In-vivo visualization device
The in-vivo visualization device uses a sensor array and advanced algorithms to achieve high-resolution, non-invasive imaging of body fluids and tissues, addressing the limitations of existing technologies by accurately identifying local changes and diagnosing diseases.
Patent Information
- Application Number
- JP2023085482
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-24
- Publication Date
- 2024-12-05
AI Technical Summary
Existing in-vivo visualization technologies, such as X-ray CT and MRI, are expensive and bulky, and conventional electrode-based devices like Patent Document 1 cannot provide high-resolution spatial and temporal measurements of body fluids and tissues, making it difficult to accurately evaluate disease states.
An in-vivo visualization device with a sensor array of electrodes that measures electrical properties, applies currents and potential differences, and uses image reconstruction and learning algorithms to create high-resolution electrical property distributions, identifying target areas based on sparsity, time correlation, and noise optimization.
The device provides non-invasive, high-resolution visualization of body fluids and tissues, enabling accurate diagnosis of diseases and health conditions by identifying local changes in time and space, replacing the need for expensive and bulky imaging modalities.
Smart Images

Figure 2024168639000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an in vivo visualization device. [Background technology]
[0002] In the fields of medicine, nursing care, and healthcare, diseases associated with aging and mental stress have become major social issues. In particular, edema, which often occurs together with geriatric diseases, and sarcopenia, a decrease in muscle mass that accompanies aging, cause significant social losses, and the importance of visualizing abnormalities in body fluids and tissues in the body to evaluate the state of these diseases is recognized in the fields of medicine, nursing care, and healthcare.
[0003] For the purpose of evaluating the state of disease, devices used to visualize abnormalities in body fluids and tissues include, for example, X-ray computed tomography (CT) and nuclear magnetic resonance imaging (MRI). However, X-ray CT and MRI are extremely expensive and large, and ensuring appropriate imaging conditions and taking safety measures is costly and time-consuming, making it difficult to visualize the inside of a living body in a simple manner.
[0004] As an example of a simple in vivo visualization device, Patent Document 1 discloses a lymphedema monitor device having an electrode band with a plurality of electrodes that are placed around a user's arm or leg, a measuring means for measuring electrical characteristics related to a relaxation frequency specific to albumin, and an information processing device having a lymphedema prediction means for detecting early stage lymphedema based on the electrical characteristics measured by the measuring means. [Prior art documents] [Patent documents]
[0005] [Patent Document 1] Patent No. 6555715 Summary of the Invention [Problem to be solved by the invention]
[0006] The measurement technique of Patent Document 1 can measure the amount and low-resolution distribution of bodily fluids and tissues contained in the cross section to be measured, but cannot measure the distribution at a high enough resolution to evaluate the state of a disease. In other words, the spatial distribution of bodily fluids and tissues in a living body is complex, and bodily fluids, etc., change significantly even in a short time regardless of whether they are normal or abnormal. Therefore, there is a demand for an in vivo visualization technique that can specify local changes in time and space in more detail than the measurement technique of Patent Document 1.
[0007] The present invention has been made in consideration of the above circumstances, and has an object to provide an in vivo visualization device that can easily and accurately identify local changes in time and space. [Means for solving the problem]
[0008] In order to solve the above problems, the present invention proposes the following means. <1> The in vivo visualization device according to the first aspect of the present invention comprises: a sensor including a plurality of electrodes positionable on the skin of a subject at a distance from one another; a current / voltage application measurement unit that applies a current or a potential difference between the electrodes while the electrodes are in contact with the skin, and measures first measurement data which is a potential difference and a phase based on a current application / voltage measurement pattern when the current is applied, and measures second measurement data which is a current and a phase based on a voltage application / current measurement pattern when the potential difference is applied between the electrodes; an image reconstruction unit that creates an electrical property distribution in the subject's body based on the first measurement data or the second measurement data and predetermined parameters; a target region identifying unit that performs an identification process for identifying a target region with respect to the electrical property distribution and creates a post-identification electrical property distribution; Equipped with. <2> A second aspect of the present invention is the in vivo visualization device of the first aspect, The classification process may be performed based on a predetermined threshold value. <3> A third aspect of the present invention is an in vivo visualization device according to the first or second aspect, The target area identification unit, A target area index distribution is generated by performing standard deviation processing on the electrical property distribution; The classification process may be performed on the target region index distribution to generate the classified electrical property distribution. <4> A fourth aspect of the present invention is an in vivo visualization device according to the first or second aspect, The image reconstruction unit, creating a multiple measurement matrix from at least one of impedance, resistance, reactance, capacitance, admittance, conductance, susceptance, and phase obtained from the first measurement data or the second measurement data; creating the electrical property distribution from the multiple measurement matrix and the predetermined parameters; The predetermined parameters may include a prior variance vector γ for sparsity, a positive definite matrix B for time correlation, and a variance value λ of a multidimensional normal distribution for noise. <5> A fifth aspect of the present invention is an in vivo visualization device according to the first aspect, The imaging device may further include a learning unit that updates the predetermined parameters based on at least the post-identification electrical property distribution. <6> A sixth aspect of the present invention is the in vivo visualization device of the fifth aspect, The learning unit may update the predetermined parameters by maximizing an average data likelihood using an expectation-maximization algorithm. Effect of the Invention
[0009] According to the above-mentioned aspects of the present invention, a non-invasive in vivo visualization device can be provided to replace X-ray CT devices and MRI devices. In addition, since it is possible to perform discrimination processing and use parameters related to optimized sparsity, time correlation, and noise, it is possible to provide an in vivo visualization device capable of identifying local changes in time and space, which was difficult with conventional technology. In addition, by evaluating the local state of body fluids and tissues in a living body, it is possible to diagnose various diseases and health conditions. [Brief description of the drawings]
[0010] [Figure 1] FIG. 1 is a schematic diagram of an internal skin measuring device according to an embodiment. [Diagram 2] FIG. 2 is a schematic diagram of a current / voltage application measurement unit in FIG. 1. [Figure 3A] FIG. 13 is a diagram for explaining a current application voltage measurement pattern using a counter electrode method. [Figure 3B] FIG. 13 is a diagram for explaining a current application voltage measurement pattern using the adjacent method. [Figure 3C] FIG. 13 is a diagram for explaining a current application voltage measurement pattern according to a reference method. [Figure 4] 1 is a flowchart of an in vivo visualization method according to the present embodiment. [Figure 5A] This is a simplified simulated image of a human calf. [Figure 5B] FIG. 1 shows lymphedema simulation conditions (condition 1). [Figure 5C] FIG. 13 is a diagram showing the venous edema simulation condition (condition 2). [Figure 6] 13 is a diagram showing the conductivity distribution Δσ obtained by image reconstruction in the embodiment, the comparative example, and the conventional example under condition 1. FIG. [Figure 7] 13 is a diagram showing the conductivity distribution Δσ obtained by image reconstruction in the embodiment, the comparative example, and the conventional example under condition 2. FIG. [Figure 8] FIG. 1 is a schematic diagram of a measurement device used in Experimental Example 2. [Figure 9] FIG. 1 illustrates the experimental protocol for prolonged standing and leg elevation to induce local spatiotemporal changes (LSTC) in extracellular fluid (ECF) in subcutaneous adipose tissue (SAT). [Figure 10] 1 shows the frequency difference conductivity distribution Δσ of a subject reconstructed by an in vivo visualization device. [Figure 11A] FIG. 1 is a cross-sectional view of a typical human calf. [Figure 11B] This is the Δσ of subject No. 12. [Figure 11C] This is an ultrasound image near the EIT electrode No. (4) of the subject. [Figure 12] FIG. 1 shows normalized spatially averaged conductivity <Δσ>SAT in isolated SAT and normalized conventional BIA impedance zBIA during prolonged standing and leg elevation. [Figure 13] FIG. 1 shows the correlation between the spatially averaged conductivity of isolated SAT, <Δσ>SAT, and conventional impedance zBIA for the calf of a subject. [Figure 14] FIG. 1 shows the time course of ΔLECF in the right leg of 15 subjects. [Figure 15] FIG. 13 shows the correlation between maximum volume change in ECFmax (ΔLECF) (ΔLECF at t8=40 min) and muscle mass in the right leg vRL as measured by conventional bioelectrical impedance analysis (conventional BIA) in 15 subjects. [Figure 16A] 1 is an example of a human calf including muscles, SAT, and GSV. [Figure 16B] FIG. 13 is a diagram showing conditions for the conductivity value σ for a simulation. [Figure 17A] Δσ at iter=15 for image reconstruction by each algorithm under an unstable background field. [Figure 17B] This is an ideal reconstructed image. [Figure 17C] is the corresponding normalized spatially averaged conductivity <Δσ~k>perturb. [Figure 17D] Examples, Comparative Examples and Conventional Examples <le>, SE, IN. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0011] Hereinafter, an in vivo visualization device 100 according to an embodiment of the present invention will be described with reference to the drawings. As shown in FIG. 1, the in vivo visualization device 100 includes a current / voltage application measuring unit 10 and a measurement calculation unit 50. The measurement calculation unit 50 includes an image reconstruction unit 52, a target region identification unit 53, a learning unit 54, and an output unit 55. The measurement calculation unit 50 of the in vivo visualization device 100 includes, for example, 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 image reconstruction unit 52, the target region identification unit 53, the learning unit 54, and the output unit 55 are realized by the CPU executing a predetermined program. The program may be acquired via a recording medium or via a network. Also, a dedicated hardware configuration may be used to realize the configuration of the in vivo visualization device 100. Each unit will be described below.
[0012] (Current and voltage application measurement unit 10) The current-voltage application measurement unit 10 will be described with reference to FIG. 2. FIG. 2 is a schematic diagram of the current-voltage application measurement unit 10 with the sensor 20 placed on the leg of the subject. As shown in FIG. 2, the current-voltage application measurement unit 10 includes one or more sensors 20 and a control unit 30. The sensor 20 includes a plurality of electrodes 21 (the number of electrodes Q is shown as 16 as an example) that can be placed on the skin at intervals from each other, and a support 25 that holds the electrodes 21. The current-voltage application measurement unit 10 uses one or more sensors 20, applies a predetermined current or potential difference between the electrodes 21 with each electrode 21 of the sensor 20 in contact with the skin of the subject, and measures the potential difference or current. Here, the potential difference refers to the difference in voltage between two electrodes. When applying a current, the potential difference is measured based on a predetermined current-application-voltage measurement pattern (a pattern in which two electrodes are selected in sequence from a large number of electrodes, a current is applied, and the potential difference is measured sequentially). At this time, it is desirable to also measure the phase (the time lag between the applied current and the measured potential difference). When applying a potential difference, the current is measured based on a predetermined voltage application current measurement pattern (a pattern in which two electrodes are selected from a large number of electrodes in sequence, a potential difference is applied, and the current is measured in sequence). At this time, it is preferable to also measure the phase (the time lag between the applied potential difference and the measured current). Hereinafter, the case of applying a current will be mainly described, and a detailed description of the case of applying a potential difference may be omitted. Herein, the number Q of the electrodes 21 is, for example, 3 or more. The number Q of the electrodes 21 is preferably 8 or more. The number Q of the electrodes 21 is preferably 32 or less. The number Q of the electrodes 21 may be 16 or less. In order to increase the accuracy of the concentration distribution, it is preferable to have a large number of electrodes. The arrangement position of the electrodes 21 is not particularly limited. It is preferable that the electrodes 15 are arranged at equal intervals so as to surround the subject (here, the legs).
[0013] The electrodes 21 are electrically connected to the control unit 30. There are no particular limitations on the material or shape of the electrodes 21 as long as they can apply a current or potential difference to the skin of the person being measured. Examples of materials for the electrodes 21 include metals such as Au, Ag, Cu, and stainless steel, conductive polymers, fibers whose surfaces are coated with metals, and fibers whose surfaces are coated with conductive polymers. There are no particular limitations on the shape of the electrodes 21 in a plane perpendicular to the direction of current application so that each electrode 21 comes into contact with the skin, and the shape may be, for example, a circle or a polygon. It is preferable that the electrodes 21 are non-invasive electrodes.
[0014] The method of electrically connecting the electrodes 21 and the control unit 30 is not particularly limited, and any known electrical connection method can be used. In this embodiment, the electrodes 21 and the control unit 30 are connected by electric wires 41.
[0015] Support body 25 is not particularly limited as long as it can hold electrodes 21. Support body 25 is preferably capable of placing electrodes 21 on the skin of a subject. Here, "capable of being placed on the skin of a subject" means that each electrode 21 is placed so as to come into contact with the skin when the subject wears sensor 20.
[0016] The support 25 is preferably capable of applying a predetermined pressure to the extent that the electrode 21 can be brought into close contact with the subject. This improves the close contact between the electrode 21 and the subject, and allows a current or potential difference to be applied more accurately and a potential difference or current to be measured. The material of the support 25 is not particularly limited, and is preferably an insulator such as elastomer, leather, or cloth. The shape of the support 25 is not particularly limited, and may be, for example, a band shape.
[0017] (Control unit) The control unit 30 includes, for example, a multiplexer for switching between a current application electrode (or a voltage application electrode for applying a potential difference) that applies a current and a voltage measurement electrode (or a current measurement electrode for measuring a current) that measures a potential difference, and an impedance analyzer for performing voltage measurement (or current measurement) and phase measurement. The impedance analyzer is a component that changes the applied frequency and amplitude to measure impedance, that is, the ratio of a measured potential difference (applied potential difference) to an applied current (measured current), and its phase. The control unit 30 executes a predetermined program in, for example, a CPU, and controls the multiplexer and the impedance analyzer to perform impedance measurement (measurement of the ratio of a potential difference to a current, and its phase). The control unit 30 may be controlled only within the current / voltage application measurement unit 10 to perform impedance measurement, or the control unit 30 may be controlled according to a program executed in the measurement calculation unit 50 to perform impedance measurement. The result of the impedance measurement is sent to the measurement calculation unit 50. There is no particular limitation on the method of transmitting information to the measurement calculation unit 50. The data may be sent from the control unit 30 to the measurement calculation unit 50 via a wire, or may be sent to the measurement calculation unit 50 via other transmission means.
[0018] The current application voltage measurement pattern for electrode 21 will be described using the electrode arrangement in Figures 3A, 3B, and 3C as an example. The numbers representing the positions of electrode 21 are numbered, for example, counterclockwise from the first electrode serving as the reference. The number M of current application voltage measurement patterns differs for each current application voltage measurement pattern. Each current application voltage measurement pattern will be described below. Examples of current application voltage measurement patterns will be described below, but the present invention is not limited to the following current application voltage measurement patterns.
[0019] First, the current application voltage measurement pattern by the counter electrode method will be described. In this case, a current is applied between a pair of opposing electrodes. For example, referring to FIG. 3A, a current is applied to opposing electrodes such as the first and ninth electrodes, the second and tenth electrodes, and so on. In the case of FIG. 3A, the number of electrodes Q is 16, so there are a total of eight patterns. The potential difference is measured in electrode pairs such as the second and third electrodes, and the third and fourth electrodes, excluding the electrode to which the current is applied, and is measured 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 (potential difference measurement patterns) for one current application pattern. Therefore, in the case of the counter electrode method, the number of measurements (measurement patterns) M is 104 in total. Here, when a current is applied and a potential difference is measured, the measurement pattern is a voltage measurement pattern. When a potential difference is applied and a current is measured, the measurement pattern is a current measurement pattern.
[0020] Next, the current application voltage measurement pattern in the adjacent method will be explained. In this case, a current is applied between adjacent electrodes. For example, referring to FIG. 3B, a current is applied between adjacent electrodes, such as between the first and second electrodes, and between the second and third electrodes. In the case of FIG. 3B, the number of electrodes Q is 16, so there are 16 patterns in total. The potential difference is measured in pairs of electrodes, such as the third and fourth electrodes excluding the electrodes to which the current is applied, and is measured from the third and fourth electrodes to the fifteenth and sixteenth electrodes, so there are 13 voltage measurement patterns (potential difference measurement patterns) for one current application pattern. Therefore, in the case of the adjacent method, the number of measurements (measurement patterns) M is 208 in total.
[0021] The current application voltage measurement pattern by the reference method will be explained. In this case, the potential difference is measured for all combinations between the reference electrode and the electrodes other than the reference electrode. For example, in the explanation of FIG. 3C, a current is applied between the reference electrode and the electrodes other than the reference electrode, such as the first electrode and the second electrode, the first electrode and the third electrode, etc. In the case of FIG. 3C, the number of electrodes Q is 16, so there are 16 combinations in total. The potential difference is measured in electrode pairs such as the third and fourth electrodes excluding the electrode to which the current is applied, and is measured from the electrode pair of the third and fourth electrodes to the electrode pair of the fifteenth and sixteenth electrodes, so there are 13 voltage measurement patterns (potential difference measurement patterns) for one current application pattern. Therefore, in the reference method, the number of measurements (measurement patterns) M is 208 in total.
[0022] Hereinafter, an example of measuring a potential difference using the neighbor method in the in vivo visualization device 100 of this embodiment will be described. Note that, although a calculation example for one sensor 20 will be described below, the same calculation can be performed when there are two or more sensors 20. Also, the in vivo visualization device 100 according to this embodiment learns predetermined parameters using sparse Bayesian learning.
[0023] (Image reconstruction unit) The image reconstruction unit 52 of the measurement calculation unit 50 creates an electrical property distribution inside the subject based on the predetermined parameters and the potential difference and phase current (hereinafter sometimes referred to as the first measurement data) and the current and phase (hereinafter sometimes referred to as the second measurement data) measured by the current / voltage application measurement unit 10. Specifically, the image reconstruction unit 52 creates a multiple measurement matrix from at least one of impedance, resistance, reactance, capacitance, admittance, conductance, susceptance, and phase obtained from the first measurement data or the second measurement data, and creates an electrical property distribution from the multiple measurement matrix and the predetermined parameters. Here, an example will be described in which a multiple measurement matrix is created using impedance. For example, the image reconstruction unit 52 generates a multiple measurement matrix (MMM) of impedance time difference ΔZ, and creates an electrical property distribution inside the subject's body from the MMM of the created impedance time difference ΔZ and the predetermined parameters. The predetermined parameters are, for example, a prior variance vector γ for sparsity, a positive definite matrix B for time correlation, and a variance value λ of a multidimensional normal distribution for noise. The initial values of the parameters may be input based on data. The initial values may be determined based on a priori information on electrical properties in a living body, or may be determined empirically in consideration of noise in experimental conditions. The initial electrical property distribution created by the image reconstruction unit 52 based on the initial predetermined parameters is sent to the target region identification unit 53. The initial electrical property distribution is sent to the target region identification unit 53 and subjected to identification processing. The post-identification electrical property distribution obtained after the initial electrical property distribution is identified is further sent to the learning unit 54. When the image reconstruction unit 52 creates an updated updated electrical property distribution based on the parameters updated by the learning unit 54, it sends it to the learning unit 54 or the output unit 55. Specifically, the image reconstruction unit 52 repeats the creation of the updated electrical property distribution in the image reconstruction unit 52 and the update of the predetermined parameters in the learning unit 54 until the learning unit 54 satisfies a termination condition described later. When the learning unit 54 has completed updating of the parameters, the image reconstruction unit 52 sends the electrical property distribution to the output unit 55 .
[0024] The electrical property distribution is not particularly limited, and examples thereof include spatial conductivity, spatial permittivity distribution, spatial phase distribution, temporal conductivity distribution, temporal permittivity distribution, temporal phase distribution, electrical conductivity response distribution of applied frequency, permittivity response distribution of applied frequency, phase response distribution of applied frequency, etc. Hereinafter, a method for creating an electrical property distribution will be described using a method for creating an internal electrical conductivity distribution of a subject as an example.
[0025] The initial time t0 and the time t obtained from the current, etc. measured by the current / voltage application measurement unit 10 k (k=1,2,...,K) and impedance difference ΔZ k is expressed by the following formula (1). Here, J in formula (1) is a sensitivity matrix expressed by the following formula (2). Δσk in formula (1) is a time t k The conductivity distribution in the following formula (1): k is a noise vector expressed by the following formula (4). Here, in the following formulas (2) and (3), N is the total number of meshes (total number of elements) that make up the image, and M in the following formula (2) is the number of measurements (number of measurement patterns). Also, the symbol formed by combining a circle and a / in formula (1) indicates division of each element of a vector.
[0026]
number
[0027] The image reconstruction unit 52 calculates ΔZ k ΔZ of MMM expressed by the following formula (5) is generated. In the following formula (5), M is the number of measurements (the number of measurement patterns), and K is the number of measurements at time t k The ΔZ of the MMM is expressed by the following equation (8) from the conductivity distribution Δσ of the MMM expressed by the following equation (6) and the noise matrix ε of the MMM expressed by the following equation (7). When ΔZ of the MMM in equation (8) is converted into a column vector, the following equation (9) is obtained. Here, * is the symbol for column vectorization. ΔZ in equation (9) * is expressed by the following equation (10). * is expressed by the following equation (11). * is expressed by the following formula (12). ~ is expressed by the following formula (13). I in the following formula (13) is a unit matrix expressed by the following formula (14). The symbol formed by combining a circle and a cross in the following formula (13) represents the Kronecker product symbol.
[0028]
number
[0029]
number
[0030] Here, the prior probability distribution p(Δσ in sparse Bayesian learning is * ;Σ pre ), likelihood p(ΔZ * |Δσ * ;ε * ) and the posterior probability distribution p(Δσ * |ΔZ * ;Σ past In sparse Bayesian learning using temporal correlation, we define a prior covariance matrix Σ consisting of a zero-mean vector 0, a prior variance vector γ for sparsity, and a positive definite matrix B for temporal correlation. pre Assuming a multivariate normal distribution, the prior probability distribution p(Δσ * ;Σ pre ) is expressed by the following equation (15). The prior variance vector γ is expressed by the following equation (16), the positive definite matrix B is expressed by the following equation (17), and the prior covariance matrix Σ pre are expressed by the following formulas (18A) and (18B). The locality can be adjusted by changing the magnitude of the prior variance vector γ related to sparsity. In the following formula (18B), it is preferable to make the positive definite matrix B common to all elements N, since this can prevent overfitting in learning. Overfitting refers to undesirable machine learning behavior that occurs when a machine learning model provides accurate predictions for training data but does not predict for new data.
[0031]
number
[0032] The noise probability distribution p(ε * If we assume that the noise covariance matrix λI is a multidimensional normal distribution with a zero-mean vector 0 and each element is independent and has the same noise covariance matrix λI, it can be expressed by the following equation (19). Here, the noise covariance matrix λI is expressed by the following equation (20). Here, λ in equation (20) is the variance value of the multidimensional normal distribution. From equations (9) and (19), the impedance time difference ΔZ * The likelihood of p(ΔZ * |Δσ * ;ε * ) is expressed as a conditional probability distribution in the following equation (21).
[0033]
number
[0034] The relationship between Bayes' theorem, that is, the prior probability distribution (19) and the likelihood (21), is as shown in the following equation (22). Since the posterior probability distribution can also be assumed to be a normal distribution by the following equation (22), the posterior probability distribution is expressed as the mean vector of the conductivity and the posterior covariance matrix Σ post Therefore, it is expressed by the following equation (23). The average vector of the conductivity is expressed by the following equation (24), and the posterior covariance matrix Σ post is expressed by the following equation (25). Since the normal distribution maximizes the probability at the expected value, the maximum a posteriori probability estimate (MAP estimate) of Δσ is the average vector of conductivity in equation (23), and is expressed by the following equations (26) and (27). Here, T is the transpose symbol. The conductivity distribution Δσ is obtained by converting Δσ* in the following equation (26) back into an N×K-dimensional Δσ. This makes it possible to create the electrical property distribution (here, the conductivity distribution) inside the subject's body.
[0035]
number
[0036] (Target area identification unit 53) The target region identifying section 53 identifies the target region by performing a predetermined process on the initial electrical property distribution obtained by the image reconstruction section 52. A method for identifying the target region by the predetermined process will be described below.
[0037] The target region identification unit 53 performs preprocessing on the electrical property distribution obtained by the image reconstruction unit 52 to generate a target region index distribution. The preprocessing is not particularly limited as long as it can identify the biological tissue that is the target region, and may be standard deviation processing, difference processing between the maximum and minimum values at a specified pixel a (max(a)-min(a)), normalized difference processing of the change between the maximum and minimum values ((max(a)-min(a)) / min(a)), first-order differentiation processing (∂y / ∂k: y is the target electrical property), second-order differentiation processing (∂ 2 y / ∂k 2 ), Fourier transform, wavelet transform and various other frequency analyses. These are performed by treating the total number of pixels N and the total number of frames K as an N x K matrix (singular value decomposition, principal component analysis, independent component analysis, etc.). The above preprocessing can be performed temporally if you wish to see local spatial changes, and spatially if you wish to see local temporal changes. Below, we will explain using standard deviation processing as an example to see spatial changes.
[0038] Using the conductivity distribution obtained by the image reconstruction unit 52, the target region index distribution (here, the temporal standard deviation image v) expressed by the following equation (28) is calculated according to the following equation (29). k,n is the time t k is the conductivity value of the nth element in <σ n > is the time average of the value of the nth element.
[0039] In order to extract the time-space local change of the electrical property in the region of interest (ROI) in a field (here, a field where the electrical property of the background (conductivity) changes significantly in time or space (here, a field where the electrical property changes in time), elements other than the ROI are deleted, and in order to identify the elements of the ROI, a classification process is performed based on the target region index distribution (here, a temporal standard deviation image v) and a predetermined threshold c to create a classified electrical property distribution. The classification process is not particularly limited as long as it can delete elements other than the ROI, and may be, for example, a process in which it is determined whether each value of the target region index distribution falls within the range of the threshold, and if it does not fall within the range of the threshold, the element is deleted. For example, the process is performed according to the following formula (30), and the classified electrical property distribution obtained from formula (30) is converted into a column vector again and sent to the learning unit 54. Here, the predetermined threshold c is determined based on the relationship between the σ of the target region ROI and the σ of the background, based on a priori information on the σ of each biological tissue. For example, the predetermined threshold c may be determined according to the ratio of the number of pixels occupied by the target region (ROI) to the total number of pixels of the image.
[0040]
number
[0041] The learning unit 54 updates predetermined parameters based on the updated electrical property distribution obtained by the image reconstruction unit 52 or the post-classification electrical property distribution obtained by the target region classification unit 53. Specifically, the learning unit 54 learns the parameter matrix Θ {e.g., λ, γ, B} by maximizing the negative log marginal likelihood -logp(ΔZ*|Θ) or maximizing the average data likelihood Q(Θ). For example, when each probability distribution is assumed to be a normal distribution, the likelihood function L(Θ) is obtained by considering only the logarithm of the marginal likelihood.
[0042] Here, the likelihood function L(Θ) is expressed by the following equation (31). likelihood is expressed by the following equation (32). Maximization of the log marginal likelihood -logp(ΔZ*|Θ) may be performed by maximizing L(Θ). In this embodiment, a method of learning using the average data likelihood Q(Θ) will be described, but is not limited to the following method. Specifically, a case of maximizing the average data likelihood Q(Θ) using an expectation-maximization (EM) algorithm will be described. The average data likelihood is expressed by the following equation (33). E[·] in equation (33) represents an expectation value. The update value of the EM algorithm, i.e., Θ that maximizes the average data likelihood Q(Θ), is Θ where the partial derivative of Q(Θ) with respect to Θ is zero.
[0043] The update of γ and B in the parameter matrix Θ will be explained. By ignoring the first term on the right hand side of equation (33) for λ, we obtain the following equation (34). Here, Γ in equation (34) is a diagonal matrix with γ as a diagonal element, as expressed by equation (35) below. Trace{} is the trace of a square matrix. γ, which is the n-th element of γ, n The updated value of is γn in Eq. (34). に γ where the partial differential with respect to is zero n and is updated using the iterative calculation number iter as shown in the following equation (36). Here, Δσ * n is expressed by the following equation (37) and represents the time change of the nth element. post n is expressed by the following equation (38) and represents the corresponding covariance matrix.
[0044] Similarly, the updated value of B is B such that the partial differential with respect to B in equation (34) becomes zero, and is updated as shown in equation (39) below.
[0045] The update of λ in Θ will be explained. By ignoring the second term on the right hand side for γ and B in equation (33), it is expressed by the following equation (40). The updated value of λ is λ where the partial differential with respect to λ in equation (40) becomes zero, and is updated as shown in the following equation (41). The learning process of Θ is repeated until a termination condition is satisfied. For example, the termination condition may be that either the error δ expressed in the following equation (42) becomes the minimum value δmin or iter becomes the maximum number of iterations itermax is satisfied.
[0046]
number
[0047]
number
[0048] (Output section) The output unit 55 outputs the updated electrical property distribution obtained by the image reconstruction unit 52. The destination of the updated electrical property distribution is not particularly limited. For example, the output unit 55 may output and store the updated electrical property distribution in a storage device such as a HDD in the in vivo visualization device 100. The output unit 55 may also output the updated electrical property distribution to an external display device such as a liquid crystal display.
[0049] (In vivo visualization method) Next, a description will be given of an in vivo visualization method using the in vivo visualization device 100 according to this embodiment. Fig. 4 is a flowchart of the in vivo visualization method according to this embodiment. In the in vivo visualization method, predetermined parameters (here, γ, B, λ) are initialized, and impedance measurement (measurement of the ratio of potential difference to current and its phase) is performed using the current / voltage application measurement unit 10 (S1). Then, an electrical property distribution is created based on the initialized predetermined parameters and the measured impedance (S2). Next, it is determined whether the initial electrical property distribution created based on the initialized predetermined parameters or the updated electrical property distribution created based on the predetermined parameters updated by the learning unit 54 is created (S3). In this case, the case of iter=1 is determined to be the initial electrical property distribution. In the case of the initial electrical property distribution, the target area identification unit 53 performs an identification process for identifying the target area on the initial electrical property distribution, and creates a post-identification electrical property distribution (S4). Next, the learning unit 54 updates the predetermined parameters based on the post-identification electrical property distribution or the updated electrical property distribution (S5). Next, it is determined whether the termination condition (whether the error δ is at a minimum value or a predetermined number of iterations has been performed) is satisfied (S6). If the termination condition is satisfied, the output unit 55 outputs the updated electrical property distribution (S7). If the termination condition is not satisfied, the process returns to S2.
[0050] This concludes a detailed description of the in vivo visualization device 100. The 100 according to this embodiment is capable of identifying local changes in time and space. The technical scope of the present invention is not limited to the above-described embodiment, and various modifications can be made without departing from the spirit of the present invention. In addition, within the scope of the invention, the components in the above-described embodiments may be replaced with well-known components, and the above-described modified examples may be combined as appropriate. EXAMPLES
[0051] The present invention will be described below with reference to examples, but the present invention is not limited to the following examples.
[0052] (Experimental Example 1) Next, an example of an experiment conducted to verify the effectiveness of the in vivo visualization device 100 according to this embodiment will be described.
[0053] Using a numerical simulation simulating lymphedema using the in vivo visualization device of the present invention, we confirmed whether it is possible to extract local time-space changes in σ near veins, which is a characteristic of lymphedema, in a field where the conductivity of the SAT and the background of the muscle changes significantly over time, and qualitatively examined the accuracy. GSV / SAT and ρ GSV / muscle The quantitative study was carried out using the above as a parameter.
[0054] As a conventional example, image reconstruction was performed using the conventional iterative Gauss-Newton method. The equation used for image reconstruction is the following equation (43). iter k is the time t at the iter-th iteration k In equation (43), I is the unit matrix expressed by the following equation (44), and η is the regularization parameter. J is the sensitivity matrix, and J T is the transpose of the sensitivity matrix T. k is the measured impedance. Measured impedance ΔZ k was calculated by the finite element method (FEM) using MATLAB (registered trademark) R2020a (Mathworks, Natick, MA). The number of elements for the semi-problem calculation was 2946, and the number of elements N for the inverse problem calculation was N = 4560.
[0055] Figure 5A is a simplified simulated image of a human calf. It is composed of three parts: the Great Saphenous Vein (GSV), the Subcutaneous Adipose Tissue (SAT), and the muscle. The ROI was set to the GSV.
[0056] In order to simulate the physiological phenomena of the progression of two types of edema, the conductivity value σ of each site was changed over time from t0 to t6. Figure 5B shows the lymphedema simulation condition of condition 1, where only the σ of GSV changed at a constant rate a GSV The SAT and muscle σ changed over time at a constant rate a. Figure 5C shows the venous edema simulation condition of condition 2, where the GSV and muscle σ changed over time at a constant rate a. GSV =a muscle (=0.667) and the σ of SAT changes with time at a SAT (=0.222 GSV ) which changes with time.
[0057] The σ of SAT and muscle at the initial time t0 in the two conditions was 0.022 [S / m] for SAT and 0.321 [S / m] for muscle. The σ of GSV at t0 (= 0.022 [S / m]) was the same as that of SAT. Generally, σ during edema is proportional to the water content, so the rate of change a GSV was measured from the change in calf moisture while standing for 60 minutes. GSV = 0.667. Furthermore, since SAT changes more slowly than GSV, SAT =a GSV / 3.
[0058] The sensitivity matrix J was calculated under homogeneous conditions, i.e., σ = 1.0 [S / m]. The applied current I0 and applied frequency f were set to I0 = 1 mA and f = 1 kHz, respectively, and the adjacent method was applied as the applied pattern.
[0059] The threshold c in equation (30) is set as the median of the temporal standard deviation image ν, and the initial values of λ, γ, and B are set as follows: λ is set to 0.01, γ is set as shown in the following equation (45), and B is set as shown in the following equation (46). Gaussian noise is added to Z so that the S / N ratio is 40 dB. The embodiment (image reconstruction unit 52 + target region identification unit 53 + learning unit 54: with ROI identification) and the comparative example (image reconstruction unit 52 + learning unit 54: without ROI identification) under two conditions, and the maximum number of iterations in the conventional method, the iterative Gauss-Newton method (equation (43)) are shown. max =10. The regularization parameter η in the iterative Gauss-Newton method is η = 4.17 × 10 using the L-curve method of PCHansen. -6 (Lymphedema simulation condition), η = 1.80 × 10 -5 (Venous edema simulation conditions).
[0060]
number
[0061] FIG. 6 shows the conductivity distribution Δσ reconstructed by the embodiment under condition 1 (image reconstruction unit 52 + target region identification unit 53 + learning unit 54: with ROI identification), the comparative example (image reconstruction unit 52 + learning unit 54: without ROI identification), and the conventional example (Gauss-Newton method). Under condition 1, the iterative Gauss-Newton method was able to image the time change in conductivity near the GSV, but the image was blurrier than the original image. Furthermore, the effect of ringing artifacts became greater. On the other hand, the embodiment and comparative example using sparse Bayesian learning were able to extract only local conductivity changes.
[0062] Next, FIG. 7 shows the conductivity distribution Δσ reconstructed by an embodiment under condition 2 (image reconstruction unit 52 + target region identification unit 53 + learning unit 54: with ROI identification), a comparative example (image reconstruction unit 52 + learning unit 54: without ROI identification), and a conventional example (Gauss-Newton method). Under condition 2, the iterative Gauss-Newton method was able to image the boundary between the SAT and the muscle, but was unable to image the time change in conductivity near the GSV. This is because all parts changed over time, and the difference between the initial time t0 and time t k This is considered to be because the background could not be made constant between the image and the GSV. On the other hand, in the comparative example (image reconstruction unit 52 + learning unit 54: no ROI identification), the conductivity change of the SAT could be imaged, and further, the vicinity of the GSV, which has a particularly large time change in the SAT, could be extracted. However, a circular artifact was generated due to the influence of the change in the SAT. In the case of the embodiment, only the conductivity change near the GSV could be extracted. Even though the entire part changed over time, the identification process of the target region was performed, and the conductivity change near the GSV, which has a particularly large time-space local change, could be clearly extracted in the target region ROI. In the simulation results, it was revealed that the in vivo visualization device 100 is capable of visualizing and measuring local σ that changes over time.
[0063] (Experimental Example 2) In vivo visualization was performed on 15 healthy subjects using the in vivo visualization device 100. A schematic diagram of the measurement device used in Experimental Example 2 is shown in Fig. 8. As shown in Fig. 8, conventional bioelectrical impedance analysis sensors were attached to the hands and feet, and the conventional bioelectrical impedance analysis sensors were connected to the bioelectrical impedance analyzer.
[0064] The in vivo visualization device according to the present embodiment is composed of a sensor consisting of 16 dry electrodes (5×10 mm) attached to the right calf, a digital multiplexer connected to the sensor, an impedance analyzer (IM3570, HIOKI, Japan), and a control PC. The impedance Z measured by the impedance analyzer is transmitted to the PC using a USB cable by the impedance analyzer connected to the coaxial cable, the digital multiplexer, and the PC containing the edema discrimination method software. The conventional BIA is composed of electrodes (two for the hands and two for the feet) (InBodyS10, InBody Japan Inc.). It was confirmed in a preliminary test that the in vivo visualization device according to the present embodiment and the conventional BIA do not interfere with each other, even if the electrodes are attached, as long as both measurements are not performed simultaneously.
[0065] Figure 9 showed the experimental protocol for prolonged standing and leg elevation to induce local spatiotemporal changes (LSTC) in extracellular fluid (ECF) of subcutaneous adipose tissue (SAT). At the beginning of the experiment, subjects lay on a reclining bed and performed leg elevation at an angle φ = 30 degrees to reduce leg swelling affected by previous activities such as exercise and muscle fatigue. During prolonged standing, subjects stood for 40 min without small movements to induce an increase in ECF volume, and measurements by the in vivo visualization device and conventional BIA were performed consecutively, taking 3 min and 2 min, respectively. The above process was repeated 8 times (k = 1-8:t k =5 / 10 / 15 / 20 / 25 / 30 / 35 / 40 min). During leg elevation, subjects used a reclining bed to elevate their legs for 3 min and then slowly stood up. Afterwards, in vivo visualization and conventional BIA measurements were performed consecutively, for a total of 8 min to induce a decrease in ECF volume. The above process was repeated three times for each subject (k=8-11:t k =50 / 60 / 70 minutes).
[0066] In Experimental Example 2, 15 healthy subjects (male: 9, female: 6, age: 23.1 ± 2.3 years, body mass index: 23.1 ± 4.9 kg / m2) participated. Impedance Z was measured in various time windows (k = 1-11) using current I0 = 1 mA at two frequencies f1 = 1.6 kHz and f2 = 4.7 kHz, selected considering the current path penetrating only the extracellular space. Impedance Z was then used to image-reconstruct the frequency difference conductivity distribution Δσ on a mesh of finite element method (FEM) generated by NETGEN supported by MATLAB R2020a (Mathworks). To quantify the local space-time change (LSTC) for evaluating lower limb edema, the spatial average conductivity 〈Δσ〉SAT was calculated on the separated SAT.
[0067] As a comparison, conventional BIA was measured in the right leg at a frequency of f = 5 kHz to determine ECF. These two parameters were compared by the correlation coefficient R, with the significance level of the t-test set at 0.05. Furthermore, the normalized values of the in vivo visualization device measurements and conventional BIA measurements with a mean of 0 and a standard deviation of 1, i.e., the normalized spatial average conductivity (<Δσ> SAT ) and the normalized conventional impedance Z BIA were calculated and their trends were evaluated.
[0068] To obtain the location of the great saphenous vein (GSV) of the LSTC, ultrasound images were acquired using a high-frequency matrix probe (LOGIQePremium, GE Healthcare, Japan) for the location of the electrodes of the sensor disclosed herein. The minimum error bound and maximum number of iterations for the in vivo visualization device were δmin = 1 × 10, respectively. -2 and iter max = 15 and optimized. The effects of these parameters are explained below.
[0069] The regularization parameter β for the time correlation was set to β = 4. The threshold c was set to 25% of the time standard deviation image ν. Figure 10 shows the frequency difference conductivity distribution Δσ of one subject reconstructed by the in vivo visualization device. It is cropped between the angles θ0 = 120° and θ1 = -150° (illustrated in Figure 11B) of the calf of one subject over time. The dimensions of the human calf are based on an xyz Cartesian coordinate system with the x- and y-axes orthogonal to the z-axis, and the z-axis is along the direction of gravity. The dashed circle indicates the local maximum location of Δσ related to the location of the great saphenous vein (GSV). Because the local spatiotemporal changes (LSTC) associated with the GSV are small and difficult to distinguish in the whole image, the whole circular image is shown only in a representative example in Figure 11B, and the Δσ is cropped for all 15 subjects. The angles θ0 and θ1 in Figure 10 are chosen to highlight the Δσ near the GSV location located on the anteromedial side of the tibia. The EIT electrode numbers where GSV was observed in the ultrasound images were located near EIT electrode numbers (3)-(5), although there was some variability among subjects. Therefore, the LSTC of the extracellular fluid (ECF) of isolated subcutaneous adipose tissue (SAT) can be accurately monitored by Δσ. That is, LSTC decreases during prolonged standing, and LSTC increases during leg elevation. High Δσ is also observed in locations other than near the GSV. This is likely because ECF is retained by capillaries near relatively large veins. More interestingly, locations close to the GSV showed the largest Δσ in most subjects.
[0070] FIG. 11A shows a cross section of a typical human calf, FIG. 11B shows Δσ of subject No. 12, and FIG. 11C shows an ultrasound image near EIT electrode No. (4) of the same subject. In FIG. 11B, the white dashed lines indicate the boundaries of the separated SAT. In contrast, the dashed lines in FIG. 11C show the direction from the origin of the ultrasound image to the GSV position, and the direction from the origin to the local maximum position, and the angle θ EITG = 150°, and the angle θ on the dashed line USG = 150°. These figures confirm that the LSTC location of Δσ is shown in the isolated SAT, not in the muscle compartment or bone. In particular, the location of the Δσ maximum, shown as the dashed circle in Figure 11B, can be correlated to the GSV location.
[0071] We explain why the maxima of Δσ (dashed circles in Figure 10) are associated with the location of the great saphenous vein (GSV). The global minimum of the sparse Bayesian learning framework is always the sparsest due to the prior variance vector γ, which is related to the sparsity that controls the individual elements of Δσ. The GSV is located between the superficial fascia and the saphenous fascia, and its depth location depends on the thickness of the subcutaneous tissue. Furthermore, the GSV plays an important role in returning blood from the subcutaneous tissue of the leg to the heart. Several large maxima were observed in Figure 10, which are speculated to be related to ECF diffused from the capillaries near the GSV. From the above, we found that the in-vivo visualization device can extract the temporal correlation of Δσ of the isolated SAT.
[0072] Figure 12 shows the normalized spatially averaged conductivity <Δσ> in the separated SATs. SAT and normalized conventional BIA impedance z during prolonged standing and leg elevation. BIA In FIG. 12, <Δσ> SAT and z BIA decreases with prolonged standing, and <Δσ> SAT and z BIA increases with leg elevation. SAT The standard deviation of z BIA Although it is relatively higher than the standard deviation of SAT showed a good trend of ECF.
[0073] Figure 13 shows the spatial average conductivity <Δσ> of the isolated SAT for the calf of a subject. SAT and the conventional impedance z BIA The vertical axis shows the correlation with spatial average conductivity <Δσ> SAT The horizontal axis is z BIA It is [Ω]. The correlation coefficient R and p-value between the two parameters are also shown. As shown in Fig. 13, there is a strong positive correlation between Δσ and z BIA and this tendency was the same for all subjects. The scores were 0.715 < R < 0.957 (p < 0.05). Therefore, it was confirmed that LSTC showed the same tendency as the temporal change of zBIA, and furthermore, LSTC, which was impossible with conventional BIA, could be visualized. Regarding the range of each value, z BIA was in the range of 170 to 340 [Ω], while <Δσ> SAT was in the range of 10 to 1400 [-]. Conventional BIA mainly evaluates the static state based on absolute values, and the in-vivo visualization device mainly evaluates the dynamic tendency based on relative values. The variation of <Δσ> SAT was larger than the variation of z BIA and there was a difference of about two orders of magnitude at most.
[0074] Explain the reason for the occurrence of local spatio-temporal change (LSTC) of the frequency-difference conductivity distribution Δσ in isolated subcutaneous adipose tissue (SAT). When standing for a long time, the hydrostatic pressure in the lower limbs increases due to the gravity of the blood, and the capillary filtration into the extracellular space of tissues such as SAT increases. At the same time, the increase in the osmotic pressure of the extracellular space, which mainly depends on the sodium ion concentration, reduces the reabsorption in the extracellular fluid (ECF). Therefore, the proportional relationship of the ratio value shown in the following formula (47) is derived. Here, L ECF and L ICF are the volume of ECF and the volume of intracellular fluid (ICF), respectively, and Na +ECF and Na +ICF are the sodium ion concentration of ECF and the sodium ion concentration of ICF, respectively. The conductivity distribution σ of the NaCl solution, which is the main component of ECF, does not depend on the frequency at f < 1 MHz, while the σ of the protein solution, which is the sub-component of ECF, depends on the frequency at low frequencies. The impedance Z measured at a low frequency of f < 5 kHz represents ECF. Since the protein concentration of healthy people is kept constant during long standing, the normalized spatial average conductivity <Δσ> in Fig. 12 SAT is Na + associated with a frequency-dependent decrease in the σ of ECF caused by an increase in From the above, in the case of a healthy person, the following relationship (48) is derived. According to (48), the change in the ECF volume of the right leg during prolonged standing and leg elevation, ΔL ECF is the spatially averaged conductivity <Δσ> SAT Here, ΔL in the kth time frame is estimated as follows: ECF and <Δσ> SAT is taken into account. w is the subject's weight, ρ TBF is the ratio of total body fluid (TBF) to total body weight (here, 0.6), and ρ TBF / ECF is the ratio of ECF to TBF (here 0.29), and ρ RL is the ratio of the right leg to the whole body (here, 0.1).
[0075]
number
[0076] Figure 14 shows the ΔL of the right leg of 15 subjects. ECF The time change of ΔL is shown. ECF increases during prolonged standing, and ΔL ECF has decreased. SAT represents only LSTC, but ΔL in Fig. 14 ECF is 〈Δσ〉 SAT The tendency of the right leg is ΔL ECF is reasonable, based on the assumption that it is consistent with Next, the range of ΔL ECF was investigated. Figure 15 shows the ECFmax (ΔL ECF ) maximum volume change (t 8 = ΔL at 40 min ECF ) and right leg v RL There was a correlation between muscle mass and max(ΔL ECF ) and v RL There was a strong positive correlation between the score and the muscle mass, with R = 0.729 (p < 0.01). Since the change in blood flow is correlated with muscle mass, the increase in intravascular pressure caused by max (ΔL ECF ) is also estimated to be correlated with muscle mass.
[0077] (Experimental Example 3) In order to quantitatively evaluate this high-precision EIT method, we compared the embodiment (image reconstruction unit 52 + target region identification unit 53 + learning unit 54: with ROI identification), the comparative example (image reconstruction unit 52 + learning unit 54: without ROI identification), and the conventional iterative Gauss-Newton method (equation (43), Gauss-Newton algorithm Tikhonov). Here, Δσ iter k is the conductivity distribution at the iter iteration at the kth arbitrary time (1≦k≦3), and η=0.01 is the regularization parameter. The measured impedance ΔZ k was calculated by the same FEM calculation as above. The element numbers of the forward problem and the inverse problem are 3219 and 4261, respectively. The impedance Z was measured at two frequencies, f1 = 1.6 kHz and f2 = 4.7 kHz, with SNR = 40 dB due to additive Gaussian noise.
[0078] FIG. 16A shows an example of a human calf including muscle, SAT, and GSV with the same boundaries as those used in FIG. 11B. FIG. 16B shows the conductivity values σ for the simulation. Under a simple unstable background field, only the σ of the GSV is assumed to be time-dependent. Then, the σ of the SAT and muscle are assumed to be time-independent, and furthermore, the σ of the muscle is assumed to be frequency-independent. Even if the muscle and SAT are time-independent, image reconstruction of local spatiotemporal changes (LSTC) due to the influence of very small GSV is difficult under frequency-dependent SAT. The σ of the GSV at frequency f1 is 2, 3, and 4 times the σ of the SAT at any t1-th, t2-th, and t3-th, respectively. The σ of the GSV at frequency f2 is adjusted to satisfy the following equation (50). Also, a GSV = 0.905 is estimated from the experimental data during long-term storage in Figure 12. This is because the Na +ECF This reflects the frequency-dependent decrease in the σ of ECF caused by an increase in k When temporarily increasing, Δσ k The hyperparameters of the sparse Bayesian learning algorithm are the regularization parameter β = 4 and the minimum error bound δmin = 1 × 10 -5 The threshold c of the in vivo visualization device was set to 75% of the time standard deviation image ν, and kept higher. This has a different sign from that in (30). This c is one of the important parameters that greatly affect the reconstruction results, so it needs to be heuristically optimized for each application. For example, the target LSTC will disappear with an inappropriate c. The maximum number of iterations for all algorithms is iter max is the time-averaged localization error expressed by the following equation (51). <le>, the spectrum error SE expressed by the following equation (52), and the image noise expressed by the following equation (53) <in>We set itermax=50 to examine the effect of different iterations on the three metrics: k is the reconstructed perturbation Δσ k perturb Distance between the origin and the center of gravity, <Δσ k > perturb is the perturbation Δσ k perturb The spatially averaged conductivity at the kth arbitrary time in <Δσ k > is a normalized value with a mean of 0 and a standard deviation of 1. std(-) is the symbol for standard deviation, x k is Δσ k perturb Here, Δσ k perturb is defined as the connected pixels with a conductivity distribution higher than half the global maximum. The superscript true denotes the true value of each metric. In ideal conditions, <le>, S.E., and <in>The value is 0.
[0079]
number
[0080] Figures 17A and 17C show the Δσ at iter=15 reconstructed by each algorithm under an unstable background field and the corresponding normalized spatially averaged conductivity <Δσ ~ k> perturb FIG. 17B shows an ideal reconstructed image. <le>, SE, and IN are shown in Figure 17D.
[0081] The reconstruction performance does not change significantly with the number of iterations, so we show Δσ at it=15. Reconstructed perturbation Δσ due to the presence of GSV k perturb is distinguished by the sparse Bayesian learning algorithm, while the traditional Gauss-Newton algorithm is Δσ k perturb The results were not clear and there were artifacts. k > perturb was the furthest from the positive values among the algorithms tested. This is thought to be because the Gauss-Newton algorithm simply obtains the frequency difference without using time correlation or sparsity, and therefore cannot cancel the effects of an unstable background field. On the other hand, in the sparse Bayes learning algorithm, Δσ k perturb are clearly distinguished at the correct position. k perturb Regarding , the in vivo visualization device of the embodiment was slightly closer to a positive value than the comparative example, but both of them estimated the spectrum appropriately, although artifacts were present in the comparative example. The three algorithms were not significantly dependent on the number of iterations, and the SBL algorithm performed better than the Gauss-Newton algorithm. <le>is the same in the examples and comparative examples, and the score <le>=0.139±0.012. For GSV localization, iter=2 was sufficient for both algorithms under these conditions. The spectral error SE was 0.031±0.004 and 0.061±0.001 for the example and comparative example, respectively. This shows that the in vivo visualization device of the example estimates the time change of the frequency difference more appropriately due to the improvement of noise robustness by SAT separation. The image noise IN was 0.072±0.009 and 0.132±0.001 for the example and comparative example, respectively. Similar to the SE result, it was found that SAT separation contributes to noise suppression, which is important for monitoring LSTC. [Explanation of symbols]
[0082] 10 Current / voltage application / measurement unit, 52 Image reconstruction unit, 53 Target area identification unit, 54 Learning unit, 55 Output unit, 100 In-vivo visualization device< / le> < / le> < / le> < / in> < / le> < / in> < / le> < / le>
Claims
1. a sensor including a plurality of electrodes spaced apart from one another and positionable on the skin of a subject; a current / voltage application measurement unit that applies a current or a potential difference between the electrodes while the electrodes are in contact with the skin, and measures first measurement data which is a potential difference and a phase based on a current application / voltage measurement pattern when the current is applied, and measures second measurement data which is a current and a phase based on a voltage application / current measurement pattern when the potential difference is applied between the electrodes; an image reconstruction unit that creates an electrical property distribution in the subject's body based on the first measurement data or the second measurement data and predetermined parameters; a target region identifying unit that performs an identification process for identifying a target region with respect to the electrical property distribution and creates a post-identification electrical property distribution; An in vivo visualization device comprising:
2. The in-vivo visualization apparatus according to claim 1 , wherein the classification process is performed based on a predetermined threshold value.
3. The target area identification unit, A target area index distribution is generated by performing standard deviation processing on the electrical property distribution; The in vivo visualization apparatus according to claim 1 , wherein the discrimination process is performed on the target region index distribution to generate the post-discrimination electrical property distribution.
4. The image reconstruction unit, creating a multiple measurement matrix from at least one of impedance, resistance, reactance, capacitance, admittance, conductance, susceptance, and phase obtained from the first measurement data or the second measurement data; creating the electrical property distribution from the multiple measurement matrix and the predetermined parameters; The in-vivo visualization device according to claim 1 or 2, wherein the predetermined parameters include a prior variance vector γ for sparsity, a positive definite matrix B for time correlation, and a variance value λ of a multidimensional normal distribution for noise.
5. The in vivo visualization apparatus according to claim 1 , further comprising a learning unit that updates the predetermined parameters based on at least the post-identification electrical property distribution.
6. 6. The in vivo visualization apparatus according to claim 5, wherein the learning unit updates the predetermined parameters by maximizing an average data likelihood using an expectation-maximization algorithm.
Citation Information
Patent Citations
Lymphedema monitoring device
JP6555715B2
