Osteoarthritis Diagnosis System

Abstract

To provide a simple and accurate method for diagnosing knee arthropathy.SOLUTION: An arthropathy diagnosis system (S) includes: a sound measuring member (1) for measuring a sound from a joint; an angle measuring member (7) for measuring a joint angle of the knee; frequency analysis means (54) for executing frequency analysis of a sound signal for each joint angle; and a classifier (57) having a non-linear classifier (57b) for discriminating a disease on the basis of information entropy and energy, and a linear classifier (57a) for discriminating a disease on the basis of energy and an auto-correlation coefficient, for discriminating a disease of a joint on the basis of the discrimination result by the non-linear classifier (57b) and the discrimination result by the linear classifier (57a).SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to an osteoarthritis diagnosis system. [Background technology]

[0002] Abnormalities in joints such as the knee and elbow joints are common clinical diseases that gradually degenerate the function of the joints due to many causes such as aging, obesity, and mechanical load. There are known non-invasive and invasive diagnostic methods for joint diseases such as knee osteoarthritis (knee OA). Invasive diagnoses (which damage a part of the body) include synovial fluid testing and arthroscopy. As a non-invasive diagnostic method, imaging diagnosis using medical images such as X-rays and MRI scans is performed. As a non-invasive diagnostic method other than imaging diagnosis, research and development is also being conducted on diagnostic methods that measure vibrations and sounds (joint sounds) generated in the joint during flexion and extension of the knee joint. As a technique for measuring joint sounds, the techniques described in Patent Documents 1 to 4 are conventionally known.

[0003] Patent Document 1 (WO 2011 / 096419) describes a technology in which a bioacoustic sensor and an angle sensor are attached to the skin near the joint of a subject, and the acceleration of the subject's joint when flexed and extended is measured with a weight meter while also measuring acoustic data and angle data. The waveform of the acoustic data is frequency-analyzed using Fourier analysis to generate frequency spectrum characteristic data, which is compared with previously prepared spectrum characteristic patterns, and the symptom grade corresponding to the most correlated pattern is used to diagnose the symptom grade of the subject's knee osteoarthritis.

[0004] Patent Document 2 (JP Patent Publication 2004-261525A) describes a technique for analyzing the waveform of acceleration data detected by an acceleration sensor attached to the tibia or heel of a subject, and comparing the time between peaks of the waveform and the ratio of the power spectrum with a threshold value to determine whether the condition is normal or knee OA. In Patent Document 2, when calculating the power spectrum, a fast Fourier transform is performed to analyze the frequency components of the acceleration.

[0005] Patent Document 3 (JP Patent Publication No. 2016-168209) and Patent Document 4 (JP Patent Publication No. 2016-168219) describe a technology in which vibrations associated with bending and extension movements of an object to be examined are detected by vibration sensors (12a-12d), and the bending and extension angles during bending and extension movements are detected by an angle sensor (14), the vibration signals are divided according to the bending and extension angles of 0°-30°, 30°-60°, and 60°-90°, frequency analysis is performed on each divided vibration signal, and the power values ​​of the frequency bands are compared with the values ​​of healthy subjects to determine whether the cartilage is deteriorating. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] International Publication No. 2011 / 096419 ("0085" to "0098") [Patent Document 2] JP 2004-261525 A ("0043" to "0106") [Patent Document 3] JP 2016-168209 A ("0026" to "0041") [Patent Document 4] JP 2016-168219 A ("0026" to "0041") Summary of the Invention [Problem to be solved by the invention]

[0007] (Problems with the Prior Art) The aforementioned X-ray examination has the problem of radiation exposure, and medical images such as MRI are time-consuming and expensive, and require large-scale equipment, making them difficult to use. There are also problems such as the fact that findings on medical images do not necessarily correlate with the pain reported by patients themselves. Although the methods using joint sounds and acceleration as in Patent Documents 1-4 are simpler than those using medical images, there is a problem that the diagnostic accuracy is insufficient. In particular, when the frequency characteristics are obtained using Fourier transform, which is generally used in frequency analysis, as in the techniques described in Patent Documents 1 and 2, information in the time domain is lost, so it is not possible to know how the frequency characteristics of the signal change from moment to moment over time. For example, it is not possible to know whether a signal of a certain frequency is observed at the beginning or end of bending a joint, and the same analysis result is obtained whether the joint sound is observed at the beginning or end of bending a joint, so that analysis corresponding to the movement cannot be performed, and it is difficult to improve the diagnostic accuracy.

[0008] In addition, the techniques described in Patent Documents 1-4 have a problem in that noises generated by body movements during bending and stretching are adversely affected. Furthermore, in the techniques described in Patent Documents 3 and 4, time-frequency analysis is performed to determine the possibility of disease, but the determination is made by a doctor who looks at a graph of the analysis signal, and diagnosis itself is not possible. In other words, there is no description of a classifier that classifies the presence or absence of disease or the grade (malignancy) of the disease. In particular, when a disease is caused by an injury or illness, the malignancy often rapidly worsens from a normal state, but when a disease is caused by aging, the malignancy often gradually worsens, and there is a demand for accurate diagnosis of the malignancy.

[0009] The present invention is a simple and highly accurate Arthropathy The technical objective is to provide a method for diagnosing the above-mentioned condition. [Means for solving the problem]

[0010] In order to solve the above technical problem, the osteoarthritis diagnostic system of the invention described in claim 1 comprises: a sound measuring member that is attached to a subject and measures sounds from the joints; When worn by the subject, Joint angle An angle measuring member for measuring a frequency analysis means for performing a frequency analysis of a sound signal measured by the sound measuring member for each joint angle measured by the angle measuring member; an autocorrelation coefficient deriving means for calculating an autocorrelation coefficient, which is a similarity of a signal waveform in a time domain, based on the signal waveform of the sound; an information entropy calculation means for calculating an information entropy for each joint angle based on a distribution of energy with respect to the frequency; Based on the information entropy and the energy The joint A nonlinear classifier for discriminating a disease based on the energy and the autocorrelation coefficient. The joint and a linear classifier for discriminating a disease, based on a discrimination result of the nonlinear classifier and a discrimination result of the linear classifier, The above A classifier for identifying joint diseases; The present invention is characterized by comprising: Effect of the Invention

[0011] According to the invention described in claim 1, a simple and highly accurate Arthropathy It is possible to provide a method for diagnosing the above-mentioned condition. [Brief description of the drawings]

[0012] [Figure 1] FIG. 1 is an explanatory diagram of an osteoarthritis diagnosis system according to a first embodiment of the present invention. [Diagram 2] FIG. 2 is a functional block diagram of the terminal body in the osteoarthritis diagnosis system of the first embodiment. [Diagram 3] FIG. 3 is an explanatory diagram of a process flow in the osteoarthritis diagnosis system of the first embodiment. [Figure 4] FIG. 4 is an explanatory diagram of an example of a joint angle signal in the first embodiment, in which FIG. 4A is an explanatory diagram of a waveform before noise removal, and FIG. 4B is an explanatory diagram of a waveform after noise removal. [Diagram 5] FIG. 5 is an explanatory diagram of an example of time, joint angle, and energy frequency-analyzed in the first embodiment, where FIG. 5A is a graph showing the relationship between time and angle, and FIG. 5B is a graph showing the relationship between time and energy. [Figure 6]FIG. 6 is an explanatory diagram of feature extraction. [Figure 7] FIG. 7 is an explanatory diagram of a classifier, where FIG. 7A is an explanatory diagram of an example of a linear classifier, and FIG. 7B is an explanatory diagram of an example of a non-linear classifier. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0013] Next, specific examples of the present invention (hereinafter, referred to as examples) will be described with reference to the drawings, but the present invention is not limited to the following examples. In the following description using the drawings, illustrations of components other than those necessary for the description are omitted as appropriate to facilitate understanding. EXAMPLES

[0014] FIG. 1 is an explanatory diagram of an osteoarthritis diagnosis system according to a first embodiment of the present invention. In FIG. 1, the osteoarthritis diagnosis system S of the first embodiment has a sound measuring member 1 to be worn by a subject. As an example, the sound measuring member 1 of the first embodiment is configured by a pair of vibration sensors installed on the left and right (inner thigh side and outer thigh side) in contact with the skin at the position of the tibia 15 cm below the knee of the subject. Note that, in the first embodiment, a configuration for detecting vibrations when sounds (crepitus sounds) at a joint are transmitted as vibrations is exemplified, but is not limited to this. For example, a configuration for measuring sounds with a microphone via a stethoscope, a microphone directly contacting the skin, or any measuring device capable of measuring sounds other than those of a microphone or vibrations corresponding to sounds can be adopted. In the first embodiment, the sound measuring member 1 measures sounds with a pair of left and right sounds, that is, two channels in total, in order to simultaneously measure the right leg and the left leg, but is not limited to this. For example, if only one leg needs to be measured, it is possible to use one channel. It is also possible to use three or more channels, for example, three channels corresponding to three locations of both knees and ankles.

[0015] In the first embodiment, the subject is fitted with an angle measuring member 7 for measuring the knee angle. The angle measuring member 7 in the first embodiment is, for example, a three-dimensional acceleration sensor fixed to the thigh via a band 8. That is, the angle measuring member 7 detects the direction of gravity (directly downward) from the gravitational acceleration, and measures the angle of the thigh (thigh) based on the direction of gravity. In the embodiment, a three-dimensional acceleration sensor is used as the angle measuring member 7, but the present invention is not limited to this. For example, a configuration in which markers are attached to the positions of the waist, knee, and heel and a camera is used to capture the image (so-called motion capture) to detect the knee joint angle, or any measuring device such as a potentiometer that can directly or indirectly measure the rotation angle, or an image analysis device that does not use a marker, can be used. The joint angle measuring device can be not only a wearable configuration as in the first embodiment, but also a non-contact configuration. In the first embodiment, the diagnosis of knee osteoarthritis is performed by measuring the joint sounds while the subject stands up from a seated position in a chair. In the first embodiment, as an example, the joint sounds are measured while the subject repeats the sitting and standing up movements three times, but the number of times can be changed appropriately according to the required accuracy and the subject's ability to bear, such as only one standing up movement. In the first embodiment, the standing up movement from a chair is exemplified, but the number of times is not limited to this, and it is also possible to evaluate movements during daily life movements such as going up and down stairs and walking. Note that excessive movements (large movements or intense movements) are not very preferable because they may complicate noise.

[0016] The sound measuring member 1 and the angle measuring member 7 are connected to a personal computer 11 as an example of a terminal via a communication cable 10. The personal computer 11 has a computer main body 12, a display 13 as an example of a display unit, and a keyboard 14 and a mouse 15 as examples of an input unit. Note that, in the first embodiment, a configuration in which the personal computer 11, the sound measuring member 1, and the angle measuring member 7 are connected via the cable 10 is exemplified, but this is not limiting, and it is also possible to transmit and receive information using any wireless communication method such as a mobile phone line, Bluetooth (registered trademark), or wireless LAN.

[0017] (Description of the control unit of the computer main body 12 in the first embodiment) FIG. 2 is a functional block diagram of the terminal body in the osteoarthritis diagnosis system of the first embodiment. FIG. 3 is an explanatory diagram of a process flow in the osteoarthritis diagnosis system of the first embodiment. In FIG. 2, the control unit 41 of the computer main body 12 of the first embodiment is composed of a computer device having an I / O (input / output interface) for inputting and outputting signals from and to the outside and adjusting input / output signal levels, a ROM (read-only memory) in which programs and data for performing necessary startup processing are stored, a RAM (random access memory) for temporarily storing necessary data and programs, a CPU (central processing unit) for performing processing according to the startup program stored in the ROM or the like, and a clock oscillator, and various functions can be realized by executing the programs stored in the ROM and RAM or the like. The control unit 41 stores basic software for controlling basic operations, a so-called operating system OS, a knee osteoarthritis diagnosis program AP1 as an example of an application program, and other software (not shown).

[0018] (Elements connected to the control unit 41 in the first embodiment) The control unit 41 receives output signals from signal output elements such as the keyboard 14, the mouse 15, the sound measuring member 1, the angle measuring member 7, and the like. Moreover, the control unit 41 in the first embodiment outputs a control signal to a controlled element such as the display 13 .

[0019] (Function of control unit 41) The knee osteoarthritis diagnosis program AP1 of the control unit 41 in the first embodiment has the following functional means (program modules) 51 to 57.

[0020] The joint sound measuring means 51 measures the sound (joint sound s0) from the knee joint based on the signal from the sound measuring member 1. The joint sound measuring means 51 of the first embodiment measures historical data of the joint sound s0. The joint angle measurement means 52 acquires the rotation angle θ0 of the knee joint based on a signal from the angle measurement member 7. The joint angle measurement means 52 of the first embodiment measures history data of the joint angle θ0. Note that, in the first embodiment, the measurement of the joint angle θ0 is started simultaneously with the measurement of the joint sound s0. That is, the timing of the waveform data of the joint sound s0 and the waveform data of the joint angle θ0 are synchronized. The noise removal means 53 has noise removal means 53a for the angle signal and noise removal means 53b for the joint sound signal, and removes noise from each of the measured signals.

[0021] FIG. 4 is an explanatory diagram of an example of a joint angle signal in the first embodiment, in which FIG. 4A is an explanatory diagram of a waveform before noise removal, and FIG. 4B is an explanatory diagram of a waveform after noise removal. The noise removal means 53a of the angle signal performs a fast Fourier transform (FFT) on the waveform of the joint angle θ0 to extract high-frequency components (20 Hz or higher) and low-frequency components (0 to 20 Hz). In the first embodiment, for the low-frequency components, the center frequency (middle of the spectrum intensity) of the low-frequency components is set as the upper limit frequency of the low-pass filter, and components higher than the center frequency are removed as noise. Here, the period of the angle change during the subject's standing up and sitting down movements was approximately 2 to 2.5 seconds, and the frequency was 0.4 Hz to 0.5 Hz. Then, the low frequency of the signal obtained from the joint angle sensor (angle measurement member 7) (i.e., the signal after passing through the low-pass filter) was obtained, and the center frequency was about 2 Hz. This is determined to represent the fluctuation (movement) of the body.

[0022] In addition, in the first embodiment, after applying a low-pass filter to the low-frequency components by the noise removal means 53a of the angle signal, data for only one movement (one standing and sitting movement or one sitting movement) is extracted from the measurement data of three consecutive repeated standing and sitting movements as a series of movements of the subject. Specifically, the start of the standing movement is the timing when the angle starts to become small (maximum value). The end time of the standing movement (start time of the sitting movement) is the timing when the angle becomes minimum (minimum value). The end time of the sitting movement is the timing when the angle becomes maximum (maximum value).

[0023] The joint sound signal noise removal means 53b removes high frequency noise (noise) from the waveform of the joint sound s0. The joint sound signal noise removal means 53b of the first embodiment performs a short time Fourier transform (STFT) on the waveform of the joint sound s0 to perform frequency analysis. Note that the fast Fourier transform (FFT) and the short time Fourier transform (STFT) are conventionally known, and therefore detailed explanations are omitted. A high pass filter is applied to the frequency analysis results derived by the STFT. In the first embodiment, as an example, 2.5 Hz, which is the same as the upper limit frequency for removing baseline drift, is used as the lower limit value of the high pass filter to remove low frequency noise of 2.5 Hz or less. The portion corresponding to the body fluctuation described above is then removed by the high pass filter from the joint sound signal. In other words, the high pass processing is performed not on the signal of the joint angle sensor (angle measuring member 7) but on the joint sound signal. As mentioned above, the body fluctuation component can be considered to be around 2 Hz, but when correcting the baseline drift of the waveform of the actual joint sound signal, the threshold of the high-pass filter is set to 2.5 Hz rather than 2 Hz in order to make it a slightly softer threshold (intentionally weakening the filter effect).

[0024] In addition, in the first embodiment, high frequency noise is removed by removing baseline drift using a high pass filter and by performing principal component analysis (PCA) on a spectrogram (a result of frequency analysis arranged in chronological order). The high-pass filter for removing baseline drift is set to the aforementioned soft threshold (a deliberately weakened threshold) of 2.5 Hz. In the principal component analysis for removing high frequency noise, first, a spectrogram is obtained by performing principal component analysis on the signal. Then, signals that are not considered as principal components are removed as noise. In this case, the first to third principal components are considered as principal components, and the fourth and higher principal components are removed as noise. After the noise removal process is performed, the joint sound signal noise removal means 53b of the first embodiment performs an inverse short-time Fourier transform (Inverse-STFT) to reconstruct the original time series signal (joint sound signal) from which the noise has been removed.

[0025] The analysis means 54 performs frequency analysis on the joint sound signal. The analysis means 54 of the embodiment performs frequency analysis by performing a short-time Fourier transform on the waveform of the joint sound signal from which the noise components have been removed. When analyzing the characteristics of the frequency domain, the selection of the window function, the window length, and the shift distance are changed from those in the joint sound signal noise removal means 53b. As an example, the window function during noise removal can be a square (rectangular) window function with a window length of 30 and an overlap of 20%, and the window function during frequency analysis can be a Hamming window function with a window length of 20 and an overlap of 20%. It is preferable to select parameters such as the window function, the window length, and the overlap according to the balance between the time resolution and the frequency resolution. In frequency analysis, if it is desired to increase the time resolution, it is preferable to shorten the window length. If it is desired to increase the frequency resolution, it is preferable to increase the window length. From the spectrogram obtained by short-time Fourier transform, the energy in the time domain and the frequency domain can be compressed to obtain a graph of time-energy and frequency-energy.

[0026] The autocorrelation coefficient derivation means 55 derives an autocorrelation coefficient based on frequency analysis. The autocorrelation coefficient derivation means 55 of the first embodiment derives an autocorrelation coefficient, which is the similarity of the signal waveform in the time domain, based on the waveform of the joint sound signal (sound signal waveform). That is, a feature amount in the time domain, which indicates the degree to which a similar pattern appears in the signal, is calculated by using the autocorrelation coefficient. As an example, the power spectral density derived by performing STFT on the waveform of the joint sound signal can be further Fourier transformed to obtain the autocorrelation coefficient (Wiener-Khintchine theorem). Specifically, consider a graph (waveform) with time (discrete time) on the horizontal axis and the autocorrelation coefficient of the signal on the vertical axis, calculate the time integral of that waveform, and use this as one of the features in the time domain. However, since the waveform of the autocorrelation coefficient mentioned above is calculated for discrete time, when integrating the waveform over time, the waveform is approximated to a continuous time waveform before being integrated over time.

[0027] FIG. 5 is an explanatory diagram of an example of time, joint angle, and energy frequency-analyzed in the first embodiment, where FIG. 5A is a graph showing the relationship between time and angle, and FIG. 5B is a graph showing the relationship between time and energy. The information entropy calculation means 56 calculates information entropy from the distribution of energy for each joint angle (i.e., for each time of standing up and sitting down). The information entropy in the first embodiment indicates the amount of information and the complexity of the energy, and the difficulty of generating (the difficulty of occurring) the energy, and is expressed by the following formula, where P(A) is the probability that energy A is generated, and S(A) is the information entropy. S(A)=-Σ[P(A) logP(A)] In FIG. 5, the information entropy calculation means 56 of the first embodiment calculates information entropy for the following four periods A, B1, B2, and C, using 120° as the reference angle at which the subject's center of gravity changes significantly during standing up, and extracts features. Period A: From the start of standing to when the joint angle reaches 120° Period B1: From joint angle 120° to the end of standing up (start of sitting down) Period B2: From the start of sitting to the time when the joint angle reaches 120° Period C: From joint angle 120° to the end of sitting (start of standing) As described above, by dividing the data by joint angle, it is expected that the relationship between the degree of bending of the joint, i.e., the contact position between the femur and tibia, and the information entropy (i.e., the characteristic quantity of damage) can be understood. In other words, it is expected that the joint and which part is damaged can be understood. In addition, since the load on the joint is considered to be different in the standing up and sitting down movements, it is expected that by dividing the standing up and sitting down, the relationship between the load on the joint and the information entropy (the characteristic quantity of damage) can be understood.

[0028] FIG. 6 is an explanatory diagram of feature extraction. Information entropy S(A) is a feature obtained by compressing the spectrum along the vertical axis in a graph with joint angle (i.e., passage of time) on the horizontal axis and frequency (energy) on the vertical axis (see spectrum compression 1 in Figure 6). More specifically, it is a feature calculated by summing up the energy at each joint angle.

[0029] The classification means (classifier) ​​57 has a linear classifier 57a and a non-linear classifier 57b, and discriminates joint diseases. The linear classifier 57a performs classification (disease discrimination) based on the value of the linear combination of features. The linear classifier 57a of the first embodiment performs disease discrimination based on the energy feature amount and the autocorrelation coefficient. The energy feature amount is a feature amount obtained by compressing the spectrum in the horizontal axis direction (see spectrum compression 2 in FIG. 6) in a graph in which the horizontal axis represents the joint angle (i.e., the passage of time) and the vertical axis represents the frequency (energy), specifically, a feature amount obtained by calculating the total value of the intensity of each frequency band (energy) over the entire period.

[0030] FIG. 7 is an explanatory diagram of a classifier, where FIG. 7A is an explanatory diagram of an example of a linear classifier, and FIG. 7B is an explanatory diagram of an example of a non-linear classifier. The linear classifier 57a determines the presence or absence of a disease based on whether or not a value obtained by linearly combining the energy feature amount and the autocorrelation coefficient reaches a predetermined threshold (see the straight line 61 in FIG. 7A). As an example, if the linearly combined value reaches the threshold, it is determined that "there is a disease" and if it does not reach the threshold, it is determined that "there is no disease." In FIG. 7A, the linear classifier 57a is based on a model represented by energy features and an autocorrelation coefficient integral value. When the features and the autocorrelation coefficient integral value are taken as coordinate values, a linear (straight-line) boundary line 61 is defined there to classify the data. As an example, it is possible to determine that "disease is present" when the right side of the boundary line is reached, and "disease is not present" when the left side of the boundary line is reached. Although FIG. 7A is shown as a two-dimensional graph for ease of understanding, in reality it is an N-dimensional feature vector space. However, even in N dimensions, the boundary line for discrimination is linear (straight-line). As an example of the linear classifier 57a, a logistic regression analysis can be used, but is not limited to this. As other examples of the linear classifier, a perceptron, a linear support vector machine, a latent Dirichlet Allocation (LDA) classification, or other conventionally known linear classifiers can be used.

[0031] The nonlinear classifier 57b discriminates diseases based on the information entropy and the feature of energy. In the nonlinear classifier 57b of the first embodiment, the distribution of test data is basically examined based on the (feature of) energy and the information entropy. In the knn method used here, classification is performed into disease and disease-free based on the length of each cluster center distance. The nonlinear classifier is also extended to N-dimensional information instead of 2-dimensional information. The nonlinear classifier 57b of the first embodiment can use the knn method as the nonlinear classifier 57b, but is not limited to this. As another example of the nonlinear classifier, a conventionally known nonlinear classifier such as a neural network, a support vector machine (SVM), or a random forest classification can be used.

[0032] In the classifier 57 of Example 1, if both the linear classifier 57a and the nonlinear classifier 57b determine that there is a disease, the classifier 57 determines that there is a disease as "disease present (knee OA)", if either the linear classifier 57a or the nonlinear classifier 57b determines that there is a disease, the classifier 57 determines that there is a disease as "possible disease (suspected knee OA)", and if both the linear classifier 57a and the nonlinear classifier 57b determine that there is no disease, the classifier 57 determines that there is no disease (healthy).

[0033] (Function of Example 1) In the osteoarthritis diagnosis system S of the first embodiment having the above-mentioned configuration, the knee joint sound s0 and joint angle θ0 of the subject are measured. The noise of the measured joint sound signal and joint angle signal are each removed by the noise removal means 53. Therefore, the accuracy of osteoarthritis diagnosis is improved compared to the case where the noise is not removed. In particular, the noise removal means 53a for the angle signal separates the signal into high-frequency components and low-frequency components and filters each of them, so that not only noise due to body movement but also high-frequency noise such as disturbances such as high-frequency vibration of the building in which the subject is located can be removed. In addition, the noise removal means 53b for the joint sound signal also removes fluctuation noise such as baseline drift by a band-pass filter, and can also remove high-frequency noise due to vibration.

[0034] In addition, in the first embodiment, not only the autocorrelation coefficient and the energy feature but also the information entropy is used as the feature for classification and discrimination. Therefore, it is possible to derive, as the information entropy, the sound and vibration in the frequency band (energy) that is unlikely to be generated in a normal subject and is likely to be generated in a disease, and it is possible to discriminate the disease from the measured sound (vibration) with higher accuracy than when the information entropy is not used. In particular, in the information entropy of the first embodiment, the spectrum is compressed in the direction of spectrum compression 1 in which frequency band information remains, rather than in the direction of spectrum compression 2 in which frequency band features do not remain. In the direction of spectrum compression 2, it is possible to extract time-related features, but it is not possible to extract frequency band features, and the problem of energy spectrum leakage occurs. Note that "energy spectrum leakage" refers to the phenomenon in which, when considering a spectrum (signal energy strength) in frequency analysis, energy (signal strength) that should be in the spectrum of interest is transmitted or transferred as if it is leaking into the surrounding spectra. In contrast to this, in the first embodiment, the spectrum is compressed in the direction of spectrum compression 1, so that the energy leakage is minimized and the diagnostic accuracy can be improved.

[0035] Furthermore, in the first embodiment, the information entropy is discriminated using the nonlinear classifier 57b, and the accuracy of diagnosis can be improved compared to the case where the linear classifier 57a, which has a lower discrimination accuracy than the nonlinear classifier, is used. Moreover, in the first embodiment, not only the nonlinear classifier 57b but also the linear classifier 57a is used. The nonlinear classifier 57b generally tends to require a longer calculation time than the linear classifier 57a, and there is a problem in that the calculation time is longer when only the nonlinear classifier 57b is used. However, by using both the nonlinear classifier 57b and the linear classifier 57a as in the first embodiment, it is possible to suppress the calculation time from becoming excessively long while suppressing the deterioration of the discrimination accuracy.

[0036] (Experimental Example) Next, an experiment was carried out to confirm the effect of the first embodiment. In the experiment, the "younger group" was assumed to be "disease-free" and the "elderly group" to be "disease-present." The number of knees measured in the younger group was 92, and the number of knees measured in the elderly group was 44. The average age of the younger group was 25 years old, and the average age of the elderly group was 86 years old. The average weight of the younger group was 55 kg, and the average weight of the elderly group was 63 kg. An Intel Core i7 7700k was used as the CPU for the information processing device. In the experimental example, as in the first embodiment, a linear classifier 57a and a non-linear classifier 57b were used using information entropy. As a comparative example, a linear classifier 57a and a non-linear classifier 57b were used to make a judgment without using information entropy.

[0037] In the comparative example, the accuracy rate of the linear classifier was 85%, and the accuracy rate of the nonlinear classifier was 89%. In contrast, in the experimental example, when the linear classifier 57a and the nonlinear classifier 57b were used to perform the judgment using the information entropy, the accuracy rate of the linear classifier was 89%, and the accuracy rate of the nonlinear classifier was 91%. The processing time of the linear classifier 57a was 0.16 seconds, and the processing time of the nonlinear classifier 57b was 14.00 seconds.

[0038] Furthermore, in the first embodiment, the sound measuring member 1 is installed in two channels. In the case of one channel, depending on the condition of the bones and muscles, sound and vibration may be attenuated during the propagation and may be difficult to measure. However, by using two channels, even if one cannot measure, the other can be easily measured. Therefore, in the first embodiment, poor measurement of joint sounds and misdiagnosis due to weak measured sounds are suppressed. In addition, by installing the sound measuring member 1 on each of the left and right knees, it is possible to measure two places simultaneously, and it is possible to prevent the subject from having to measure multiple times by switching the sound measuring member 1 to each knee, thereby reducing the burden on the subject. In addition, in Example 1, an acceleration sensor is used as the angle measurement member 7, and compared to using an angle sensor such as a potentiometer, it is possible to obtain not only the timing of the movement from the change in acceleration, but also information on noise components such as the movement (swaying) of the entire body and low-frequency vibrations, thereby improving the diagnostic accuracy.

[0039] (Example of change) Although the embodiments of the present invention have been described above in detail, the present invention is not limited to the above embodiments, and various modifications can be made within the scope of the gist of the present invention described in the claims. Modifications (H01) to (H04) of the present invention are illustrated below. (H01) In the above embodiment, the tibia was used as an example of the position where the joint sound s0 is observed, but this is not limited, and any position where the joint sound s0 of the knee joint can be observed is possible. In this case, it is preferable that the observation position is a part that is less affected by skin and muscles. Therefore, although the inner or outer part can be used, it is more preferable to use the tibia or patella, which are closer to the bone and can be observed, than the inner or outer part, which are more likely to be affected by muscles.

[0040] (H02) In the above embodiment, the exemplified values ​​are not limited to the specific numerical values ​​exemplified, and can be changed to appropriate numerical values ​​observed in an experiment, etc., depending on the sensitivity of a measuring device such as a vibration sensor, differences in experimental conditions (for example, changing from standing up from a chair to going up and down stairs), etc. Also, instead of using the specific numerical values ​​exemplified as threshold values, it is also possible to use something like a function or a distribution curve as a threshold value.

[0041] (H03) In the above embodiment, three judgment results were provided: healthy, knee OA, and suspected knee OA. However, this is not limited to this. It is also possible to eliminate knee OA and provide only two judgment results: healthy and knee OA, or to add other judgment results (e.g., suspected positive, suspected negative). (H04) In the above embodiment, the present invention is described as being used to distinguish diseases of the knee joint, but is not limited thereto. It can also be applied to other joints of the human body, such as the elbow, shoulder, and waist. [Explanation of symbols]

[0042] 1...Sound measuring member, 7...Angle measuring member, 54...Frequency analysis means, 55...Means for deriving autocorrelation coefficients, 56... Information entropy calculation means, 57...classifier, 57a...linear classifier, 57b...Nonlinear classifiers, S... Osteoarthritis Diagnostic System.

Claims

[Claim 1] a sound measuring member that is attached to a subject and measures sounds from the joints; an angle measuring member that is attached to the subject and measures a joint angle; a frequency analysis means for performing a frequency analysis of a sound signal measured by the sound measuring member for each joint angle measured by the angle measuring member; an autocorrelation coefficient deriving means for calculating an autocorrelation coefficient, which is a similarity of a signal waveform in a time domain, based on the signal waveform of the sound; an information entropy calculation means for calculating an information entropy for each of the joint angles based on a distribution of energy with respect to the frequency; a non-linear classifier that discriminates a disease of the joint based on the information entropy and the energy, and a linear classifier that discriminates a disease of the joint based on the energy and the autocorrelation coefficient, and a classifier that discriminates a disease of the joint based on a discrimination result of the non-linear classifier and a discrimination result of the linear classifier; An arthropathy diagnosis system comprising:

Images