency range considered.
Previously, the analysis target frequency for the 3SR was set to 10-30Hz, but based on the above experimental results, the frequency of the F-APW that the 4SR measures is set to 0.5-80Hz, taking into consideration that the microphone's performance guarantee range is 0.1Hz or higher, and for heart sounds, the analysis target frequency is set to 25-45Hz or 40-80Hz, rather than 100Hz or higher, because most sounds are concentrated in the low audible range. In addition, the sampling frequency was set to 200Hz for the 3SR, but to 1000Hz for the 4SR.

Here, a comparative experiment was conducted between the biosignal detection sensor 1 (4SR) of this embodiment and a structure (4SR without gel pack) in which the gel pack 1B was removed from the 4SR and only the air pack 1A was used to measure acoustic vibration. Figure 6 shows the state of the experiment, in which the biosignal detection sensor 1 (4SR) was placed on the right side of the figure on a urethane base, and the structure (4SR without gel pack) with only the air pack 1A used was placed on the left side. The other experimental conditions were the same as those of the comparative experiment between 3SR and 4SR described above. The results are shown in Figure 7.
7(a)-(c) show the experimental results of 4SR. 7(d)-(f) show the experimental results of the structure with only the air pack 1A (4SR without gel pack). (a) and (d) show the time series waveforms of the acoustic vibration information captured when a person is the sound source, (b) and (e) are time series waveforms in which the center of the amplitude of the waveforms in (a) and (d) is the zero point, and (c) and (f) show the time series waveforms of the acoustic vibration information captured by inputting the Acceleration PCG. FIG. 7(g) is an enlarged view of (d), and FIG. 7(h) is an enlarged view of (e).
As is clear from these figures, 4SR has a higher detection sensitivity for acoustic vibration information than a structure consisting of only air pack 1A (4SR without gel pack). In particular, when a person was used as the sound source, 4SR caused an amplitude change of about ±0.2 [V], and was able to clearly capture acoustic vibration information, whereas it was very difficult to capture the amplitude change with a structure consisting of only air pack 1A. Also, as shown in Figure 7 (j), 4SR had a smaller hysteresis loss in the load-deflection characteristics compared to a structure consisting of only air pack 1A.
The microphone 30 is placed on a thin, disk-shaped support member (member 30b in FIG. 6) on the bottom surface of a synthetic resin case 40, but the case 40 is about 0.1 to 0.2 mm thick and has lower rigidity than the disk-shaped support member 30b. Therefore, the case 40 easily vibrates in response to acoustic vibrations, resonates with the acoustic vibrations, and plays a role similar to the diaphragm and enclosure of a speaker, thereby increasing the detection sensitivity of the microphone 30 and increasing the amplitude. FIG. 7(k) shows the amplitude ratio of the two (4SR/4SR without gel pack). From this figure, it can be seen that 4SR not only has high detection sensitivity for extremely low-frequency acoustic vibrations such as apical pulsation, but also has high detection sensitivity in the range of heart sounds (25 to 80 Hz).
In addition, the time waveforms of the microacoustic vibration (MAV) output from the trunk measured by a microphone attached directly to the chest and the F-APW (response of the mechanical vibration system of the above-mentioned components that make up the 4SR, and acoustic vibration information affected by resonance due to the natural oscillator and stochastic resonance) measured by the 4SR were frequency-analyzed, and the amplitude magnification was calculated to obtain the gain of the F-APW/MAV. The gain of the F-APW/MAV was 42 dB in the frequency band of the fundamental harmonic of the apex beat, slightly more than 30 dB in the frequency band of the harmonics, and 20 dB in the minimum frequency band of the heart sound near 20 to 30 Hz, which is the frequency band of the natural oscillator. The response of the mechanical vibration system and the effect of stochastic resonance were expressed as a maximum of more than 40 dB and a minimum of 10 dB, with an amplitude difference of just under 200 times. The waveform of the F-APW was observed in the 1 to 1.5 Hz band, and included information on the heart sound and the apex beat. The time lag between the MAV and the F-APW was 0.12 seconds.

Figures 8 to 11 show the results of 3-second time-series waveform measurements of F-APW, R-APW, L-APW and PCG for four subjects with average heart rates of 58, 71, 79 and 93/min. These three types of APW contain information on heart sounds and apex beats. Waveforms in the 1-1.5 Hz band were observed for F-APW, while waveforms in the 5-7 Hz band were observed for R-APW and L-APW.

Biological Signal Analysis Apparatus Next, a biological signal analysis apparatus 100 having a computer function in which a computer program for processing data obtained from the biological signal detection sensor 1 of this embodiment is set will be described with reference to FIG.

The biosignal analyzer 100 processes the time series data of the biosignal acquired by the biosignal detection sensor 1 to obtain the waveform of the apex beat, and further estimates the biocondition, i.e., various health conditions of a person (presence or absence of diseases such as heart disease, identification of diseases, physical condition, etc.). The biosignal analyzer 100 is composed of a computer (including a personal computer and a microcomputer incorporated in a device), and receives the time series data of the biosignal transmitted from the microphone 30 of the biosignal detection sensor 1. It has a frequency analysis means 110 and a boundary frequency identification means 120 that perform predetermined processing using the received time series data.
The storage unit also has a correlation data storage unit 150 that stores correlation data between the boundary frequency between vibrations caused by apex beats and vibrations caused by heart sounds, which has been measured and constructed in advance, and the heart rate.

More specifically, the biosignal analysis device 100 has a computer program stored in a storage unit (including recording media such as a hard disk built into the computer (biological signal analysis device 100), as well as various removable recording media and recording media of other computers connected via communication means) that executes procedures that function as the frequency analysis means 110 and the boundary frequency identification means 120. The biosignal analysis device 100 can also be realized using an electronic circuit having one or more memory circuits incorporating computer programs that realize the frequency analysis means 110 and the boundary frequency identification means 120.

The computer program can also be provided by storing it on a recording medium. The recording medium storing the computer program may be a non-transient recording medium. There is no particular limitation on non-transient recording media, but examples of such recording media include flexible disks, hard disks, CD-ROMs, MOs (magneto-optical disks), DVD-ROMs, and memory cards. It is also possible to transmit the computer program to a computer via a communication line and install it.

Since the biosignal detection sensor 1 can capture heart sounds as described above, when the obtained biosignal data is received, the frequency analysis means 110 performs frequency analysis on the biosignal data obtained by the biosignal detection sensor 1 via the body surface.

The boundary frequency identification means 120 determines the boundary frequency (BF) between the vibration caused by the apex beat and the vibration caused by the heart sound in the biosignal data from the frequency analysis result of the biosignal data obtained from the frequency analysis means 110. The boundary frequency identification means 120 includes a means for determining the sudden change part of the power spectrum that is the boundary between the harmonic vibration and the irregular vibration in the frequency analysis result obtained by the frequency analysis means 110, and identifying the boundary frequency based on this sudden change part.

According to this embodiment, by determining the boundary frequency, it is possible to obtain the waveform of the apex beat from the biosignal separately from the waveform of the heart sounds.

The biosignal analyzer 100 of this embodiment also uses the boundary frequency itself as a judgment index. That is, as will be apparent from the experiments described below, the boundary frequency is expressed as a quadratic function of the heart rate, and the boundary frequency can be known from the correlation data between the two based on the heart rate. Looking at the correlation between the boundary frequency and the heart rate, the heart rate of an efficient state is approximately 70 to 80/min. This is a state in which the exercise load on the heart is low, and in this case the boundary frequency is at a minimum. Therefore, the state of the heart at that time can be estimated from the boundary frequency at the time of measurement, which is obtained by measuring the heart rate and using the correlation data.

As shown by the imaginary lines in FIG. 12, the biosignal analyzer 100 of this embodiment preferably has a correlation data storage unit 150 in which correlation data between the boundary frequency and the heart rate is stored, and a measurement state estimation means 130 that determines the boundary frequency based on the heart rate of the subject in reference to the correlation data stored in the correlation data storage unit 150, and estimates the boundary frequency of the subject at the time of measurement.

When estimating the state by the measurement state estimation means 130, information on the heart rate is required, and the heart rate can be calculated by the heart rate calculation means 140 shown in phantom lines in FIG. 12. In this case, the biosignal detection sensor 1 and the heart rate calculation means 140 that processes the biosignal data to calculate the heart rate function together as a heart rate measurement unit. Note that the heart rate can be measured by electrocardiogram, phonocardiogram, photoplethysmography, sphygmomanometer, etc., and these measuring devices can also be used as the heart rate measurement unit. However, the biosignal detection sensor 1 of this embodiment is excellent in terms of ease of measurement, as it can measure without restraint just by contacting it with the surface of the human body.

The correlation data stored in the correlation data storage unit 150 consists of correlation data between the boundary frequency between vibrations caused by the apex beat and vibrations caused by heart sounds, which have been measured in advance, and the heart rate. This correlation data is constructed using a quadratic function that shows the correlation between the boundary frequency and the heart rate when the boundary frequency was determined. Therefore, the boundary frequency can be known by comparing the measured heart rate with the correlation data stored in the storage unit of the correlation data storage unit 150.

The boundary frequency stored in the correlation data storage unit 150 is determined in advance by analyzing a large amount of biological signal data. The boundary frequency is determined as the above-mentioned sudden change portion identified by the boundary frequency identification means 120.

The area of sudden change can be determined by first frequency analyzing the biosignal data obtained by the biosignal detection sensor 1 via the body surface, and then adding the frequency analysis results of the heart sound data measured at the same time to the frequency analysis results, and extracting and determining the boundary frequency between the vibrations caused by the apex beat and the vibrations caused by the heart sounds in the biosignal data.

Specifically, the frequency analysis results of the biosignal data obtained by the frequency analysis means 110 are averaged and the resulting waveform is displayed on a double logarithmic axis using frequency and power spectrum, while the frequency analysis results of the heart sound data are averaged and the resulting waveform is displayed on a double logarithmic axis using frequency and power spectrum. A fluctuation change point is obtained from the logarithmic difference between the two waveforms displayed on the double logarithmic axis, and this fluctuation change point is regarded as the effective vanishing point where the harmonic components of the apex beat become extremely small. The frequency of the vanishing point corresponds to the above-mentioned sudden change part, making it possible to obtain the boundary frequency.

Furthermore, the frequency analysis means 110 may employ a short-time Fourier transform (STFT) technique. In this case, the boundary frequency identification means 120 determines the sudden change part of the power spectrum from the analysis result of the short-time Fourier transform. The short-time Fourier transform preferably outputs the analysis result as image data indicating the time, frequency, and degree of fluctuation of the power spectrum. This allows the boundary frequency identification means 120 to determine the sudden change part of the power spectrum from the image data of the short-time Fourier transform. Specific means for identifying the sudden change part, i.e., for determining the boundary frequency, will be described later.

Here, the biosignal detection sensor 1 is attached to the back, chest, waist, etc. of a person's body and captures sounds and vibrations within the body, and the obtained biosignal is a collection of various biosounds and internal vibrations. On the other hand, the apical pulsation is a vibration hidden in the waveform of the heart sounds, and it is difficult to separate them. However, in this embodiment, the above-mentioned boundary frequency (BF) that distinguishes between the apical pulsation waveform and the waveform of the heart sounds is found, making it possible to estimate the biological state related to the apical pulsation based on the heart rate.

The method of performing frequency analysis using the frequency analysis means 110 and identifying the boundary frequency using the boundary frequency identification means 120 will be described in detail below. In this embodiment, there are two methods: a method of using a waveform displayed on both logarithmic axes as described above, determining a fluctuation change point based on the frequency at which harmonic vibrations in the waveform essentially disappear, and treating this fluctuation change point as a sudden change part in the power spectrum, and a method of determining a sudden change part in the power spectrum from the analysis results of a short-time Fourier transform.

[Boundary frequency (BF) extraction method]
(BF extraction method based on APW harmonic oscillation vanishing frequency)
As shown in Non-Patent Document 1, the heart sound is an irregular vibration, but the experimental results described below have revealed that the apex beat wave is mainly composed of harmonic vibration consisting of fundamental and higher harmonics. Therefore, we focused on frequency analysis of the APW waveform obtained from a 4SR sensing system attached to the front of the chest, performing arithmetic averaging, and finding the change point of the power spectrum of the harmonic harmonic vibration and the irregular vibration, that is, the sudden change part of the power spectrum. The power spectrum of the harmonic component of the CAB (Cardiac Apex Beat) of the apex beat wave becomes smaller as the frequency increases. On the other hand, the power spectrum of the irregular vibration system of the CAS (Cardiac Acoustic Sound) of the heart sound does not depend on the frequency, and the magnitude of the power spectrum changes. Here, the frequency at which the power spectrum of the harmonic component of the CAB becomes smaller and the fluctuation behavior of the CAS changes is called the boundary frequency, and hereinafter this is called BF. In addition, when heart rate or blood pressure fluctuations occur, the APW waveform changes instantaneously. One of the factors that causes the instantaneous fluctuations in the APW waveform is thought to be changes in the dominant frequency of the PCG.

Depending on the individual differences of the subjects, impedance also differs with regard to the transmission characteristics of the acoustic vibration of the trunk, and it may be difficult to find the BF. Therefore, the logarithmic difference method is applied to find the frequency at which the harmonic components of the apical pulsation wave disappear (the frequency at which the power spectrum of the harmonic components becomes very small and can be practically ignored). The arithmetic average processing of the APW is performed with a time window of 8.2 seconds and a Fourier transform with 90% overlap. The arithmetic average processed waveform is displayed on a double logarithmic axis. This is because the power spectrum waveform displayed on a double logarithmic axis can be used for the addition theorem. Note that the PCG waveform is used for the heart sounds. By applying the logarithmic difference method, the absolute value of the power spectrum of the PCG in the 0.5 to 20 Hz frequency band is increased, and the power spectrum waveform can be made in opposite phase. This reduces the power spectrum of the apical pulsation harmonic components and the irregular vibration of the heart sounds, making it easier to find the BF.

(BF extraction method using short-time Fourier transform)
The short-time Fourier transform (STFT) of the power spectrum is applied to F-APW, PCG, APW×PCG, and APW×PCG ^-1 to visualize the change points of the fluctuations. The formula for the short-time Fourier transform used in this analysis is:

となる。

It becomes.

x(t) is the measured signal data, and W(t-τ) is the window function. In order to set the frequency resolution to 0.5 Hz or less, the number of window function points is set to 2 ¹² = 4096. If the sampling frequency is 1000 Hz, 4096 points corresponds to 4.096 seconds, and 4.096 seconds is 0.244 Hz. This allows the frequency resolution to be set to 0.244 Hz, which is 0.5 Hz or less. The shift width of the window function is set to 50 points. t indicates the measurement time, and τ indicates the time series data based on the shift time of the window function. By multiplying the measured signal data by the window function, it is possible to create frequency information that changes over time. Note that frequency information is generated every 50/1000 = 0.05 seconds.

Next, using the BF determined based on the above two BF extraction methods, the apical beat waveform (Front CAB) and heart sounds (Front CAS) are plotted from the F-APW, and similarly, the Rear CAB, Rear CAS, Lumbar CAB, and Lumbar CAS are plotted from the R-APW and L-APW.In addition, since it was found in the preliminary experiment and preliminary study that HR is involved in BF, a correlation diagram between BF and HR is created using data from 50 subjects to examine the correlation between the two.
The BF extraction method based on the frequency at which harmonic components disappear is a method for determining the BF from the frequency, while the BF extraction method using the short-time Fourier transform is a method for determining the BF using the power spectrum as a threshold. Specifically, the former is a method for determining the frequency at which the power spectrum height changes from decreasing to increasing as the BF by changing the frequency, while the latter is a method for determining the BF by setting 70% of the average value of the power spectrum group near the BF as the threshold.
Figures 19 to 21 show a comparison of ECG, F-APW, PCG, Front CAB, and Front CAS for four subjects with HR (heart rates): 58/min, 71/min, 79/min, and 93/min.
In the case of a subject with a HR of 58/min, the systolic wave measured by the Front CAB showed a late systolic bulge in the ACG, but the ECG and PCG (Front CAS) made it difficult to consider dilated cardiomyopathy, and it is thought that the apical beat approached a sustained pattern due to the influence of gravity.
In the case of a subject with a HR of 71/min, the P wave was unclear and small and not visible on the ECG. The A wave (atrial upstroke wave) originating from the left atrial contraction confirmed by the ACG was confirmed on the front CAB, and the front CAS also had a low-frequency wave earlier than the I sound of the PCG, so this was not an ectopic rhythm.
In the subject case with a HR of 79/min, the waveform measured by Front CAB exhibited a sustained pattern, but the waveforms of the A waves and Rapid Filling Wave (RFW) and Front CAS suggested that this was not a case of left ventricular hypertrophy or leakage from the left ventricle during systole (mitral valve regurgitation, etc.).
In the case of a subject with a HR of 93/min, the ECG showed low voltage and a high heart rate at rest, but the Front CAB had low and high frequency components, and the First sound was neither accentuated nor attenuated, eliminating suspicion of underlying heart disease.

(Fluctuation characteristics of heart rate variability and correlation with heart rate and BF)
The slope of the power spectrum displayed on both logarithmic axes is called the fluctuation coefficient here, and the point where the fluctuation coefficient changes sharply is hereafter called the fluctuation change point. At the fluctuation change point, a loss of complexity or a breakpoint occurs. The fluctuation coefficient is expressed by the power spectrum P(f) and frequency (f), and is log P(f) and log(f) on both logarithmic axes. If there is a linear relationship between log P(f) and log(f), then

で表される。

It is expressed as:

To evaluate the fluctuation characteristics of heart rate variability obtained from ECG, experimental data measured using the APW measurement experimental protocol is used. The fluctuation characteristics are calculated using formula (5). The obtained value of n is a negative number (approximately -0.5 to -1.5), but according to Non-Patent Document 3, it is defined as the 1/f fluctuation of the cardiac cycle, so it is expressed as an absolute value.

To evaluate the fluctuation of heart rate variability, data from 30 seconds of apnea, natural breathing, and apnea is used, totaling 90 seconds. The heart rate here is the average value over the 90 seconds.

[BF extraction results for each extraction method]
(Results of BF extraction from harmonic loss frequencies of F-APW)
Figure 13 is a flowchart showing the procedure for extracting BF using F-APW×PCG ⁻¹ , and the following (1) to (10) correspond to the circled numbers in Figure 13. The calculation procedure and points of view are explained below.

(1) Calculate the F-APW from the 4SR data without filtering and perform frequency analysis. Hereinafter, this is referred to as F-APW.
(2) Perform averaging on the F-APW (time window of 8.2 seconds, 90% overlap).
(3) Extract the waveform from the PCG and perform frequency analysis. Hereinafter, this will be referred to as PCG.
(4) Perform the same averaging process as in (2) (time window 8.2 seconds, 90% overlap).
(5) Generate F-APW×PCG ^-1 using the power spectra F-APW and PCG obtained in (2) and (4), and find the breakpoint.
(6) From the breakpoints on the F-APW×PCG ^-1 waveform, the harmonic disappearance frequency of the fluctuating apical pulsation wave and the appearance frequency of the irregular vibration components of the heart sound that becomes white noise are identified.
(7) Draw a BF line from the harmonic vanishing point on F-APW×PCG ^-1 , and let the intersection of the BF line and F-APW be BF.
(8) 0.5 Hz to BF is the CAB band (hereinafter, CAB (0.5-BF)), and BF to 50 Hz is the CAS band (hereinafter, CAS (BF-50)).
(9) Apply a bandpass filter of 0.5 Hz to BF to the F-APW and verify that the harmonic components of the CAB are reduced near the BF.
(10) Apply a bandpass filter from BF to 50 Hz to the F-APW and confirm that the power spectrum changes regardless of frequency near the BF.

Figure 14 shows the results of frequency analysis of CAB and CAS, organized by heart rate (HR) (58-93/min) in a biaxial linear display. From this figure, it is possible to understand the disappearance of the harmonic components of CAB in BF, the appearance of irregular oscillations of CAS near BF, and the HR dependency of BF, and it is clear that the method of the present invention can separate CAB and CAS and visualize their characteristics.

Next, Figures 15 to 18 show CAB and CAS waveforms traced using the BF extraction method based on the harmonic loss frequency of the APW. The processed waveforms of Front CAB (0.5-BF), Front CAS (5-BF), Rear CAB (0.5-BF), Lumbar CAB (0.5-BF), and PCG, as well as Front CAS (BF-50), Rear CAS (BF-50), and Lumbar CAS (BF-50), extracted from the F-APW, R-APW, and L-APW obtained from subjects with average heart rates of 58, 71, 79, and 93/min, are shown. The tracing time is 1.5 seconds. The frequency band of the applied filter is shown in parentheses. The vertical and horizontal ranges of each CAB and CAS waveform are standardized for each measurement site. It can be seen that the Front CAB (0.5-BF) of four subjects shows that the apical pulsation wave, which is dominated by the fundamental harmonic, is depicted.

From the analysis of the results, it was found that the Front CAB (5-BF) was formed by a combination of multiple 5-10 Hz waveforms. On the other hand, the Rear CAB (0.5-BF) and Lumbar CAB (0.5-BF) were harmonic vibration waveforms dominated by waveforms with higher frequencies than the waveform shown in the Front CAB (0.5-BF), resulting in waveforms different from those of the Front CAB (0.5-BF). In addition, the heart sounds were irregular vibrations consisting of high-frequency waveforms that were a combination of multiple vibration components and multiple vibration waveforms.

(Results of BF extraction using short-time Fourier transform)
22 to 27 show the results of STFT analysis of power spectra by frequency analysis of F-APW and PCG for subjects with average heart rates of 93/min, 58/min, 71/min, and 79/min over a 10-second measurement period.
In the STFT diagram, the vertical axis represents frequency and the horizontal axis represents the STFT output time. The shade of color indicates the degree of fluctuation in the power spectrum, and the magnitude of the fluctuation coefficient expressed by equation (5) can be expressed by changing the range of the shade of color. The magnitude of the fluctuation in the power spectrum is shown by the shade of red, green, and blue (RGB). In areas where the fluctuation of the power spectrum is small, areas where the fluctuation coefficient does not change suddenly, or areas with small amplitude irregular vibration, red, green, and blue appear as planes, while in areas where the fluctuation of the power spectrum is large or areas where the fluctuation coefficient changes suddenly, red, green, and blue appear as lines.

The logarithmic difference (F-APW×PCG ⁻¹ ) is shown in Figures 23(a) and 23(b). Figure 24(b) shows the frequency analysis result, where the BF corresponding to the breakpoint is 15 Hz.

In Figure 24 (a), the logarithmic sum of APW and PCG (APW x PCG) was calculated to emphasize the harmonic components and irregular vibrations in the frequency band of 0.5 to 30 Hz. The STFT emphasizes both the apical pulsation wave and the irregular vibration of the heart sounds, with the BF as the boundary. As a result, the BF appears in the band of 14 to 17 Hz, which is dominated by blue and yellow.

From these two linear tracing lines, APW×PCG ^-1 and APW×PCG, the fluctuating behavior of the apical beat harmonics and the irregular vibration of heart sounds was detected, and the frequency at which the apical beat harmonics disappeared, the frequency at which the heart sounds appeared, and the BF were identified.

Next, points to consider when considering each of FIGS. 22 to 27 will be described.
22 to 24 show the results of STFT analysis for a subject whose average heart rate over a 10-second measurement period was 93/min.
Figure 22(a) shows that the APW has components of 9th harmonic with 1.6Hz as the fundamental harmonic. The BF in this figure is the breakpoint at the sudden change in the power spectrum, so the BF exists between 14 and 17Hz, as shown in blue.
In Figure 22(b), the PCG has vibration components with large changes in the power spectrum between 10 and 50 Hz. When viewed together with Figure 22(a), it can be seen that the PCG captures everything from the 5th harmonic to irregular vibrations.
Figure 23(a) shows APW×PCG ^-1 , and BF has a breakpoint at a sudden change in the band with little fluctuation in the power spectrum, so BF exists between 14 and 17 Hz, as shown by Red, Blue, and Red. Figure 23(b) shows an enlarged view of APW×PCG ^-1 , where a linear red line appears near 15 Hz, indicating a breakpoint.
In the APW×PCG in Fig. 24(a), the fundamental harmonic of 1.49Hz is the blue line, the second harmonic is the green line, and the harmonics from the third harmonic to 15Hz are the red line, and it can be seen that the power spectrum of the harmonics is larger than that of the fundamental harmonic. There is also a breakpoint in the blue range shown by Red-Blue-Red near 15Hz, indicating that irregular vibrations occur from 15 to 20Hz where Red begins.
24B shows the power spectra of APW×PCG ⁻¹ , APW, PCG, and APW×PCG. APW× ^PCG had a breakpoint at 15 Hz, and BF occurred at 15 Hz.

25 to 28 show the BFs obtained by the BF extraction method using APW×PCG ⁻¹ and STFT described above. The subjects' heart rates were 58, 71, and 79/min.
Figure 25(a) shows a case in which the breakpoint appears in a mountain-shaped power spectrum, which is the opposite of the case in Figure 22(a), while in Figure 25(b) the BF appears as a linear Red in the band sandwiched between Red-Green-Red-Blue.
FIG. 26(a) shows an example in which the breakpoint shown in FIG. 22(a) appears on the ridge of the mountain-shaped power spectrum, while in FIG. 26(b) the BF appears as a linear Red sandwiched between Red-Blue-Red-Blue.
Figures 27(a) and (b) show examples where the fluctuation of the power spectrum has changed. It is difficult to judge from the frequency analysis result in Figure 27(a), but it is possible to judge the BF by the STFT method in Figure 27(b). In Figure 27(b), the BF appears as a red line on the border between Red and Blue.

Figures 28(a) and (b) show the correlation diagrams (BF-HR correlation diagrams) between BF and heart rate of 50 subjects, which were obtained by the BF extraction method from harmonic disappearance frequency and confirmed by the BF extraction method using STFT. Figure 28(a) shows the frequency of BF on the vertical axis and HR on the horizontal axis. When the relationship between the two is approximated by a quadratic function, the relationship is expressed as y = ^0.0173x2 - 2.5847x + 107.6111, and the minimum value is near 10Hz when HR is 75/min. In this case, R2 = 0.8242. By using this diagram, BF can be known without having to measure PCG.
In addition, BF was not correlated with age, SBP, DBP, or BMI. Furthermore, in a study of BF in a group of 20 young controls, a strong correlation was observed between BF obtained by STFT and BF determined by the BF-HR curve (R2=0.85). In other words, since the microacoustic vibration (MAV) output from the trunk is collected using the biosignal detection sensor 1 (4SR) of this embodiment, by analyzing it with the biosignal analysis device 100, it is possible to identify BF and apical pulsation waves and classify waveforms (sine wave or triangular wave) even from subjects in the "group with thick chest walls," which was previously difficult to identify.

Figure 28(b) shows a correlation diagram between the harmonic disappearance order and HR. If the order is taken on the vertical axis and HR on the horizontal axis, and the relationship between the two is approximated by a quadratic function, the relationship is expressed as y = 0.0203x ² -3.2258x + 135.4587, and the minimum value is near 10Hz when HR is 79/min. In this case, R2 = 0.9479. Note that Figures 28(a) and (b) are composed of data measured for 10 seconds in a resting state with RR fluctuations of 15% or less.

Therefore, by storing the correlation diagrams of Figures 28(a) and (b) and their quadratic functions in advance in the correlation data storage unit 150, the measurement state estimation means 130 can quickly determine the boundary frequency at the time of measurement from the heart rate, and can quickly determine and output the state of the person's heart and health at the time of measurement according to the boundary frequency.

(The relationship between heart rate fluctuation characteristics, heart rate, and BF)
The correlation between the fluctuation characteristics of heart rate variability obtained from ECG in the APW measurement experiment and the heart rate was examined for 50 subjects. As shown in Figure 29 (a), the relationship between the two was approximated by a quadratic function, and when the HR was 72/min, the fluctuation was 1/f (|n| = 1). On the other hand, when the HR was around 50 or 90/min, the fluctuation was 1/f ² (|n| = 2).
Figure 29(b) shows the correlation between the fluctuation characteristics of heart rate variability and BF. The relationship between the two was approximated by a linear function. When BF was around 8 to 12 Hz, it showed 1/f fluctuation (|n| = 1), and when BF was around 20 Hz or more, it showed 1/ ^f2 fluctuation (|n| = 2).
Figures 29(c)-(e) show the experimental data for 90 seconds for the same subject, which was divided into 9 equal parts, and the BF fluctuations when heart rate fluctuations occurred during respiratory inhibition at rest were calculated and plotted on the regression curve of the BF-HR correlation diagram. For the three subjects, BF moved along the regression curve in accordance with heart rate fluctuations.

When the heart rate is 70-80/min in a resting state, the fluctuation of the heart rate fluctuation is 1/f. In light of this, when the BF is in the vicinity of 8-12 Hz, the fluctuation of the heart rate fluctuation becomes 1/f. When the BF shows a value in this vicinity, it can be estimated that the heart is moving efficiently. Therefore, by storing the correlation diagram in FIG. 29(b) in the correlation data storage unit 150, the measurement state estimation means 130 can quickly estimate the fluctuation characteristics of the heart rate fluctuation during measurement from the BF obtained by the boundary frequency identification means 120. If the fluctuation characteristics can be estimated as 1/f, it can be estimated that the heart is moving well, that is, that the patient is healthy in terms of the heart's movement. The further the BF value deviates from the vicinity of 8-12 Hz, which shows 1/f fluctuation, the greater the influence of aging, disease, etc. can be estimated.

Second Embodiment
Next, a second embodiment of the present invention will be described with reference to Fig. 30. In this embodiment, a biosignal analyzer 100 includes an apex beat waveform extracting means 160 and a waveform classifying means 170 in addition to the configuration of the first embodiment.

The apex beat waveform extraction means 160 filters the biosignal data by setting the boundary frequency (BF) identified by the boundary frequency identification means 120 as the upper limit value, and obtains the waveform of the vibration caused by the apex beat. In this embodiment, filtering is performed using a band pass filter of 0.5 Hz to BF as described above. Note that the apex beat waveform (CAB) and phonocardial waveform (CAS) are depicted in Figures 15 to 18 of the first embodiment. The apex beat waveform in this case is obtained using a band pass filter of 0.5 Hz to BF using a method similar to that used by the apex beat waveform extraction means 160 of this embodiment. The phonocardial waveform is obtained by applying a band pass filter of BF to 50 Hz.

The waveform classification means 170 classifies the waveform of vibrations caused by the apex pulsation (apex pulsation waveform) output by the apex pulsation waveform extraction means 160. In this embodiment, classification is performed by combining (A) a means for classifying the waveform by a mathematical approach using Fourier series expansion, and (B) a means for classifying the waveform from data obtained by an engineering approach. As will be described in detail later, in this embodiment, the apex pulsation waveforms of healthy subjects are classified into five types by combining the two approaches.
In addition, in the means for classifying waveforms from data obtained by an engineering approach, the waveforms are classified using at least one of the following four judgment factors. In order to improve the classification accuracy, it is preferable to classify using a combination of two or more judgment factors.

(1) Data showing the shape of the time waveform of the vibration generated by the apex beat in a predetermined time range (the diagrams (b) in each of Figures 32 to 36)
(2) Graph data in which the horizontal and vertical axes represent the frequency and power spectrum obtained by the frequency analysis means of the time waveform (Figures (c) in each of Figures 32 to 36)
(3) Image data in a predetermined time range showing the degree of fluctuation in time, frequency and power spectrum, which are the analysis results of the short-time Fourier transform of the time waveform (Figures (d) in each of Figures 32 to 36)
(4) Data on the type and period of waveforms extracted by correlograms over a specified time range (Figures (e) in each of Figures 32 to 36)

First, we sought to classify the time waveforms of healthy subjects into five waveforms using a mathematical approach that uses Fourier series expansion, and to prove this, we used an engineering approach to analyze the amplitude, frequency, and time phase, and obtained data corresponding to the five waveforms from the shape data of the time waveform of Front CAB (0.5-BF) (Figures 32 to 36 (b)), the graph data with the power spectrum on the horizontal and vertical axes (Figures 32 to 36 (c)), the graphic data of the STFT analysis results (Figures 32 to 36 (d)), and the data extracted by the correlogram (Figures 32 to 36 (e)). The method for evaluating these waveforms will be described in more detail later.

The information on the waveform classification results classified by the waveform classification means 170 is stored in the correlation data storage unit 150. Here, the information on the waveform classification results is stored in association with information on health status (presence or absence of illness such as heart disease, identification of illness, physical condition, etc.). This is the result of measuring the apical pulsation waveforms of as many people as possible and digitizing them in accordance with their health status at that time.

The measurement condition estimation means 130 classifies the waveform of vibrations caused by the apex beat of the subject at the time of measurement, which is obtained by the apex beat waveform extraction means 160, using the waveform classification means 170, and accesses the correlation data storage unit 150 to compare the waveform classification result at the time of measurement with information on the waveform classification result stored in advance. As a result, the health condition corresponding to the waveform classification result at the time of measurement is output.

As shown in FIG. 31, the configuration may include a model creation means 180 that uses the waveform information of the apical pulsation waveform output by the apical pulsation waveform extraction means 160 and to be classified by the waveform classification means 170, and the health condition, as teacher data, to generate an estimation model that estimates the health condition from the waveform information by machine learning. Furthermore, the accuracy of the estimation model is improved by performing machine learning while taking into account information on the classification results by the waveform classification means 170 (classification information).

In this case, the measurement state estimation means 130 uses the waveform information of the vibration generated by the apex pulsation of the subject at the time of measurement as input, and outputs the value of the health state at the time of measurement from the acquired waveform information at the time of measurement using the estimation model created by the model creation means 180. By adopting such a configuration incorporating machine learning, the accuracy of estimating the health state from the waveform information is improved. Note that the feature amount of the waveform information of the learning object is the amplitude, frequency, time phase, etc. constructed in each waveform shown in Figures 32 to 36, but the image data as the analysis result of STFT shown in each (c) of Figures 32 to 36 has a large amount of information, and it is easy to extract the feature amount of the learning object. Therefore, in order to grasp the waveform information, it is preferable to include the image data of STFT in the learning object. In addition, the upper and lower diagrams of each (c) of Figures 32 to 36 show the apex pulsation waveform of 0.5 Hz to BF divided into two, 1.95 Hz and 7.81 Hz, which further increases the feature amount of the learning object and contributes to improving the estimation accuracy.
Other configurations of the second embodiment, including the biological signal detection sensor 1, are similar to those of the first embodiment.

The classification and evaluation methods of the waveforms will be explained in more detail below.
[Evaluation method for time waveform of Front CAB]
(Front CAB time waveform analysis)
Next, the method for evaluating the Front CAB waveform will be described.
The apex cardiogram (ACG) is known as one of the physical diagnostic tools that non-invasively evaluates the dynamics of the cardiovascular system and serves as an indicator of pathology and treatment, but the position for examination is not necessarily supine, and the patient may be in a sitting or left lateral position. Therefore, if there is information on the normal apex cardiogram pattern by measurement position, covering the supine, left lateral, and sitting positions, it would be possible to examine the relationship with waveform analysis and time phase analysis of PCG and ECG.

Therefore, we perform frequency analysis on the F-APW waveform measured with a microphone placed on the chest wall, extracting the frequency band from 0.5 Hz to BF, i.e., Front CAB (0.5-BF), and compare it with the clinically known apical beat wave. The focus of the waveform analysis is on the tapping of the E wave, which is a short-lasting beat that can only be seen by visual inspection or palpation during early systole, and the sustained beat, which is a long-lasting, powerful beat. Here, we perform waveform analysis using the shape of the Front CAB (0.5-BF) waveform and the time phase at which the ACG peak appears.

The amplitude of each time waveform in the Front CAB (0.5-BF) is evaluated during the systolic time phase. The A wave and traces of the A wave resulting from left atrial contraction are confirmed in the Front CAB (0.5-BF). The bottom value after the A wave is taken as the C point of the ACG, and the total amplitude up to the peak value corresponding to the E wave of the ACG is measured and taken as the amplitude of the systolic wave. The total amplitude from the C point of the ACG to the E wave is also calculated for the time waveforms of the Front CAB (5-BF), Rear CAB (0.5-BF), and Lumbar CAB (0.5-BF). When comparing the amplitude of time waveforms at different measurement sites, the Front CAB (0.5-BF) is used as the reference value. The ratio of each waveform is taken as the amplitude ratio. The bottom value of the amplitude of the time waveform is between the P wave and the R wave of the electrocardiogram. The peak value of the amplitude of the time waveform is after the R wave and up to the start of the second sound of the PCG. Next, we classify the time waveforms of the Front CAB (0.5-BF) of 50 subjects.

(Harmonic component analysis using short-time Fourier transform)
Next, the waveform gradient analysis method and the harmonic component analysis method using the short-time Fourier transform will be explained.
Equation (4) was used to examine which parts of the time waveform the harmonic components are superimposed on, focusing on the harmonics that make up the time waveform of the Front CAB (0.5-BF). The frequency was divided into 1-4 Hz and 4-16 Hz for analysis. The number of window points for 1-4 Hz waveform analysis was set to 2 ⁹ = 512. With a sampling frequency of 1000 Hz, 512 points equals 0.512 seconds, and a power spectrum component of 1.95 Hz appears in the calculations. The shift width of the Hanning window function is set to 51 points. At that time, frequency information is generated every 51/1000 = 0.051 seconds.

The number of window points to capture the E wave, which is made up of harmonics, was set to 2 ⁷ =128. With a sampling frequency of 1000 Hz, 128 points corresponds to 0.128 seconds, and power spectrum components appear at 7.8 Hz. The reason for setting the resolution to 0.12 seconds (7.8 Hz) is that the duration of systole at a heart rate of 90 beats per minute is assumed to be 0.26 seconds, and the required resolution of the detection frequency to capture the peak of the systolic duration is set to 0.13 seconds. In this case, the movement width of the Hanning window function was set to 12 points. At that time, frequency information is generated every 12/1000 = 0.012 seconds.

(Distinguishing vibrations using the autocorrelation function)
A correlogram is drawn using the autocorrelation function (ACF) to distinguish between waveforms. The vertical axis of the correlogram is the ACF, and the horizontal axis is time. The ACF is calculated for each time interval T, and the type and period of the waveform are identified from the correlogram. The ACF, i.e., R _x (τ), is expressed as follows:

で表される。τは時間間隔、ラグ (Lag)を示し、Tは全区間を示す。

where τ is the time interval or lag, and T is the entire interval.

The normalized autocorrelation function has a maximum value of 1 when τ = 0, and this value is the mean square value. When the ACF value is 0.2 or less, it is within a 95% or higher confidence interval for no autocorrelation. Vibration type analysis is determined using 10 seconds of data.

[Front CAB time waveform evaluation results]
Using the normal patterns in the sitting position, Front CAB (0.5-BF) measured from 50 subjects was used for classification and testing. The results are shown in Figures 32 to 36. The waveforms of Front CAB (0.5-BF) are shown in each of Figures 32 to 36 (b), and correspond to the time waveforms of the ECG over one second shown in each of Figures 32 to 36 (a). The waveforms of Front CAB (0.5-BF) are classified into five waveforms. Waveforms similar to the A wave (atrial contraction wave: arrowhead) and the ventricular systolic wave starting from the E wave (marked with an asterisk) of the apical cardiogram are observed.
The classification method for the five waveform patterns is based on Fourier series expansion. However, mathematical expressions are not suitable for intuitive understanding. Therefore, when examining the Front CAB waveform, we classified the characteristics of the Front CAB waveform by using PSD (power spectral density) when focusing on the amplitude, STFT when focusing on the amplitude and phase, and Correlogram when examining whether it has periodicity.
In this specification, the five classified waveforms are named pseudo-sine waves (S waves), pseudo-triangular waves (T waves (triangular waves), C waves (cosine waves), G waves (Gaussian waves)), and pseudo-full-wave rectified waves (R waves). Of these, S waves and G waves correspond to sustained, as seen in visual and palpation based on the duration of the systolic wave, while T waves, C waves, and R waves correspond to tapping. Although these waveforms are not perfect sine waves, triangular waves, or full-wave rectified waves, the waveform shapes are similar, so the word "pseudo" is added to the prefix.

What is needed for waveform analysis is the fundamental harmonic waveform and the amplitude, frequency, and phase of each waveform near the bottom and peak.

In each of Figures 32 to 36 (c), the power spectrum density (PSD) was calculated by frequency analysis and the relationship between amplitude and frequency was classified, and in each of Figures 32 to 36 (e), the type of vibration was classified by correlogram. Using these parameters, the time waveforms of the Front CAB (0.5-BF) were classified into five representative waveforms.

More specifically, the PSD allows classification of waveform patterns as follows:
First, the systolic E wave shown by the S wave of the Front CAB has a small contribution of high frequency components and is close to a sine wave. In other words, only the fundamental frequency component is dominant (the "X" pattern shown in the figure).
Next, the systolic E wave shown by the Front CAB T, C, and G waves is in the process of converging into a triangular wave, and can be classified as a triangular wave. In other words, odd-order frequency components are dominant, and the amplitude decreases in proportion to the inverse of the square of the order as the order increases (the "Y" pattern shown in the figure).
The R wave of the Front CAB has a small amplitude of the fundamental frequency component and a small difference with the amplitude of the integral multiple frequency component, so the triangular wave part is replaced by a full-wave rectified sine wave. The corresponding PSD has even-order frequency components that are dominant, and the amplitude becomes smaller as the order increases (the "Z" pattern shown in the figure).
STFT is suitable for investigating the influence of phase information, and the accuracy of waveform classification is improved by superimposing PSD and STFT. By examining the periodicity between the A and E waves using Correlogram, any waveform can be classified as a variant of the normal waveform in terms of amplitude, phase, and periodicity.

The STFT diagrams in (d) of Figures 32 to 36 are obtained by dividing the power spectrum information from the fundamental harmonic to 5 Hz appearing in (c) of Figures 32 to 36 into two frequencies, center frequencies of 1.95 Hz and 7.81 Hz, doubling the amount of information about the power spectrum and making pattern matching easier (the upper diagram in (d) is 1.95 Hz, and the lower diagram is 7.8 Hz). The information on the fundamental harmonic and the harmonics between 1 and 5 Hz in (c) of Figures 32 to 36 is plotted between 1 and 5 Hz in the STFT. The time waveform information is represented by red, green, and blue contour lines.

Waveform I (S wave) has three time periods where the fundamental harmonic appears, two time periods where 5-10 Hz components appear, and the interval between the atrial bump and systolic wave is short, indicating that it is a square wave. Waveform II (T wave) has STFTs indicating the appearance of components below 5 Hz concentrated at 2.3-2.5 seconds, and 5-10 Hz components concentrated at 2.5-2.7 seconds, indicating that these components are approximately contained in the systolic phase, indicating that it is a triangular wave. Similarly, waveform III (C wave) is also contained in the systolic phase, and the interval between the atrial bump and systolic wave is short, indicating that it is a triangular wave. Waveform IV (G wave) has a peak value interval of 5-10 Hz close to that of waveform I (S wave), and has a shape that resembles a combination of S waves and T waves, and overall shows the characteristics of a triangular wave. Waveform V (R wave) shows that the STFT peak values below 5Hz show a downward sloping contour line, indicating that the power spectrum from the fundamental harmonic to the higher harmonics is of the same magnitude, and furthermore that there are three or more peak values between 5Hz and 10Hz, and the time phase of their appearance is the same as that of the components below 5Hz, showing the characteristics of a full-wave rectified wave.

The correlograms in Figures 32 to 36 (e) were used to classify the type of vibration (periodicity between the waveforms of A and E waves) by the appearance of sine wave notches, with the autocorrelation coefficient set at 0.5 or more and 0.2 or less as standard values. From the correlogram waveforms of the Front CAB (0.5-BF), i, a sine wave, ii, a waveform with one notch in the sine wave, and v, a waveform with two or more notches, were extracted. The remaining waveforms were divided into iii, with envelopes that decay on one side, and iv, with envelopes that decay on both sides.

[Theoretical basis and objective classification criteria for waveform classification]
Here, we will explain in detail the mathematical approach to the apex wave using Fourier series expansion. These mathematical approaches show that the apex wave of a healthy person can be classified into five types, and the above PSD, STFT, and Correlogram support this classification into five types through an engineering approach.
First, if any periodic function is f(t), f(t) can be expanded into the following Fourier series.
Here, ω = 2π/T, where T is the period.

S波は、その波形形状が正弦波に近いため、式（７）で、1次成分が卓越する状況にあたる。ただし、若干の高調波成分も含まれているため、STFT解析結果からは、基本周波数が主体となり、これに高次周波数成分が伴ったものと解釈できる。
これは、収縮期の持続時間が長い、Sustainedの所見に合致する。

Since the waveform of an S wave is close to a sine wave, the first-order component is dominant in equation (7). However, since it also contains some harmonic components, the STFT analysis results show that the fundamental frequency is dominant, accompanied by higher-order frequency components.
This is consistent with the findings of Sustained, where the duration of systole is long.

Since the T wave is close to a triangular wave, it can be expanded into a Fourier series as follows:

式（８）よりnの次数が高くなるにつれて、急激に展開係数が小さくなってゆくという特徴があることが分かる。
また、スペクトルとして発現する次数は、奇数次であることが分かる。
これによってPSDの形が説明できる。
Correlogramで示される自己相関係数がt=0でピークを持ち、それ以外の時間では自己相関係数び絶対値が１より小さくなることは、正弦波とは別の高調波の波形が影響を及ぼしていることから説明可能である。
これは、正弦波に、収縮早期の持続の短い尖鋭な心室収縮に由来する波形の重畳を意味し、収縮早期の持続時間が短いTappingの所見に合致する。

It can be seen from equation (8) that the expansion coefficients become smaller rapidly as the degree of n increases.
It can also be seen that the orders appearing as spectra are odd orders.
This explains the shape of the PSD.
The autocorrelation coefficient shown in the correlogram has a peak at t=0, and the absolute value of the autocorrelation coefficient is less than 1 at other times. This can be explained by the fact that this is due to the influence of harmonic waveforms other than the sine wave.
This means that a waveform resulting from a sharp ventricular contraction with a short duration of early systole is superimposed on the sine wave, which is consistent with the findings of Tapping, in which the duration of the early systole is short.

The waveform of the C wave resulting from ventricular contraction can be considered in the same way as the T wave. The difference between the C wave and the T wave is that in the Fourier series expansion of equation (8), the C wave has a higher dominance of the first-order component than the T wave.
The results of the STFT analysis show that the waveform resulting from ventricular contraction is dominated by the fundamental frequency and the second harmonic, with frequency components up to the seventh harmonic. The correlogram matches the findings of tapping, in which the duration of the early contraction is short, similar to the T wave.

From STFT analysis, G waves can be considered as a combination of S waves and T waves, with the first-order component being predominant. Because they are closer to S waves than T waves, the duration of the systole is longer, which is consistent with the findings of Sustained.

The waveform of the R wave resulting from ventricular contraction is not single-peaked but multi-peaked, but if we hypothesize that it is similar to a full-wave rectified waveform, it can be expanded into the following Fourier series.

式（９）は、２次成分から始まり、これに偶数次の高調波が重畳したものと考えられる。nが大きくなるとともに展開係数は小さくなる。なお、R 波には1次成分も含まれているため、PSDの図に示すような形をとるものと考えられる。
Correlogramの分析から、T波と同様に1次成分以外が卓越している理由が分かる。これは収縮早期の持続時間が短いTappingの所見に合致する。
以上のとおり、Correlogram からは正弦波の基本周期を持つS波とG波に分類され、高調波が顕在化するT波、C波、R波に分類される。また、ラグに対する自己相関関数の差に注目すると、Correlogramの＋側でS波とG波が区別でき、Correlogramの－側でT波とC波が区別でき、多峰性を示すノッチによりR波が区別できる。
すなわち、Correlogramによる分類は、S波とG波は正弦波的な周期性に着目し、他方T波、C波、R波はひずみ波的な周期性に着目したものである。収縮期波の周期性はCorrelogramの＋側の自己相関関数で分類され、心房性隆起波と他の波形の周期性はCorrelogramの－側の自己相関関数で分類される。S波とG波は、ラグに対する自己相関の差からさらに仕分け、T波、C波、R波は－側の自己相関係数の値(上記の例では、0,-0.2,-0.5)で仕分けることができた。そして心房性隆起波と他の波形の凸凹は描かれた自己相関関数の形(ノッチ)を利用した。R波は多峰であるため、ノッチが2つ以上となった。

Equation (9) can be considered to start with the second-order component and then superimpose even-order harmonics on it. As n increases, the expansion coefficients become smaller. Note that the R wave also contains the first-order component, so it is considered to take the form shown in the PSD diagram.
Analysis of the correlogram reveals why non-primary components are predominant, as with the T wave, and this is consistent with the findings of tapping, where the duration of the early systole is short.
As described above, the correlogram classifies waves into S and G waves, which have a fundamental sine wave period, and into T, C, and R waves, which are manifestations of harmonics. In addition, by looking at the difference in the autocorrelation function against the lag, S and G waves can be distinguished on the positive side of the correlogram, T and C waves can be distinguished on the negative side of the correlogram, and R waves can be distinguished by the notches that show multi-peaks.
In other words, classification by correlogram focuses on the sinusoidal periodicity of S waves and G waves, while the distorted wave-like periodicity of T waves, C waves, and R waves. The periodicity of systolic waves is classified by the autocorrelation function on the positive side of the correlogram, while the periodicity of atrial upswing waves and other waveforms is classified by the autocorrelation function on the negative side of the correlogram. S waves and G waves are further classified based on the difference in autocorrelation against the lag, while T waves, C waves, and R waves can be classified by the value of the autocorrelation coefficient on the negative side (0, -0.2, -0.5 in the above example). The bumps and dips of atrial upswing waves and other waveforms are determined by the shape (notches) of the autocorrelation function drawn. Since R waves are multi-peaked, there are more than two notches.

The waveforms were examined by pattern matching 50 cases using (b) to (d) in Figures 32 to 36. As a result, the occurrence rates of pseudo-sine waves (S waves), triangular waves (T waves), cosine waves (C waves), Gaussian waves (G waves), and pseudo-full-wave rectified waves (R waves) were 4/50=8%, 8/50=16%, 15/50=30%, 11/50=22%, and 12/50=24%, respectively.

FIG. 37 shows the results of waveform analysis using the power spectrum, two types of STFT, and correlogram shown in each of (b) to (e) of FIG. 32 to FIG. 36, without using the waveform test determined in each of (a) of FIG. 32 to FIG. 36. The bold numbers in the figure indicate the values that could be tested. For STFT and correlogram, 43 out of 50 waveforms could be tested, except for one triangular wave. For power spectrum, 38 out of 50 waveforms could be tested. The P values were P=2.2×10 ^-16 for STFT test, P=7.9×10 ^-5 for power spectrum test, and P=2.2×10 ^-16 for correlogram test. According to the Wiener-Khintchine theorem, STFT and correlogram (ACF) are inverse Fourier transform related, so the fact that the P values of both were the same supports the validity of the analysis method. Therefore, by performing these three tests together, it is possible to improve the accuracy of the Front CAB (0.5-BF) waveform test, and it has been found possible to classify with high accuracy the evaluation that was previously performed based on the shape of the ACG waveform. It was also found that the 4SR sensing system has the waveform resolution necessary to detect abnormal findings, such as distinguishing between normal hearts and hypertrophied/enlarged hearts.

Therefore, in this embodiment, the measurement condition estimation means 130 judges the subject to be healthy and outputs the result if the classification result of the apical pulsation wave of the subject obtained by the waveform classification means 170 corresponds to the five patterns of the waveform classification of healthy subjects described above, and judges the subject to have abnormal findings related to cardiac disease and outputs the result if the classification result does not correspond to any of the five patterns.

Figures 38 (a) to (e) show the frequency analysis results of Front CAB(0.5-BF), Front CAB(5-BF), Rear CAB(0.5-BF), and Lumbar CAB(0.5-BF) of five subjects as representative examples of the five waveforms defined in each of Figures 32 to 36 (a). Figure 39 shows the average value and standard deviation of the amplitude ratio of CAB by measurement site and the frequency analysis results. The amplitude ratio is an index that indicates the ease of measurement based on the difference in the amplitude of CAB by measurement site relative to Front CAB(0.5-BF), and is calculated using the maximum amplitude value of the ventricular contraction wave from the start of the left ventricular pressure rise of the systolic wave shown in the above section (Front CAB time waveform analysis). The amplitude ratio of Front CAB(5-BF) can be calculated from the ratio of Front CAB(5-BF) to Front CAB(0.5-BF). Similarly, the amplitude ratio of the Rear CAB(0.5-BF) is calculated from the ratio of the Rear CAB(0.5-BF) to the Front CAB(0.5-BF), and the amplitude ratio of the Lumbar CAB(0.5-BF) is calculated from the ratio of the Lumbar CAB(0.5-BF) to the Front CAB(0.5-BF). The amplitude ratio of the Front CAB(5-BF) to the Front CAB(0.5-BF) is 0.2 because it does not contain frequency components between 0.5 Hz and 5 Hz. The amplitude ratio of the Rear CAB(0.5-BF) is 1.1, and the amplitude ratio of the Lumbar CAB(0.5-BF) is 2.6. The average and standard deviation were calculated using the amplitude ratios of 50 subjects.

Next, we will consider the power spectra in Figures 38(a)-(e) and the amplitude in Figure 39. The power spectrum near 5 Hz increased in the Rear CAB (0.5-BF) and Lumbar CAB (0.5-BF). Specifically, the power spectrum near 5 Hz increased by 1.8 to 4.8 times in the Rear CAB (0.5-BF), and by 3.7 to 7.9 times in the Lumbar CAB (0.5-BF). The difference in the power spectrum of the harmonics near 5 Hz is due to the length of the aorta, suggesting that a liquid column resonance phenomenon occurred due to the aorta and pressure pulsation.

As described above, according to the present invention, the following specific effects were obtained.
The conventional 3SR is a sensing system developed for mounting on vehicles, which attenuates external vibrations by the air damping effect of the air pack, and extracts the first and second heart sounds, which contain frequency components of 20 Hz or higher, and the periodic components of heart rate fluctuations. On the other hand, the 4SR of the present invention was developed for indoor use, and was found to be suitable as a sensor for capturing the period and amplitude of the apex beat, which is in the vicinity of 1 to 2 Hz.

The F-APW produced a waveform similar to the apex beat wave. The waveforms of the apex beat wave and heart sounds could be generated from the F-APW. It is believed that the R-APW and L-APW produced different waveforms due to a resonance phenomenon caused by pressure pulsation. However, even for the R-APW and L-APW, which produced waveforms different from the F-APW, vibration information (CAB) related to cardiac movement, including the apex beat, and heart sounds (CAS) could be extracted by applying vibration analysis, demonstrating the possibility that this could be of clinical significance.

By combining the BF extraction method from the harmonic disappearance frequency of the F-APW with the BF extraction method using the STFT method, a correlation diagram of BF-heart rate was created. From the actual measurement results, BF could be expressed as a quadratic function of heart rate, making it possible to determine BF in real time from the heart rate using only the F-APW data without using PCG. It was also suggested that the visualization method using the STFT method may enable not only vibration analysis of the heart vibration and aortic pressure pulsation, but also diagnosis based on the degree of change.

The results of the experiment on heart rate fluctuations showed that when the heart rate is 70-80 beats per minute in a resting state, the fluctuation is 1/f, suggesting that efficient exercise is being performed.The results of the experiment on heart rate fluctuations showed that when the BF is around 10 Hz, the fluctuation is 1/f, suggesting that efficient exercise is being performed. Utilizing this characteristic may make it possible to quantify the changes that drugs have on cardiovascular function.

As shown above, BF is determined from the heart rate (for 10 seconds in this case). The value of heart rate variability fluctuation (for 90 seconds) obtained from that BF tended to match the value of heart rate variability fluctuation (for 90 seconds) obtained directly from the heart rate (for 90 seconds). The results obtained by substituting these relationship equations support the validity of BF.

Furthermore, by applying a bandpass filter to the APW in the range from 0.5 Hz to BF, it was possible to draw a waveform that approximated the ACG, and the Front CAB (0.5-BF), which represents the apical beat, could be broadly classified into five waveforms using frequency analysis, time series analysis using the STFT method, and waveform analysis using a correlogram.

When heart rate fluctuations occur during respiratory depression at rest, BF moves along the regression curve of the BF-HR correlation diagram (see Figure 40). This suggests the possibility that the vibration analysis method and BF-HR correlation diagram described above can be used to analyze cardiac motion and pressure pulsation, as well as evaluate cardiac motion and intraventricular blood flow.

In addition, the same BF-HR correlation diagram was obtained in the supine and left and right lateral positions as in the sitting position (see Figure 41). This suggests the possibility that the BF-HR correlation diagram can be estimated from a single diagram, regardless of posture, and the measurement method using 4SR is highly practical as data can be easily obtained, making it possible to apply it to objectify visual and palpation findings. APW has the potential to be a physical diagnostic tool that non-invasively evaluates the dynamics of the cardiovascular system and serves as an indicator of pathology and treatment.

According to the present invention, signals spanning two frequency bands are separated by identifying the BFs.
In other words, by separating the signals corresponding to the apex beat and heart sounds from the APW, it becomes possible to quantitatively recognize the results of different types of cardiac activity by observing one physiological signal, and useful complementary information on the cardiac state can be obtained. In addition, by increasing the waveform resolution through signal separation, for example, the frequency at the boundary between two frequency bands and the harmonic components of the Front CAB and the dominant frequency components of the Front CAS in the vicinity of the frequency band may become new parameters of the cardiocirculatory system.
In addition, Front CAB has pressure information of the systolic wave of the apexcardiogram, and Front CAS has time-phase information regarding the cardiac cycle. By combining the pressure and time-phase information, it is believed that the complementary information can be used for diagnostic purposes.

For example, the first heart sound can be identified from the Front CAS, and the notch corresponding to the first sound in the ascending limb of the systolic wave can be observed in a wide or narrow range in the STFT shown in (b) and (c) of Figures 32 to 36. The observable range can be changed by changing the window width of the 0 to 20 Hz band, and the notch can be easily identified by adjusting the contrast. This can be used to determine the presence or absence of a notch in the systolic pressure wave and to identify the cause of abnormal cardiac activity, and can be used for diagnosis. Front CAB, which is vibration information of the heart, and Front CAS, which is acoustic information of the heart, mainly indicate the mechanical properties of the atrium, ventricle, valves, blood, and aorta. The result of their complementarity is expressed as fluctuations in blood pressure, heart rate, and respiratory rate. It can also be used to quantify changes that occur during drug therapy while a patient is being administered medication. For example, in diagnosing left ventricular failure, comparing the mechanical properties of the left ventricle that should be watched out for with those of healthy people of the same age group can also be useful in diagnosing the presence of left ventricular failure. Furthermore, by correlating with ECG information such as conduction pathways, velocity characteristics, and potentials, it may be possible to estimate the health of myocardial cells. Front CAB and Front CAS information may provide data relevant for cardiac status monitoring and functional evaluation. As a result, it may be possible to model the ventricular size, heart wall thickness, and mechanical properties of the heart, and to describe the motion mathematically.
Furthermore, by measuring data along the aorta, it may be possible to separate aortic hemodynamics from that resulting from the apical pulsation or heart sounds, which may provide new insight into aortic hemodynamics.

As is clear from the above, by extracting acoustic vibration information (Acoustic Pulse Wave: APW) from 0.5 to 80 Hz and applying a specified filter process, it is possible to generate an ACG equivalent waveform (Cardiac Apex Beat: CAB) and a heart sound equivalent waveform (Cardiac Acoustic Sound: CAS).
CAB and CAS have different sources and transmission pathways, and the superposition of these two vibrations forms the APW. If contractility is considered to be an inherent property of the myocardium and independent of preload and afterload, the CAB waveform during systole contains information about the influence of chemical stimuli and hormones during contraction. If CAS is considered to contain information about the movement of blood ejected from the ventricle during systole, the APW contains information about preload, afterload, and contractility.
Therefore, APW, CAB, and CAS contain a lot of information about sound, vibration, and pressure, and by observing them in conjunction with the polarization phenomenon of ECG, it is possible to diagnose cardiac malfunctions.
Real-time measurement of APW, CAB, and CAS adds clinically significant sound, vibration, and pressure information, and it was found that the engineering approach used for machine failure diagnosis can be applied to health checks. In addition, the 4SR system of the present invention that measures APW can become a physical diagnostic tool that can obtain bioinformation non-invasively and repeatedly over a long period of time, and is expected to have a wide range of applications, including analysis using AI including the above-mentioned machine learning.

1. Biosignal detection sensor (4SR)
REFERENCE SIGNS LIST 10 Three-dimensional knitted fabric 20 Storage film 30 Microphone 40 Case 50 Gel 100 Biological signal analysis device 110 Frequency analysis means 120 Boundary frequency identification means 130 Measurement state estimation means 140 Heart rate calculation means 150 Correlation data storage unit 160 Apex beat waveform extraction means 170 Waveform classification means 180 Model creation means 1000 Biological signal detection sensor (3SR)

Claims (24)

a frequency analysis means for performing frequency analysis on biosignal data obtained by a biosignal detection sensor via a body surface;
a boundary frequency specifying means for determining a boundary frequency between a vibration caused by an apex beat and a vibration caused by a heart sound in the biosignal data from a frequency analysis result of the biosignal data obtained from the frequency analysis means,
The boundary frequency identification means
The biosignal analysis device includes a means for determining a sudden change portion of the power spectrum that is a boundary between harmonic vibration and irregular vibration in the frequency analysis result, and for specifying the boundary frequency based on this sudden change portion . The boundary frequency identification means
2. The biosignal analyzer according to claim 1 , further comprising means for determining an abrupt change in said power spectrum by adding a result of frequency analysis of simultaneously measured heart sound data to the result of said frequency analysis. The boundary frequency identification means
3. The biosignal analyzer according to claim 2, further comprising: a double logarithmic axis display means for displaying, on a double logarithmic axis, a waveform obtained by averaging the results of frequency analysis of the biosignal data and the heart sound data, respectively, using a logarithmic difference method; and a sudden change portion identifying means for determining a fluctuation change point from the waveform displayed on the double logarithmic axis and identifying the fluctuation change point as a sudden change portion of the power spectrum. A short-time Fourier transform is applied as the frequency analysis means,
3. The biosignal analyzer according to claim 1 , wherein the boundary frequency identifying means includes means for determining an abrupt change portion of the power spectrum from the analysis result of the short-time Fourier transform. the frequency analysis means includes means for outputting an analysis result of the short-time Fourier transform into image data indicating the degree of fluctuation in time, frequency, and power spectrum;
5. The biosignal analyzing apparatus according to claim 4 , wherein said boundary frequency identifying means includes means for determining an abrupt change portion of said power spectrum from said image data. a correlation data storage unit in which correlation data relating to the boundary frequency is stored;
6. The biosignal analyzer according to claim 1 , further comprising a measurement-time condition estimating means for estimating the health condition of the subject at the time of measurement by collating the correlation data. the correlation data is correlation data between the boundary frequency and a heart rate,
The biosignal analysis device of claim 6, wherein the measurement state estimation means includes means for estimating the boundary frequency of the subject at the time of measurement by comparing the subject's heart rate at the time of measurement with correlation data between the boundary frequency and the heart rate. the correlation data is correlation data between the boundary frequency and a fluctuation characteristic of heart rate variability,
The biosignal analysis device of claim 6, wherein the measurement state estimation means includes means for estimating the fluctuation characteristics of the heart rate fluctuation of the subject during measurement by comparing the boundary frequency of the subject during measurement with correlation data between the boundary frequency and the fluctuation characteristics of the heart rate fluctuation. an apex pulsation waveform extraction means for filtering the biological signal data by setting the boundary frequency identified by the boundary frequency identification means as an upper limit value, and obtaining a waveform of vibration caused by the apex pulsation;
A waveform classification means for classifying the waveform of the vibration generated by the apex beat of the heart,
7. The biosignal analysis device of claim 6, wherein the measurement condition estimation means includes means for classifying, by the waveform classification means, the waveform of vibration generated by the apex pulsation of the subject at the time of measurement, which is obtained by the apex pulsation waveform extraction means, and estimating the health condition of the subject at the time of measurement based on data of the waveform classification result at the time of measurement. The correlation data storage unit stores correlation data between the waveform classification result and a health condition in advance,
10. The biosignal analysis device of claim 9, wherein the measurement condition estimation means includes means for classifying, by the waveform classification means, the waveform of vibration caused by the apex beat of the subject at the time of measurement, which is obtained by the apex beat waveform extraction means, and collating the waveform classification result at the time of measurement with pre-stored correlation data between the waveform classification result and health condition, to estimate the health condition of the subject at the time of measurement. a model creation means for generating an estimation model by machine learning to estimate the health condition from the waveform information by using the waveform information output by the apex beat waveform extraction means and the health condition as teacher data,
The biosignal analysis device of claim 9, wherein the measurement condition estimation means includes means for receiving as input waveform information of vibrations caused by the apex beat of the heart of the subject at the time of measurement, using an estimation model created by the model creation means, and outputting a value of the health condition at the time of measurement from the acquired waveform information at the time of measurement. The biological signal detection sensor includes:
Three-dimensional knitting and
A containing film that hermetically covers the periphery of the three-dimensional knitted fabric;
The biosignal analysis device of any one of claims 1 to 11 , comprising a microphone arranged on the outside of the containing film, a case covering the microphone, and a disturbance intrusion suppression member within the case that functions to suppress the intrusion of disturbances into the microphone. A computer program for processing biosignal data obtained by a biosignal detection sensor via a body surface and causing a computer to function as a biosignal analysis device,
A step of subjecting the biological signal data to frequency analysis;
and causing the computer to execute a procedure of identifying a boundary frequency between a vibration caused by an apex beat and a vibration caused by a heart sound in the biosignal data based on the frequency analysis result ;
In the step of identifying the boundary frequency,
A computer program for determining a sudden change in the power spectrum that is the boundary between harmonic vibration and irregular vibration from the frequency analysis results, and for specifying the boundary frequency based on this sudden change . In the step of identifying the boundary frequency,
14. The computer program according to claim 13 , further comprising: a step of calculating a sudden change in the power spectrum by adding a result of frequency analysis of simultaneously measured heart sound data to the result of the frequency analysis. In the step of identifying the boundary frequency,
15. The computer program according to claim 14, further comprising: displaying waveforms obtained by averaging the results of frequency analysis of the biosignal data and the heart sound data on a double logarithmic axis using a logarithmic difference method; determining a fluctuation change point from the waveform displayed on the double logarithmic axis; and identifying the fluctuation change point as a sudden change portion of the power spectrum. In the step of performing the frequency analysis, a short-time Fourier transform is applied,
15. The computer program according to claim 13 , wherein in the step of identifying the boundary frequency, abrupt changes in the power spectrum are obtained from the analysis results of the short-time Fourier transform. In the step of performing the frequency analysis, the analysis result of the short-time Fourier transform is outputted as image data indicating the degree of fluctuation in time, frequency and power spectrum;
17. The computer program according to claim 16 , wherein in the step of identifying the boundary frequency, a sudden change portion of the power spectrum is obtained from the image data. The computer program according to any one of claims 13 to 17 , which executes a procedure of estimating the health condition of the subject at the time of measurement by comparing the correlation data related to the boundary frequency stored in a correlation data storage unit. the correlation data is correlation data between the boundary frequency and a heart rate,
20. The computer program of claim 18, wherein the step of estimating the health condition at the time of measurement includes estimating the boundary frequency of the subject at the time of measurement by comparing the subject's heart rate at the time of measurement with correlation data between the boundary frequency and the heart rate. the correlation data is correlation data between the boundary frequency and a fluctuation characteristic of heart rate variability,
The computer program of claim 18, wherein the step of estimating the health condition at the time of measurement includes comparing the boundary frequency at the time of measurement of the subject with correlation data between the boundary frequency and the fluctuation characteristics of heart rate fluctuation to estimate the fluctuation characteristics of the heart rate fluctuation of the subject at the time of measurement. Further, a step of filtering the biological signal data by setting the boundary frequency as an upper limit value, and extracting an apex pulsation waveform for obtaining a waveform of vibration caused by the apex pulsation;
and classifying the waveform of the vibration caused by the apex beat of the heart.
20. The computer program of claim 18, wherein the step of estimating the health condition at the time of measurement includes classifying a waveform of vibration generated by the apex beat of the subject at the time of measurement by executing a step of classifying the waveform, and estimating the health condition of the subject at the time of measurement based on the waveform classification result. the correlation data storage unit stores correlation data between the waveform classification result and a health condition,
22. The computer program of claim 21, wherein the step of estimating the health condition at the time of measurement includes classifying a waveform of vibration caused by the apex beat of the subject at the time of measurement by executing a step of classifying the waveform of vibration caused by the apex beat of the subject, and estimating the health condition of the subject at the time of measurement by comparing the waveform classification result at the time of measurement with correlation data between the waveform classification result and the health condition. generating an estimation model by machine learning that estimates the health condition from the waveform information, using waveform information that is output by executing the procedure of extracting the apex beat waveform and is to be classified in the procedure of classifying the waveform and the health condition as teacher data;
The computer program according to claim 21, wherein the step of estimating the health condition at the time of measurement uses waveform information of vibrations generated by the apex beat of the subject at the time of measurement as input, and uses the estimation model to output a value of the health condition at the time of measurement from the acquired waveform information at the time of measurement. A recording medium on which the computer program according to any one of claims 13 to 23 is recorded.

Images