JP7497863B2 - Diagnostic Equipment - Google Patents

Description

The present invention relates to diagnostic devices.

In the meat and seafood industries, there is a need to measure the fat, lean and bone content and their location in meat to determine meat quality and for automated meat butchering.

One example of a device for judging the quality of meat is one that irradiates light onto a cross section of a tuna tail and measures the fat content from the spectrum of reflected near-infrared wavelengths. Such a device makes it easy to measure the fat content of tuna without relying on experience or intuition. Patent document 1 also discloses a quality status measuring device that includes an impedance measuring unit that measures a voltage to obtain the impedance of a measurement object, including fish meat and meat, and a calculation unit that calculates a quality index value that indicates the quality of the measurement object based on the impedance.

As an example of equipment for butchering meat, Patent Document 2 discloses a left/right discrimination system for the arm parts of a pig carcass that uses X-rays to determine the coordinates of the bone surface, which are necessary for scoring and deboning.

JP 2018-159591 A Patent No. 5788108

However, the technology in Patent Document 1 measures the fat percentage of specific parts of fish and meat, so there is a problem in that the overall fat percentage cannot be determined. In addition, the technology in Patent Document 1 has a problem in that it cannot determine the location and amount of components such as bones. The technology in Patent Document 2 uses X-rays, so the equipment is expensive, and in addition, since X-rays are irradiated onto food, there are safety concerns from general consumers.

The present invention has been made in consideration of the above problems, and has an object to provide a diagnostic device capable of diagnosing the constituents of a subject to be measured at low cost and in a short time.

In order to solve the above problems, the present invention proposes the following means.
<1> A diagnostic device according to one aspect of the present invention is a diagnostic device using an electrical impedance tomography sensor, comprising four or more flexible electrodes, the electrodes being arranged around a subject to be measured and displaced along a surface shape of the subject to be measured, and a displacement meter for measuring the amount of displacement of the electrodes, the electrodes being displaced in a direction connecting the origin and the electrodes when the center of a shape boundary formed by the electrodes before the subject to be measured is taken as the origin, the diagnostic device measures the amount of displacement of the electrodes after the subject to be measured is displaced along the surface shape of the subject to be measured after the subject to be measured is placed on the electrical impedance tomography sensor, and the flexible electrodes are arranged to measure the amount of displacement of the electrodes. the contour estimation unit estimates the contour of the object to be measured from the displacement of the electrode of each of the flexible electrodes measured by the displacement amount and voltage/current measurement unit; and a Jacobian matrix prediction unit predicts the Jacobian matrix of the object to be measured using machine learning based on input variables consisting of the voltage measurement pattern or current measurement pattern measured by the displacement amount and voltage/current measurement unit, the coordinates of a mesh obtained by dividing the contour of the object to be measured estimated by the contour estimation unit, and the distance of each of the electrodes from the origin.
<2> The diagnostic device described above in <1> may further include a conductivity distribution calculation unit that calculates a conductivity distribution of the object under test from the Jacobian matrix of the object under test predicted by the Jacobian matrix prediction unit and the voltage or current measured by the displacement amount and voltage/current measurement unit.
<3> The diagnostic device described in <2> above may further include a constituent distribution calculation unit that obtains a constituent distribution of the object to be measured by classifying the conductivity distribution of the object to be measured obtained from the conductivity distribution calculation unit into clusters corresponding to the constituent components of the object to be measured.
<4> The diagnostic device according to <3> above may further include a boundary detection unit that detects a boundary line between each constituent component from the constituent distribution of the subject obtained by the constituent distribution calculation unit.

The above aspects of the present invention provide a diagnostic device capable of diagnosing the constituents of a measurement target at low cost and in a short time, and an electrical impedance tomography sensor for use in the diagnostic device.

FIG. 1 is a schematic diagram of a diagnostic device according to an embodiment of the present invention. FIG. 2 is a schematic diagram of electrical impedance tomography before a measurement target is placed; FIG. 1 is a schematic diagram of electrical impedance tomography after a measurement target is placed. FIG. 13 is a diagram for explaining a flexible electrode. 11A and 11B are diagrams for explaining a flexible electrode before an object to be measured is placed thereon; 11A and 11B are diagrams for explaining the flexible electrode after a measurement target is placed thereon; 11A and 11B are diagrams for explaining a method of applying a current to a flexible electrode. FIG. 13 is a diagram for explaining a contour ∂Ω of a measurement object obtained by a Lame curve. FIG. 1 illustrates a data set of input variables for a known contour ∂Ωg. 13 is a flowchart of Jacobian matrix prediction in a Jacobian matrix prediction unit. FIG. 1 is a diagram for explaining a method for predicting a Jacobian matrix J ^* using the k-nearest neighbor method. FIG. 13 is a diagram showing the conductivity distribution of the object to be measured before and after classification into clusters corresponding to the constituent components of the object to be measured. 10 is a flowchart illustrating a flow of classifying the conductivity distribution of a measurement target into clusters corresponding to the components constituting the measurement target. FIG. FIG. 1 shows a comparison of predicted and actual contours. FIG. 13 shows the predicted Jacobian matrix of a measurement sample. FIG. 13 is a diagram showing the calculation time of the Jacobian matrix of a measurement sample. FIG. 13 is a diagram showing the conductivity distribution of a measurement sample. FIG. 13 is a diagram showing a comparison between the component distribution of a measurement sample and the component distribution obtained by the k-means method. FIG. 1 is a diagram showing a comparison between the true values of the component ratios of the measured sample and the component ratios obtained by the k-means method. FIG. 13 is a diagram showing boundary lines detected by the Canny method.

(Diagnostic device)
Hereinafter, a diagnosis device 100 according to an embodiment of the present invention will be described with reference to the drawings. As shown in Fig. 1, the diagnosis device 100 includes a displacement and voltage/current measuring unit 1, a contour estimating unit 2, a Jacobian matrix predicting unit 3, a conductivity distribution calculating unit 4, a constituent component distribution calculating unit 5, a boundary detecting unit 6, and an output unit 7.

The diagnostic device 100 includes a central processing unit (CPU), a read only memory (ROM), a random access memory (RAM), and a hard disk drive (HDD)/solid state drive (SSD). The contour estimation unit 2, the Jacobian matrix prediction unit 3, the conductivity distribution calculation unit 4, the constituent component distribution calculation unit 5, the boundary detection unit 6, and the output unit 7 are realized by the CPU executing a predetermined program. The program may be acquired via a recording medium or via a network. A dedicated hardware configuration may also be used to realize the configuration of the diagnostic device 100. Each unit is described below.

(Displacement and voltage/current measuring section)
The displacement and voltage/current measuring unit 1 will be described with reference to Figures 2A and 2B. The drawings used in the following description may show characteristic parts in an enlarged scale for the sake of convenience in order to make the characteristics easier to understand, and the dimensional ratios of each component may differ from the actual ones. The materials, dimensions, etc. exemplified in the following description are merely examples, and the present invention is not limited thereto, and may be modified as appropriate within the scope of the effects of the present invention.

First, let us define the directions. One direction on the top surface of meat stand E is the x-direction, and the direction along the top surface that is perpendicular to the x-direction is the y-direction. The z-direction is perpendicular to the top surface of meat stand E. The z-direction is perpendicular to the x- and y-directions. Below, the +z direction may be expressed as "up" and the -z direction as "down". Up and down do not necessarily coincide with the direction in which gravity is applied.

As shown in FIG. 2A, the displacement and voltage/current measuring unit 1 includes an electrical impedance tomography sensor 10 and a control measuring unit 30. The electrical impedance tomography sensor 10 includes Q flexible electrodes 20 and an outer frame 25 and an inner frame 26 to which the flexible electrodes 20 are fixed. Here, the number Q of the flexible electrodes 20 is 4 or more. By having 4 or more flexible electrodes 20, it is possible to measure the contour of the object to be measured and predict the Jacobian matrix. The flexible electrodes 20 are not particularly limited in their positions, but are preferably arranged evenly in the circumferential direction of the outer frame 25 and the inner frame 26. After placing the object to be measured on the electrical impedance tomography sensor 10, the displacement and voltage/current measuring unit 1 measures the displacement of the electrode 21 after it is displaced along the surface shape of the object to be measured, and applies a predetermined current or voltage between the flexible electrodes 20 to measure the voltage or current.

The object to be measured is placed inside the inner frame 26 of the electrical impedance tomography sensor 10. Here, the object to be measured is, for example, meat and fish. ∂Ω ⁰ in FIG. 2A means a shape boundary formed by the Q electrodes 21. The origin O is the center of the shape boundary ∂Ω ⁰ formed by the Q electrodes before the object to be measured is placed. The shape of the shape boundary ∂Ω ⁰ is not particularly limited, and may be, for example, a circle or a rectangle. When the object to be measured is placed inside the inner frame 26, as shown in FIG. 2B, the flexible electrode 20 comes into contact with the object to be measured and is displaced along the surface shape of the object to be measured. The surface shape (contour) ∂Ω of the object to be measured is measured according to the position of the electrode 21 after the displacement.

As shown in FIG. 3A, the flexible electrode 20 includes a housing 15, an electrode 21, a support rod 22, a coil spring 23, and a displacement meter 24. The housing 15 of the flexible electrode 20 is fixed to an outer frame 25 and an inner frame 26, and is electrically connected to the control measurement unit 30 by a conductor or the like. The electrode 21 of the flexible electrode 20 displaces along the surface shape of the object to be measured, and the amount of displacement can be measured. The electrode 21 of the flexible electrode 20 displaces in the direction (radial distance) connecting the origin O and the electrode 21. In addition, the flexible electrode 20 can measure the voltage or current of the object to be measured by applying a current or voltage from the electrode 21.

The housing 15 is a hollow cylinder that is open on one side. One end of the coil spring 23 is attached to a closed lid inside the housing 15. One end of the support rod 22 is attached to the other end of the coil spring 23. An electrode 21 is attached to the other end of the support rod 22. The displacement meter 24 measures the spring length of the coil spring 23. Each part will be described below.

The coil spring 23 expands and contracts according to the load transmitted from the support rod 22. It is preferable that the coil spring 23 has a spring constant that does not change the surface shape of the object to be measured.

The electrode 21 is electrically connected to the control and measurement unit 30. There are no particular limitations on the material or shape of the electrode 21 as long as it can apply a current or voltage. An example of the material of the electrode 21 is gold-plated Cu.

The support rod 22 is connected to the electrode 21 and the coil spring 23. The support rod 22 transmits the load to the coil spring 23 after the electrode 21 comes into contact with the surface of the object to be measured.

The displacement meter 24 is, for example, a potentiometer. The potentiometer is a device that measures the length Sq of the coil spring 23 from the change in electrical resistance value, and the measured electrical resistance value is transmitted to the control measurement unit 30 via a conductor.

As shown in Figures 3B and 3C, the length s _q of the coil spring 23 and the position (radial distance) r _q of the electrode are displaced before and after the placement of the object to be measured. The radial distance from the origin O to each electrode 21 before the placement of the object to be measured is expressed by the following formula (1). The radial distance from the origin O to each electrode after the placement of the object to be measured is expressed by the following formula (2). The difference in the radial movement distance of the electrode 21 before and after the placement of the object to be measured, Δr _q = rq - r _q ⁰ , is calculated by the following formula (3) using the displacement meter 24. Here, r _q indicates the radial distance of the electrode 21 after the displacement, r _q ⁰ indicates the radial distance of the electrode 21 before the displacement, s _q indicates the length of the coil spring 23 after the displacement, and s _q ⁰ indicates the length of the coil spring 23 before the displacement.

(Control and measurement section)
The control and measurement unit 30 includes, for example, a multiplexer for switching between application electrodes and measurement electrodes, and an impedance analyzer for performing impedance measurement. The control and measurement unit 30 executes a predetermined program in a CPU and controls the multiplexer and the impedance analyzer to perform displacement measurement and impedance measurement. The obtained measurement result of the displacement of the electrode 21 is sent to the contour estimation unit 2, and the impedance measurement result is sent to the Jacobian matrix prediction unit 3.

The control measurement unit 30 measures the amount of displacement at the point when the electrode 21 is displaced by the object being measured and the amount of displacement stops fluctuating. The obtained results are sent to the contour estimation unit 2.

The control measurement unit 30 applies a current between the electrodes 21 and measures the voltage. Alternatively, the control measurement unit 30 applies a voltage between the electrodes 21 and measures the current. When applying a current, there is no particular limitation as to which electrodes the current is applied between. Similarly, when applying a voltage, there is no particular limitation as to which electrodes the voltage is applied between. Below, a case where a current is applied and a voltage is measured will be described. Current application methods include, for example, the counter electrode method, the adjacent method, and the reference method. The number of measurements M differs depending on each current application method. Below, each current application method will be described. Note that the current application method described below can also be applied to the voltage application method.

The counter electrode method is a method of applying a current between a pair of opposing electrodes. For example, in the case of FIG. 4(a), a current is applied to opposing electrodes such as electrodes 1 and 9, and electrodes 2 and 10. In the case of FIG. 4(a), the number of electrodes Q is 16, so there are a total of eight patterns. Voltage is measured in electrode pairs such as the second and third electrodes, and the third and fourth electrodes, excluding the electrodes to which the current is applied, and measurements are performed from the electrode pair of the second and third electrodes to the electrode pair of the fifteenth and sixteenth electrodes, so there are 13 voltage measurement patterns for one current application pattern. Therefore, in this case, in the counter electrode method, the number of measurements (measurement patterns) M is 104 in total. Here, when a current is applied and a voltage is measured, the measurement pattern is a voltage measurement pattern. When a voltage is applied and a current is measured, the measurement pattern is a current measurement pattern.

The adjacent method is a method in which a current is applied between adjacent electrodes. For example, referring to FIG. 4(b), a current is applied between adjacent electrodes, such as between electrode 1 and electrode 2, and between electrode 2 and electrode 3. In the case of FIG. 4(b), the number of electrodes Q is 16, so there are 16 combinations in total. Voltage is measured in electrode pairs, such as the third and fourth electrodes excluding the electrodes to which current is applied, and measurements are made from the third and fourth electrodes to the fifteenth and sixteenth electrodes, so there are 13 voltage measurement patterns for one current application pattern. Therefore, with the adjacent method, the number of measurements (measurement patterns) M is 208 in total.

The reference method is a method of measuring voltages in all combinations between a reference electrode and an electrode other than the reference electrode. For example, in the case of FIG. 4(c), a current is applied between the reference electrode and an electrode other than the reference electrode, such as between the first and second electrodes, or between the first and third electrodes. In the case of FIG. 4(c), the number of electrodes Q is 16, so there are 16 combinations in total. Voltage is measured in electrode pairs such as the third and fourth electrodes excluding the electrode to which the current is applied, and measurements are made from the third and fourth electrodes to the fifteenth and sixteenth electrodes, so there are 13 voltage measurement patterns for one current application pattern. Therefore, in the reference method, the number of measurements (measurement patterns) M is 208 in total.

Below, we will explain an example of measuring voltage using the adjacent method with the diagnostic device 100 of this embodiment.

(Contour Estimation Unit 2)
The contour estimation unit 2 estimates the surface shape of the object to be measured from the displacement of the electrode 21 of the flexible electrode 20 sent from the displacement amount and voltage current measurement unit 1. The displacement amount data (coordinate data) sent from the displacement amount and voltage current measurement unit 1 is only the same number as the number of electrodes Q, and in order to obtain the contour ∂Ω of the object to be measured, it is necessary to supplement the intervals (between each coordinate point). The contour estimation unit 2 estimates the contour ∂Ω of the object to be measured from the displacement amount data (position coordinates of the electrode 21) and the Lame curve. Here, the contour ∂Ω of the object to be measured is estimated using a Lame curve, but a spline curve or a Bezier curve may also be used. FIG. 5 shows the contour ∂Ω of the object to be measured estimated using a Lame curve. As shown in FIG. 5, the position of each electrode 21 is expressed by the radial length r and the angle θ between the line connecting the center of the electrode 21 and the origin O and the x-axis. The coordinates (x, y) of the contour ∂Ω are expressed by the following equation (4). The obtained coordinate information of the contour ∂Ω of the object to be measured is sent to the Jacobian matrix prediction unit 3 .

(Jacobian matrix prediction unit 3)
The Jacobian matrix prediction unit 3 predicts the Jacobian matrix of the object to be measured by machine learning such as a nearest neighbor search method or a neural network using the displacement amount and measurement pattern of the electrode 21 sent from the displacement amount and voltage current measurement unit 1, and the contour ∂Ω of the object to be measured estimated by the contour estimation unit 2. Specifically, the Jacobian matrix prediction unit 3 predicts the Jacobian matrix of the object to be measured by, for example, a nearest neighbor search method, using the displacement amount and measurement pattern, and the contour ∂Ω of the object to be measured estimated by the contour estimation unit 2, based on input variables consisting of the measurement pattern measured by the displacement amount and voltage current measurement unit 1, the coordinates of the mesh obtained by dividing the contour ∂Ω of the object to be measured estimated by the contour estimation unit 2, and the distance from the origin O of the electrode 21 of each flexible electrode 20. The Jacobian matrix is a sensitivity matrix that represents the sensitivity of the conductivity vector (conductivity distribution) when the measurement voltage vector changes. Therefore, if the Jacobian matrix (sensitivity matrix) is known, the conductivity distribution can be calculated. The nearest neighbor search method is not particularly limited, but includes k-nearest neighbor search, approximate nearest neighbor search, locality sensitive hashing, kd tree, etc. Here, the k-nearest neighbor search will be taken as an example for explanation.

Next, we will explain a method of predicting the Jacobian matrix using the k-nearest neighbor method based on the dataset of the Jacobian matrix of a known contour, the input dataset of the known contour, the contour ∂Ω of the object to be measured estimated by the contour estimation unit 2, and the measurement pattern m (1<m<M). Here, the dataset of the Jacobian matrix of a known contour refers to data on multiple Jacobian matrices obtained by the finite element method or the like from the contours of multiple known samples. The dataset of a known contour refers to data on multiple input variables obtained from the contours of multiple known samples.

The data set of the Jacobian matrix of the known contour may be created by, for example, calculating the Jacobian matrix using the finite element method. Since the Jacobian matrix varies depending on the contour ∂Ω of the object to be measured, it is preferable to prepare a large number of known contours and Jacobian matrices. Here, the known contour is ∂Ω ^{g (} 1≦g≦G), where G is the number of known contours in the data set, and is, for example, 100 to 10,000. The type of ∂Ω ^g and the number of G can be appropriately modified depending on the object to be measured.

The known contour ^∂Ωg is divided so as to obtain an appropriate resolution according to the number of electrodes 21. For example, when there are 16 electrodes 21, the region including the known contour ^∂Ωg may be divided vertically into 64 and horizontally into 64 to create mesh n (1≦n≦N). In the case of 64×64, N is 4096. The number and shape of the meshes can be set appropriately according to the number of electrodes 21 and the required resolution.

The Jacobian matrix J ^g of the known contour Ω ^g is expressed by the following formula (5). In formula (5), M indicates the number of measurements (measurement patterns), and N indicates the number of meshes. The Jacobian matrix J _mn in the measurement pattern m and mesh n is calculated using the following formula (6). Here, σ _n indicates the conductivity in mesh n. A _n indicates the nth mesh area. Vm(e, d) indicates the measured voltage V in the measurement pattern m. e indicates the current application electrode pair in the measurement pattern m, and d means the voltage measurement electrode pair in the measurement pattern m. V(i ^e ) indicates the potential between the voltage measurement electrode pair d induced by the application of current to the current application electrode pair e. V(i ^d ) is the potential between the current application electrode pair e induced by the application of current to the voltage measurement electrode pair d.

Next, the input data set of the known contour ∂Ωg will be described with reference to FIG. 6. The input data set I is expressed by the following formula (7). I ^g is an input variable in the known contour ^∂Ωg , and is expressed by the following formula (8). I ^g _mn in formula (8) is an input variable of the measurement pattern m and mesh n in the known contour ^∂Ωg , and is expressed by the following formula (9). m in formula (9) indicates the measurement pattern. Furthermore, X ^g _n in formula (9) indicates the Cartesian coordinate of mesh n in the known contour ^∂Ωg , and is expressed by the following formula (10). r ^g in formula (9) is the distance from the origin of the electrode 21 arranged in the known contour ^∂Ωg , and is expressed by the following formula (11). Q in formula (11) indicates the number of electrodes 21.

Next, a method of predicting a Jacobian matrix using the k-nearest neighbor method will be described with reference to FIG. 7. The Jacobian matrix prediction unit 3 divides the contour ∂Ω of the object to be measured sent from the contour estimation unit 2 so that the number of meshes is the same as the data set of the known contour ∂Ω ^g (S11). Then, an input variable I ^* is created from the displacement amount and measurement pattern m sent from the displacement amount and voltage current measurement unit 1 and mesh n (S12). The input variable I ^* is an input variable in the contour ∂Ω of the object to be measured, and is expressed by the following formula (12). I ^g _mn in formula (12) is an input variable of the measurement pattern m and mesh n in the contour ∂Ω of the object to be measured, and is expressed by the following formula (13). m in formula (13) indicates the measurement pattern. Also, X ^g _n in formula (13) indicates the Cartesian coordinate of mesh n in the contour ∂Ω of the object to be measured, and is expressed by the following formula (14). In formula (13), r ^* is the distance from the origin of the electrode 21 arranged on the contour ∂Ω of the object to be measured, and is expressed by the following formula (15): Q in formula (15) represents the number of electrodes 21.

Next, the flow of prediction of the Jacobian matrix J ^* of the object to be measured will be described. The Jacobian matrix prediction unit 3 inputs an initial value (for example, n=1, m=1) (S13). Next, the Jacobian matrix prediction unit 3 predicts the Jacobian matrix J ^* _mn at the contour ∂Ω of the object to be measured based on k Jacobian matrices ^Jgmn with small Euclidean distance between I ^* _and ^Ig according to the following formula (16) (S14). Here, d(I ^* _{mn ,} ^Igmn ) is calculated by the following formula (17) and indicates the Euclidean distance between ^Igmn and I ^* _mn . ^Igmn indicates the input at the measurement pattern m and mesh n at the known contour ^Ωg . _I ^* _mn is an input variable measured by the electrical impedance tomography sensor 10 at the measurement pattern m and mesh n sent from the displacement amount and voltage _current measurement unit 1. C _mn is expressed by the following formula (18) and indicates k input variables I ^g _mn having a small Euclidean distance from I ^* _mn . The number k is not particularly limited and is 10, for example.
The prediction of the Jacobian matrix J ^* using the k-nearest neighbor method will be explained with reference to Fig. 8. In this example, the Jacobian matrix J*1,8 _is predicted from the Jacobian matrix ^J11,9 of the known contour ^Ω1 _, the Jacobian matrix ^J81,12 of the known contour ^Ω8 , and the Jacobian matrix ^J151,10 of the known contour ^Ω15 , which have small _Euclidean distances with the input _variables of the ^object to be measured.

After the prediction of J _mn ^* is completed, the Jacobian matrix prediction unit 3 judges whether the number of n is equal to the number of meshes N (S15). If n and N are not equal, the number of n is incremented by one, and the process returns to S14 again (S16). If n and N are equal, the process next judges whether m is equal to the number of measurement patterns M (S17). If m and M are not equal, the number of m is incremented by one, and the process returns to S14 again (S18). If n and m are equal, the Jacobian matrix prediction unit 3 ends the prediction of the Jacobian matrix and sends the predicted Jacobian matrix J ^* of the object to be measured and the measurement pattern to the conductivity distribution calculation unit 4. Normally, when the Jacobian matrix is calculated using the above formula (6), it takes more than 10 minutes to calculate. In the diagnosis device 100 according to this embodiment, a data set of the Jacobian matrix of a known contour ∂Ω ^g and an input data set are prepared, and the Jacobian matrix J * of the ^object to be measured can be predicted with high accuracy in a short time by using machine learning such as the k-nearest neighbor method based on the contour and measurement pattern of the object to be measured.

(Conductivity distribution calculation part)
The conductivity distribution calculation unit 4 calculates the conductivity distribution of the object to be measured from the predicted Jacobian matrix J* of the object to be measured sent from the Jacobian matrix prediction unit 3 and the displacement and the voltage measured by the voltage/current measurement unit 1. The initial conductivity distribution is calculated from the following formula (19 ^{) using the Jacobian matrix J* of} the object to be measured. ΔV in formula (19) is expressed by the following formula (20). ΔVm in formula (20) is a voltage normalized by the voltage V(f ₀ ) at low frequency and the voltage V(f ₁ ) at high frequency, and is expressed by the following formula (21). Here, the low frequency is, for example, 500 Hz, and the high frequency is, for example, 1 kHz. m in formula (21) is a measurement pattern.
The conductivity distribution of the object to be measured is calculated from the following formula (22A). In formula (22A), i represents the number of iterations of the repeated calculation. In formula (22A), R represents a regularization matrix, and λ represents a relaxation factor scalar. R is expressed, for example, by the following formula (22B). λ is an arbitrary hyperparameter, for example, 0.01. The calculated conductivity distribution of the object to be measured is sent to the constituent distribution calculation unit 5.

(Component Distribution Calculation Unit)
The constituent distribution calculation unit 5 obtains the constituent distribution of the object to be measured by classifying the conductivity distribution of the object to be measured obtained from the conductivity distribution calculation unit 4 into clusters corresponding to the constituent components of the object to be measured. There is no particular limitation on the method of classifying the conductivity distribution of the object to be measured into clusters corresponding to the constituent components of the object to be measured, but examples of the method include the k-means method, the EM algorithm, the ISODATA method, Fuzzy clustering, and the longest distance method. Here, an example of the k-means method will be described.

Fig. 9(a) shows the conductivity distribution of 25x20 mesh before classification into clusters corresponding to the components of the object to be measured. The flow of classification into clusters corresponding to the components of the object to be measured will be explained below with reference to Fig. 10. The conductivity distribution (Fig. 9(a)) sent from the conductivity distribution calculation unit 4 is divided into three types of clusters (fat ^{k=1σ0 ,} red meat ^{k= 2σ0} , bone ^{k=3σ0 )} and the clusters are initialized (S21). Next, the center of gravity position ^krc of component k is calculated from the coordinates ^krn =( _{xn , yn} ) of the mesh n _clustered to the conductivity ^kσn using the following formula (23) (S22). After calculating the center _of gravity ^krc for each cluster, the conductivity _σn in the mesh n is assigned the conductivity ^kσn of the component k that is the same as the center of gravity ^krc that has the smallest Euclidean distance from the coordinate _rn of the mesh _n calculated by the following formula (24) (S23). Then, ^the center of gravity _{krc is} calculated again for each cluster (S24). After calculating the center of gravity of each cluster, if the latest center of gravity of each cluster does not match the center of gravity of each cluster in the calculation one time before, the process returns to S23. If the latest center of gravity of each cluster matches the center of gravity of each cluster in the calculation one time before, the calculation is terminated. The component distribution calculation unit 5 sends the obtained component distribution of the measured object to the boundary detection unit 6. The component distribution of the measured object after classification into clusters corresponding to the components of the measured object is as shown in FIG. 9(b). By separating the constituent components using a method such as k-means that classifies the conductivity distribution of the object to be measured into clusters corresponding to the constituent components of the object to be measured, the constituent components can be classified with higher accuracy than by using a conductivity threshold value.

(Boundary detection section)
The boundary detection unit 6 detects the boundary of each component from the distribution of components of the measured object sent from the component distribution calculation unit 5. The method of detecting the boundary is not particularly limited, and examples thereof include the Canny method, the Sobel method, the Gaussian Laplacian method, etc. Here, an example of the Canny method will be described.

The Canny method detects boundaries from the gradient |Gn| of conductivity in adjacent meshes and its gradient direction θn. If there is a sudden change in conductivity between meshes, |Gn| becomes large. Therefore, a large |Gn| indicates a boundary. The gradient direction θn indicates the direction of the boundary. The boundary is detected by the following formulas (25) to (29). Here, Gn(x) indicates the gradient in the x direction of mesh n. Gn(y) indicates the gradient in the y direction of mesh n. Sx indicates a Sobel filter in the x direction. Sy means a Sobel filter in the y direction. By detecting the boundary using the boundary detection unit 6, it is possible to clarify the boundaries of each component after classifying the conductivity distribution of the object to be measured into clusters corresponding to the components of the object to be measured, and the positional relationship of each component can be made clearer. The boundary detection unit 6 sends the component distribution of the object to be measured after detecting the boundary to the output unit 7.

(Output section)
The output unit 7 outputs the component distribution of the object to be measured after the boundary line has been detected, sent from the boundary line detection unit 6. There is no particular limitation on the destination to which the component distribution of the object to be measured is output. The output destination may be a display unit such as a liquid crystal display, or a storage device such as a HDD.

The above is a detailed explanation of the diagnostic device 100 according to this embodiment. Note that the technical scope of the present invention is not limited to the above embodiment, and various modifications can be made without departing from the spirit of the present invention. For example, an air spring may be used instead of a coil spring.

In addition, the components in the above embodiment may be replaced with known components as appropriate without departing from the spirit of the present invention, and the above-mentioned modifications may be combined as appropriate.

(Example)
Next, an example of an experiment conducted to verify the effectiveness of the diagnostic device 100 according to this embodiment will be described.

(Measurement sample)
Two types of agar phantoms were prepared. These agar phantoms were created by adjusting the electrical conductivity of the fat portion to 0.024 S/m, the electrical conductivity of the lean portion to 0.351 S/m, and the electrical conductivity of the bone portion to 0.083 S/m. The electrical conductivity of each component was adjusted using distilled water and salt. Agar powder (Fisher Scientific A360) was mixed, the agar was melted at 85°C, and molded into the shape shown in Figure 11.
In addition, beef and three types of mutton were purchased from Newzee Co., Ltd. The appearance of each type of beef and mutton is as shown in FIG.

(Voltage measurement)
The voltage of the above measurement sample was measured using an impedance analyzer (IM 3570, manufactured by Hioki E.E. Corporation), a multiplexer (consisting of Arduino (manufactured by LLS, Italy) and CD74HC4067 (Texas Instrument, USA, Analog Mux/Demux)), an electrical impedance tomography sensor, and a personal computer. The number of electrodes was 16, and the electrodes were combined based on the adjacent method to apply a current and measure the voltage. There were 208 measurement patterns m. The impedance analyzer swept the complex impedance frequency from 500 Hz to 1 kHz. At the same time, the displacement of each electrode was also measured.

(Estimation of the contour of the object to be measured)
From the obtained displacement of each electrode, the Lamé curve was used to estimate the contour of the measurement sample. The obtained results are shown in FIG. 12. The dotted line in FIG. 12 indicates the true value, and the solid line indicates the predicted value. The true value of the contour was obtained from the image in FIG. 11. From these results, the error was calculated using the following formula (30). The error of agar phantom 1 was 4.81%, the error of agar phantom 2 was 3.38%, the error of beef was 10.47%, the error of mutton 1 was 4.81%, the error of mutton 2 was 3.19%, and the error of mutton 3 was 5.35%. Therefore, the average error of each measurement sample was 4.81%. This error value is small. Therefore, it was found that the contour estimated by the Lamé curve matches the actual contour with high accuracy.

(Jacobian matrix prediction)
Using the k-nearest neighbor method described above, the Jacobian matrices were predicted from the estimated contours and the measured patterns. The known Jacobian matrices were calculated using the finite element method, and a data set of G=200 was prepared. The predicted Jacobian matrices are shown in FIG. 13. Next, these predicted Jacobian matrices J ^* were compared with the Jacobian matrices J calculated using the above formula (6), which is a conventional calculation method. Specifically, the average value <CC> of the cross-correlation functions of J ^* and J was calculated according to the following formula (31). <CC> ranges from -1 to 1, and indicates positive correlation closer to 1, negative correlation closer to -1, and no correlation closer to 0. The <CC> of agar phantom 1, <CC> of agar phantom 2, <CC> of beef, <CC> of lamb ¹ , <CC> of lamb 2, <CC> of lamb 3 ...

(Jacobian matrix prediction time)
Next, the time required to predict the Jacobian matrix was compared. In FIG. 14, the white circle indicates the time t(J ^* ) required to predict the Jacobian matrix J ^* using the k-nearest neighbor method, and the black circle indicates the time t(J) required to calculate using the above formula (6). In order to compare the calculation speed, the calculation speed ratio sp was calculated from the following formula (32), and the results are shown in a bar graph. The calculation time required by the conventional method was 51.58 s for agar phantom 1, 13.32 s for agar phantom 2, 14.99 s for beef, 27.69 s for mutton 1, 23.52 s for mutton 2, and 8.25 s for mutton 3. Therefore, the average calculation time required by the conventional method was 23.22 s. On the other hand, the time required for prediction using the k-nearest neighbor method was 8.07 s for agar phantom 1, 3.58 s for agar phantom 2, 3.82 s for beef, 5.28 s for mutton 1, 4.72 s for mutton 2, and 2.95 s for mutton 3. Therefore, the average time required for prediction using the k-nearest neighbor method was 4.73 s. In addition, the speed ratio sp was 6.39 for agar phantom 1, 3.72 for agar phantom 2, 3.93 for beef, 5.24 for mutton 1, 4.98 for mutton 2, and 2.79 for mutton 3. Therefore, the average speed ratio sp was 4.51. From the above, it was found that the calculation time using the k-nearest neighbor method can be significantly reduced compared to conventional calculation methods.

(Conductivity Distribution)
The conductivity distribution was calculated from the Jacobian matrix obtained by the k-nearest neighbor method using the above formula (22), and the diagram of the conductivity distribution is shown in Figure 15. As shown in Figure 15, it was found that there was consistency with the measurement sample in Figure 11.

(Constituent distribution)
The obtained conductivity distribution was used to create a component distribution of the measurement sample using the k-means method. The obtained component distribution is shown in FIG. 16(b). The true value of the component distribution obtained from the image of FIG. 11 by the image overlay method is shown in FIG. 16(a). The component ratio of each component obtained by the k-means method was ^k η, and the true value of the component obtained from FIG. 16(b) was ^k η ^true . The error was calculated from the following formula (33). ^k A ^true in formula 33 represents the area of the kth component, and A ^true represents the total area of the object to be measured. These results are shown in FIG. 17. FIG. 17(a) shows the proportion of fat, FIG. 17(b) shows the proportion of red meat, and FIG. 17(c) shows the proportion of bone. The black circles in each figure indicate the true value of the proportion of the component of the measurement sample, and the white circles indicate the proportion of the component of the measurement sample obtained by the k-means method. The error of agar phantom 1 in fat was 7.48%, the error of agar phantom 2 was 4.11%, the error of beef was 3.45%, the error of mutton 1 was 9.16%, the error of mutton 2 was 3.95%, and the error of mutton 3 was 9.82%. Therefore, the average error of each measurement sample in fat was 6.33%. In addition, the error of agar phantom 1 in red meat was 7.03%, the error of agar phantom 2 was 3.09%, the error of beef was 6.33%, the error of mutton 1 was 10.91%, the error of mutton 2 was 5.07%, and the error of mutton 3 was 11.53%. Therefore, the average error of each measurement sample in red meat was 7.03%. The error of agar phantom 1 for bone was 2.25%, the error of agar phantom 2 was 1.03, the error of beef was 9.78%, the error of mutton 1 was 1.75%, the error of mutton 2 was 1.12%, and the error of mutton 3 was 1.71%. Therefore, the average error of each measurement sample for bone was 2.94%. From the above, it was found that the component ratio of the measurement sample can be diagnosed with high accuracy by the k-means method.

(Borderline detection)
The above canny method was used to detect the boundary line of each component from the distribution of components obtained by the k-means method. The boundaries of each component are shown in FIG. 18. The error of the boundary line of the actual meat component was calculated based on the following formula (34). The true value of the boundary line was obtained from the image in FIG. 11. The average value e _e was calculated from the boundary line error ^k e _e of each cluster in each measurement sample obtained. The average value e _e of the boundary line error of each cluster of agar phantom 1 was 6.22%, the average value e _e of the boundary line error of each cluster of agar phantom 2 was 5.37%, the average value e _e of the boundary line error of each cluster of beef was 8.46%, the average value e _e of the boundary line error of each cluster of mutton 1 was 4.66%, the average value e _e of the boundary line error of each cluster of mutton 2 was 1.69%, and the average value e _e of the boundary line error of each cluster of mutton 3 was 9.01%. The average value of e _e of each measurement sample was 5.90%. Therefore, it was found that the Canny method can detect the boundary between each component with high accuracy.

As described above in detail, the diagnostic device 100 according to this embodiment can diagnose the components of the subject to be measured with high accuracy in a short time.

Reference Signs List 1 Displacement and voltage/current measuring unit 2 Contour estimating unit 3 Jacobian matrix predicting unit 4 Conductivity distribution calculating unit 5 Constituent distribution calculating unit 6 Boundary detecting unit 7 Output unit 10 Electrical impedance tomography sensor 15 Housing 20 Flexible electrode 21 Electrode 22 Support rod 23 Coil spring 24 Displacement meter 25 Outer frame 26 Inner frame 30 Control measuring unit 100 Diagnostic device

Claims (4)

an electrode disposed around the object to be measured and displaced along the surface shape of the object to be measured;
a displacement meter for measuring the amount of displacement of the electrode;
Four or more flexible electrodes comprising:
A diagnostic device using an electrical impedance tomography sensor , in which, when a center of a shape boundary formed by the electrodes before the measurement subject is placed is taken as an origin, the electrodes are displaced in a direction connecting the origin and the electrodes,
a displacement amount and voltage/current measuring unit that measures the amount of displacement of the electrodes after the object to be measured is displaced along the surface shape of the object to be measured after the object to be measured is placed on the electrical impedance tomography sensor, and applies a current between the flexible electrodes and measures a voltage, or applies a voltage between the flexible electrodes and measures a current;
a contour estimation unit that estimates a contour of the object to be measured from the displacement amount of each of the flexible electrodes measured by the displacement amount and a voltage/current measurement unit;
a Jacobian matrix prediction unit that predicts a Jacobian matrix of the object to be measured by machine learning based on input variables including the displacement amount and a voltage measurement pattern or a current measurement pattern measured by the voltage/current measurement unit, coordinates of a mesh obtained by dividing the contour of the object to be measured estimated by the contour estimation unit, and a distance of each of the electrodes from the origin;
A diagnostic device comprising: the Jacobian matrix of the device under test predicted by the Jacobian matrix prediction unit; and
2. The diagnostic device according to claim 1 , further comprising a conductivity distribution calculation unit that calculates a conductivity distribution of the object to be measured from the amount of displacement and the voltage or current measured by the voltage/current measurement unit . 3. The diagnostic device according to claim 2, further comprising a constituent distribution calculation unit that obtains a constituent distribution of the object to be measured by classifying the conductivity distribution of the object to be measured obtained from the conductivity distribution calculation unit into clusters corresponding to the constituent components of the object to be measured. 4. The diagnostic device according to claim 3 , further comprising a boundary detection unit that detects a boundary line between each constituent component from the constituent distribution of the object to be measured obtained by the constituent distribution calculation unit.

Images