JP2017023548A - Estimation method and estimation system - Google Patents

Abstract

An object of the present invention is to provide an estimation method for estimating the state of a subject more accurately than in the past. An estimation method for estimating a state of a subject includes time-series brain activity information measured when a perceptual stimulus that can be correlated with the state is given to each subject. (S1) to acquire each different region in the brain, and for each subject, each temporal change in brain activity information obtained from each region in the subject's brain, and perceptual stimulation given to the subject Using a step (S3) of calculating a similarity group consisting of the respective similarities between changes in time of each of the above, the state of each subject quantified in advance, and the similarity group corresponding to each subject And learning a correlation between the similarity group and the state of the subject (S4). [Selection] Figure 1

Description

  The present disclosure relates to a technique for estimating a state of a subject using brain activity information obtained from the subject's brain.

  2. Description of the Related Art Magnetic resonance apparatuses that measure a subject's brain activity using functional magnetic resonance imaging (hereinafter, also referred to as “fMRI”) have become widespread. In recent years, a technique has been known in which a subject is given some kind of perceptual stimulus, changes in the brain activity of the subject are measured with a magnetic resonance apparatus, and the state of the subject is estimated from the changes in the brain activity.

  Regarding this technology, Japanese Patent Application Laid-Open No. 2014-133080 (Patent Document 1) discloses an information acquisition method for evaluating the stress of a subject from brain activity information obtained by giving a perceptual stimulus to the subject. More specifically, the information acquisition method includes a step (1) of providing a sensory stimulus to at least one of the cortex visual cortex, cerebral cortex auditory cortex, and cerebral cortex equilibrium sensory area of the subject; The step (2) of measuring brain activity in a plurality of non-identical areas in the area when the sensory stimulus is given, and the stress that the sensory stimulus gives to the subject based on the brain activity data of the plurality of areas The process (3) which evaluates is included.

JP 2014-133080 A

  The information acquisition method disclosed in Patent Document 1 evaluates the stress of a subject using brain activity information in a specific part in the brain. However, the active part of the brain for sensory stimuli varies depending on the type of sensory stimulus given to the subject. Therefore, even if the information acquisition method can evaluate the stress of the subject, it cannot estimate the other state of the subject. Therefore, an estimation method capable of specifying an active part in the brain that responds to a sensory stimulus given to the subject and estimating the state of the subject from the brain activity information in the active part is desired.

  The present disclosure has been made to solve the above-described problems, and an object in one aspect thereof is to provide an estimation method capable of estimating the state of a subject more accurately than in the past. . The objective in the other situation is to provide the estimation system which can estimate a test subject's state more correctly than before.

  According to an aspect, an estimation method for estimating a state of a subject is provided. The estimation method includes obtaining time-series brain activity information measured when a perceptual stimulus that can be correlated with the state is given to each subject for each different region in the subject's brain; For each subject, the degree of similarity between each temporal change in brain activity information obtained from each region in the subject's brain and the temporal change in sensory stimulation given to the subject The correlation between the similarity group and the state of the subject is learned using the step of calculating the similarity group, the state of each subject quantified in advance, and the similarity group corresponding to each subject. Steps.

  Preferably, the estimation method further includes a step of estimating the state of the new subject from the similarity group calculated for the new subject using the correlation.

  Preferably, each brain activity information of each subject is represented as a time-series first numerical value group. The perceptual stimulus given to each subject is represented as a second numerical group in time series. The calculating step multiplies the numerical values corresponding to the same time in the first numerical value group and the second numerical value group, and adds the multiplied results to calculate the similarity.

  According to another aspect, an estimation system for estimating a state of a subject is provided. The estimation system includes: an acquisition unit that acquires time-series brain activity information measured when a sensory stimulus that can be correlated to a state is given to each subject for each different region in the subject's brain; For each subject, the degree of similarity between each temporal change in brain activity information obtained from each region in the subject's brain and the temporal change in sensory stimulation given to the subject Learning the correlation between the similarity group and the subject's state using the calculation unit for calculating the similarity group, the state of each subject that has been quantified in advance, and the similarity group corresponding to each subject A part.

In one aspect, the state of the subject can be estimated more accurately than before.
The above and other objects, features, aspects and advantages of the present invention will become apparent from the following detailed description of the present invention taken in conjunction with the accompanying drawings.

It is a conceptual diagram which shows a learning process roughly. It is a conceptual diagram which shows an estimation process roughly. It is a figure which shows the code | symbol of MATLAB (trademark). It is a block diagram which shows the main hardware structures of an estimation system. It is a block diagram which shows the main hardware constitutions of information processing apparatus. It is a conceptual diagram which shows roughly the procedure of the last common game for investigating the reaction of the brain with respect to inequality. It is a figure which shows contrast with the estimated BDI (The Beck Depression Inventory) score and an actual BDI score. It is a figure which shows the active region in the brain with respect to an inequality feeling. It is a figure which shows the brain activity pattern of the amygdala when a subject is provided with inequality.

  Embodiments according to the present invention will be described below with reference to the drawings. In the following description, the same parts and components are denoted by the same reference numerals. Their names and functions are also the same. Therefore, detailed description thereof will not be repeated. Each embodiment or modification described below may be selectively combined as appropriate.

[Method of estimating the subject's condition]
With reference to FIG. 1 and FIG. 2, a method for estimating the state of the subject will be described. The estimation system 100 (see FIG. 3) that performs the estimation executes a learning process and an estimation process. FIG. 1 is a conceptual diagram schematically showing the learning process. FIG. 2 is a conceptual diagram schematically showing the estimation process.

  In the learning process, the estimation system 100 learns a correlation between brain activity information obtained from a subject by an arbitrary sensory stimulus and the state of the subject. The perceptual stimulus refers to a stimulus given to the subject through a sensory organ such as vision and hearing. The state of the subject includes, for example, the present or future tendency of the subject. The tendency includes a tendency of the subject's illness (for example, depression tendency, etc.).

  In the estimation process, the estimation system 100 obtains the same type of perceptual stimulus as the learning process to the subject to be examined, acquires brain activity information from the subject, and uses the correlation obtained in the learning process. The state of the subject to be examined is estimated from the obtained brain activity information.

Below, a learning process and an estimation process are demonstrated in order.
(Learning process)
First, the learning process will be described with reference to FIG. As shown in FIG. 1, the learning process mainly includes the processes of steps S1 to S4.

  In step S <b> 1, the estimation system 100 acquires brain activity information measured when a subject is given a sensory stimulus that can be correlated with a state of the subject for each different region in the subject's brain. . Below, the case where the image 15 (namely, irritation | stimulation with respect to vision) is used as a perceptual stimulus is demonstrated as a typical example.

  The image 15 is, for example, for giving the subject a sense of inequality. Although details will be described later, the image 15 represents a distribution between subjects, and the inequality is given by the difference in distribution.

  The estimation system 100 obtains brain activity images 1 </ b> A to 1 </ b> C from the subject 1 by measuring the brain activity information of the subject 1 at the same time that the subject 1 is inequalized by sequentially displaying the images 15. Similarly, the estimation system 100 obtains brain activity images 2A to 2C from the subject 2. Similarly, the estimation system 100 obtains brain activity images 3 </ b> A to 3 </ b> C from the subject 3. The brain activity images 1A to 3C may be expressed in one dimension or may be expressed in a plurality of dimensions. The brain activity images 1A to 3C are obtained by, for example, fMRI measurement. Each pixel of the brain activity images 1A to 3C represents brain activity information in each region in the subject's brain. Each brain activity information represents, for example, the blood flow volume of the corresponding part in the brain. Preferably, the estimation system 100 acquires the brain activity images 1A to 3C from the amygdala in the brain that reacts when feeling inequality.

  In step S <b> 2, the estimation system 100 digitizes the change in the perceptual stimulus given to the subject 1 as the numerical value group 12. Each numerical value of the numerical value group 12 represents the degree of emotion that the subject 1 is reminded by the image 15. For example, when inequality is given to the subject 1 by the image 15, each numerical value of the numerical value group 12 represents the degree of inequality. As an example, the degree of inequality corresponds to the amount of money distributed to each subject.

  In step S3, the estimation system 100 identifies an active region in the brain that responds to the applied sensory stimulus. More specifically, the estimation system 100 calculates the similarity between the time-series brain activity information and the numerical value group 12 in the same pixel of the brain activity images 1A to 1C for each pixel of the brain activity image.

  A method of calculating the similarity will be described by taking time series brain activity information 11A to 11C obtained from a predetermined part in the brain of the subject 1 as an example. The estimation system 100 calculates the correlation value between the brain activity information 11A to 11C and the numerical value group 12 as the similarity 16A. As the brain activity information 11A to 11C is linked to the perceptual stimulus, the value of the similarity 16A increases.

  The brain activity information 11 </ b> A to 11 </ b> C indicates a blood flow volume in a predetermined portion in the subject's brain, and is shown as a time-series numerical group. The estimation system 100 multiplies each of the numerical values corresponding to the same time among the brain activity information 11A to 11C shown as a numerical value group and the numerical value group 12, and adds up the results of the multiplications to obtain a similarity 16A. Is calculated. That is, the similarity 16 </ b> A corresponds to the inner product of the brain activity information 11 </ b> A to 11 </ b> C shown as a numerical value group and the numerical value group 12.

The estimation system 100 calculates the similarity for each region in the brain of the subject 1 like the similarity 16A. Similarity group calculated for each region of the brain of the subject 1 is represented as a brain activity pattern X _1. Similarly for the subject 2, the estimation system 100 calculates the similarity between the brain activity information of the same pixel in the brain activity images 2 </ b> A to 2 </ b> C and the numerical value group 13 for each region in the brain of the subject 2. generating a brain activity pattern X ₂ consisting degrees group. Similarly for the subject 3, the estimation system 100 calculates the similarity between the brain activity information of the same pixel in the brain activity images 3 </ b> A to 3 </ b> C and the numerical value group 14 for each region in the brain of the subject 3. generating a brain activity pattern X ₃ consisting degrees group.

In step S4, the learning unit 20 of the estimation system 100 collectively refers to the state of each subject that has been quantified in advance and the brain activity patterns X _{1 to} X ₃ (hereinafter referred to as “brain activity pattern X”) corresponding to each subject. To learn the correlation between the state of the subject and the brain activity pattern X. The state of the subject used for learning is quantified as a BDI score, for example. The BDI score indicates the degree of depression and is quantified from the results of the BDI test performed on each subject. The BDI test is performed after a predetermined period of time has elapsed since the test for obtaining the brain activity patterns X _{1 to} X ₃ was performed (for example, one year later).

The learning unit 20 learns a correlation between the brain activity pattern X and the future depression degree of the subject using the teacher data 22 including the actual BDI score of each subject. “Y _n ” shown in the teacher data 22 represents the actual BDI score in the nth subject.

As an example, the BDI score Z _t of the t th subject is estimated using the kernel method included in the standard library “sparse Bayesia”. The BDI score Z _t is represented by the following formula (1), for example.

“X _i ” in Equation (1) indicates the brain activity pattern of the i-th subject. “X _t ” indicates the brain activity pattern of the t-th subject of interest. “K (X _i , X _t )” indicates the similarity between the brain activity pattern X _i and the brain activity pattern X _t . The similarity is equivalent to, for example, the inner product of the brain activity pattern X _i and the brain activity pattern X _t .

The learning unit 20 learns the weight w _i so as to minimize the difference (= Z _t −Y _t ) between the actual BDI score Y _t and the estimated BDI score Z _t . Details of the learning method by the learning unit 20 will be described later.

The learning result 25 is obtained by the learning. In the learning result 25, they are associated with each other and respective weight _w 1 to w _n obtained for each and each subject's brain activity patterns _X 1 to X _n of the subject.

  As described above, in the learning process, the estimation system 100 uses the time-series brain activity information measured when a perceptual stimulus that can be correlated with the state of the subject is given to each subject, It acquires about each mutually different area | region. Thereafter, the estimation system 100 determines, for each subject, between each temporal change in brain activity information obtained from each region in the subject's brain and temporal change in sensory stimulation given to the subject. The brain activity pattern X (similarity group) consisting of the respective similarities is calculated. The estimation system 100 learns the correlation between the brain activity pattern X and the state of the subject using the state of each subject that has been quantified in advance and the brain activity pattern X corresponding to each subject.

  Since the brain activity pattern X represents the degree of interlocking with the perceptual stimulus for each region in the brain, the part responding to the perceptual stimulus can be easily identified. As a result, the estimation system 100 can accurately identify the active part in the brain even when the type of the perceptual stimulus changes. Since the estimation system 100 estimates the state of the subject in the estimation process described later using the brain activity information in the brain linked to the perceptual stimulus, it is possible to accurately estimate the state of the subject.

  In addition, since the brain activity pattern X indicates the degree of interlocking with the perceptual stimulus, if the perceptual stimulus type is the same, even if the content of the perceptual stimulus changes, the brain activity pattern X has the same tendency among subjects. . Therefore, the number of trials and contents of the perceptual stimulus given to each subject can be arbitrarily changed, so that a highly versatile learning function can be realized.

  In addition, although the example which gives a visual stimulus to a test subject as a perceptual stimulus was demonstrated above, the kind of perceptual stimulus is not limited to a visual stimulus. For example, the sensory stimulation given to the subject includes auditory stimulation, tactile stimulation, and the like.

  Moreover, although the example which gives the test subject the visual stimulus accompanied with an unequal feeling was demonstrated above, the perceptual stimulus may be a visual stimulus accompanied with another emotion. For example, the perceptual stimulus includes a visual stimulus accompanied with a sense of well-being, a sense of disgust, and the like.

  Furthermore, in the above description, an example of learning a correlation between a brain activity pattern and a depression state when a perceptual stimulus for giving an unequal feeling was given was explained. However, it is not limited to these. That is, the estimation system 100 can learn a correlation between a brain activity pattern when an arbitrary sensory stimulus is given and an arbitrary state of the subject. For example, when the estimation system 100 is used for marketing, the correlation between the brain activity pattern and the purchase will of the subject is learned using the brain activity pattern when the product is shown and the questionnaire result for the product as learning data. You can also.

(Estimation process)
With reference to FIG. 2, estimation processing in the estimation method according to the present embodiment will be described. In the estimation process, the estimation system 100 uses the correlation between the brain activity pattern X and the state of the subject (that is, the learning result 25), and the subject from the brain activity pattern X calculated for the subject to be examined. The state of is estimated. The estimation process mainly includes the processes of steps S11 to S14.

  In step S <b> 11, the estimation system 100 gives the subject T the same type of perceptual stimulus (for example, display of the image 15) as the learning process, and acquires brain activity images 5 </ b> A to 5 </ b> C from the subject T.

  In step S <b> 12, the estimation system 100 digitizes the change in the perceptual stimulus given to the subject T as the numerical group 18 by the same method as the learning process.

In step S13, the estimation system 100 calculates the similarity between the numerical value group 18 and time-series brain activity information in the same pixel of the brain activity images 5A to 5C for each pixel. When attention is paid to the brain activity information 17A to 17C in the predetermined pixels of the brain activity images 5A to 5C, the estimation system 100 corresponds to the brain activity information 17A to 17C shown as a numerical group and the numerical group 18 at the same time. Each of the numerical values is multiplied and the result of the multiplication is added to calculate the similarity 19A. Estimation system 100 calculates the similarity 19A for each pixel of brain activity image 5A-5C, generates a brain activity pattern _{X T.}

In step S14, the estimation unit 30 of the estimation system 100 estimates the state (for example, depression tendency) of the subject T using the learning result 25 obtained by the learning process. More specifically, the estimation unit 30 calculates each and brain activity pattern X ₁ to X _n included in the learning result 25, each of the similarity between the brain activity pattern X _T of the subject T to be inspected To do. Estimating unit 30 multiplies the respective weight w ₁ to w _n to each of the calculated degree of similarity, and summing the results of multiplying to estimate the change in BDI score of the subject T. If the estimated change in the BDI score shows a positive value, this indicates that the subject T is more likely to become depressed than at present. If the estimated degree of change in the BDI score is a negative value, this indicates that the degree of depression of the subject T is likely to be improved over the current level.

[Learning unit 20]
Hereinafter, the learning process by the learning unit 20 (see FIG. 1) will be described in more detail.

As a learning method by the learning unit 20, for example, a Bayesian kernel method is used. In the Bayesian kernel method, as shown in the above equation (1), the estimated BDI score Z _t is calculated by weighting the sum of (n−1) kernel functions.

As described above, by the kernel function K (X _i , X _t ) of Equation (1), the brain activity pattern X _i of the _i th subject and the brain activity pattern X _t of the t th subject of interest are Similarity between them is calculated. The weight w _{i in} equation (1) is a parameter that can be adjusted, and the difference (= Z _t −Y _t ) between the actual BDI score Y _t and the estimated BDI score Z _t is minimized by the Bayesian method. It is adjusted to become. This model is linear with respect to the parameters shown in Equation (1), and an efficient calculation is realized. On the other hand, kernel functions are non-linear and can be complex functions.

The point to realize the sparse Baysian methodology is to define a hyperparameterized prior distribution based on the model parameter w (set of w _i ) in the following equation (2).

This type of prior distribution favors both the fit and sparseness of the data. In other words, due to the prior distribution, the probability mass of most prior distributions is w _i = 0. This means that in this learning, there are a small number of subjects whose w _i is not 0, and only w _i corresponding to the small number of subjects is used for estimation. By the Bayesian method, the posterior distribution is calculated based on the hyperparameter α _i and the most likely value is output by maximizing the marginal likelihood function. As a result, the posterior distribution based on the set parameters w _i almost to zero is obtained.

In the current implementation, optimization is performed using the software “SparseBays” (version 2.0) installed in MATLAB, where K is calculated based on the linear spline kernel K (X _i , X _t ). The A linear spline kernel K (X _i , X _t ) was proposed by Vapnik et al. For X _i and X _t like the MATLAB code shown in FIG. “LengthScale” of the code shown in FIG. 3 is a normalized parameter, and is determined by a 10-fold cross-validation procedure.

(Modification)
Below, the modification of the learning process by the learning part 20 (refer FIG. 1) of the estimation system 100 is demonstrated.

In the Bayesian kernel method described above, the objective function is expressed as the sum of kernel functions, and only a limited number of subjects are used for estimation. Therefore, the inventors investigated the applicability of a Bayesian generalized additive model instead of the Bayesian kernel method. The Basisian generalized additive model takes each subject's polynomial representation X _i (eg, x _i , x _i ² ,..., X _i ^p ) as input, as shown in equation (3) below, The estimation model is output as the component weight addition.

“X _i ^p ” in Expression (3) represents a term of p-th power of the i-th input beta value. “W _i ^p ” in Equation (3) is an adjustable parameter for the term. This model is called a generalized additive model and is sufficiently effective to represent non-linear functions that do not require interaction terms for input variables. The same optimizations and prior distributions as the Bayesian kernel method are applied to the generalized additive model. Therefore, there are a small number of subjects whose w _i ^p is not zero. The optimization was also performed using the software “Sparse Bayes” (version 2.0) installed in MATLAB.

  Due to normalized parameters and principal component analysis (PCA) dimensions in a sparse kernel (or generalized additive model order), actual BDI score changes and estimated change values Is plotted against the kernel parameters and the number of principal components using a 10-fold cross validation method to determine the optimal set of values. The diffusion coefficient is defined as the average of the absolute value differences between the estimated value Z and the regression line.

[Estimation system 100]
With reference to FIG. 4, an outline of a device configuration of estimation system 100 according to the present embodiment will be described. FIG. 4 is a block diagram illustrating a main hardware configuration of the estimation system 100. As illustrated in FIG. 4, the estimation system 100 includes an fMRI apparatus 200 and an information processing apparatus 400.

  The fMRI apparatus 200 applies a high-frequency electromagnetic field having a resonance frequency toward a region where the brain activity information of the subject S is desired to be acquired, and detects an electromagnetic wave generated by resonance of a specific nucleus (for example, a hydrogen nucleus). Measure activity information. As an example, a “Trio TIM 3T scanner” manufactured by Siemens is used for the fMRI apparatus 200. As the imaging method of the fMRI apparatus 200, for example, an EPI (Echo Planar Imaging) method is used. An image obtained by the EPI method is divided into 2 × 2 × 2 mm voxels and smoothed by an isotropic Gaussian kernel having a full width at half maximum (FWHM) of 6 mm. The data is filtered by high frequency by using the default setting of SPM 8 (cut-off frequency: 128 s).

  The fMRI apparatus 200 includes a static magnetic field magnet 101, a gradient magnetic field coil 102, a transmission unit 103, a transmission coil 104, a reception coil 105, a reception unit 106, a display unit 109, a power source 110, a bed 115, A top plate 116 and a data processing unit 150 are included.

  The data processing unit 150 performs overall control of the estimation system 100. For example, the data processing unit 150 collects brain activity information by driving each unit of the fMRI apparatus 200. The data processing unit 150 may be a dedicated computer or a general-purpose computer that realizes predetermined processing by executing a control program stored in the storage unit 153 or the like. The data processing unit 150 includes a sequence control unit 151, an image reconstruction unit 152, a storage unit 153, a display control unit 154, an interface 155, a display unit 156, and an input unit 157.

  The static magnetic field magnet 101 is a magnet formed in a hollow cylindrical shape, and generates a uniform static magnetic field in an internal space. As the static magnetic field magnet 101, for example, a permanent magnet, a superconducting magnet, or the like is used.

  The gradient magnetic field coil 102 is a coil formed in a hollow cylindrical shape, and is disposed inside the static magnetic field magnet 101. The gradient coil 102 is formed by combining three coils respectively corresponding to x, y, and z axes orthogonal to each other. These three coils are individually supplied with current from the power supply 110 and generate gradient magnetic fields for changing the magnetic field strength along the x, y, and z axes.

  The gradient magnetic fields of the x, y, and z axes generated by the gradient coil 102 correspond to, for example, the gradient magnetic field Gs for slice selection, the gradient magnetic field Gp for phase encoding, and the gradient magnetic field Gr for readout. . The slice selection gradient magnetic field Gs is used to arbitrarily determine the imaging slice plane of the subject S. The phase encoding gradient magnetic field Gp is used to change the phase of the magnetic resonance signal. The gradient magnetic field Gr for readout is used for changing the frequency of the magnetic resonance signal.

  The transmission unit 103 transmits a high frequency signal corresponding to the Larmor frequency to the transmission coil 104 based on the control by the sequence control unit 151. The transmission coil 104 is disposed inside the gradient magnetic field coil 102 and irradiates the subject S with a radio wave output from the transmission unit 103.

  As the receiving coil 105, for example, a coil capable of receiving magnetic resonance signals of a plurality of channels is used. The magnetic resonance signal detected by the receiving coil 105 is transmitted to the receiving unit 106.

  The receiving unit 106 performs A / D (Analog-to-Digital) conversion on the magnetic resonance signal output from the receiving coil 105 based on the control by the sequence control unit 151. A magnetic resonance signal is collected while a gradient magnetic field Gs for slice selection, a gradient magnetic field Gp for phase encoding, and a gradient magnetic field Gr for readout are being applied. Source location information is encoded.

  In estimation system 100 according to the present embodiment, as shown in step S1 of FIG. 1, brain activity information when subject S views image 15 is measured. Therefore, the subject S is configured to be able to see the image 15. As an example, the display unit 109 is provided in the cavity of the fMRI apparatus 200. By displaying the image 15 on the display unit 109, a visual stimulus is given to the subject S.

  A couchtop 116 on which the subject S is placed is attached to the bed 115. The top plate 116 is inserted into the cavity of the gradient coil 102 with the subject S placed thereon. Normally, the bed 115 is installed such that the longitudinal direction is parallel to the central axis of the static magnetic field magnet 101.

  The sequence control unit 151 collects data for generating a brain activity image of the subject S by driving the transmission unit 103, the reception unit 106, and the power source 110 according to the sequence execution data generated by the data processing unit 150. To do. The sequence execution data is information that defines a pulse sequence for collecting raw data from the subject S. More specifically, the sequence execution data includes the strength of the power supplied to the gradient coil 102 by the power supply 110 and the timing of supplying power, the level of the high-frequency signal transmitted from the transmission unit 103 to the transmission coil 104, and the high-frequency signal. This is information defining a procedure for collecting data such as a transmission timing and a timing at which the receiving unit 106 detects a magnetic resonance signal.

  The image reconstruction unit 152 performs a reconstruction process such as Fourier transform on the k-space magnetic resonance signal received by the reception unit 106 to generate a brain activity image of the subject S. The storage unit 153 stores a control program, parameters, image data (such as a three-dimensional model image), other electronic data, and the like for executing processing related to fMRI measurement. The display control unit 154 switches the display of the display unit 109 or switches the display of the display unit 156 through the transmission unit 103.

  The interface 155 is a communication interface for connecting the information processing unit 210 and the information processing apparatus 400 to each other. The information processing unit 210 and the information processing apparatus 400 are connected via an interface 155 wirelessly or by wire. The display unit 156 displays various images and various information related to the region of interest of the subject S on the screen. The input unit 157 receives various operations and information inputs from an operator (not shown).

[Information processing apparatus 400]
With reference to FIG. 5, an example of the hardware configuration of the information processing apparatus 400 (see FIG. 4) constituting the estimation system 100 will be described. FIG. 5 is a block diagram illustrating a main hardware configuration of the information processing apparatus 400.

  As illustrated in FIG. 5, the information processing apparatus 400 includes a ROM (Read Only Memory) 401, a CPU 402, a RAM (Random Access Memory) 403, a network interface 405, a display unit 406, and a storage device 410. Including.

  The ROM 401 stores an operating system, an estimation program executed by the estimation system 100, and the like. The CPU 402 controls the operation of the estimation system 100 by executing various programs such as an operating system and an estimation program of the estimation system 100. The CPU 402 includes the learning unit 20 and the estimation unit 30 described above as functional configurations. The RAM 403 functions as a working memory and temporarily stores various data necessary for program execution.

  The control interface 404 exchanges data with the data processing unit 150 of the fMRI apparatus 200. The network interface 405 exchanges data with an external device (for example, a data server device on the cloud). The control interface 404 and the network interface 405 are configured by arbitrary communication components such as a wired LAN (Local Area Network), a wireless LAN, a USB (Universal Serial Bus), and Bluetooth (registered trademark). The estimation system 100 may be configured to be able to download an estimation program 412 for realizing various processes according to the present embodiment via the network interface 405.

  Display unit 406 includes, for example, a liquid crystal display, an organic EL (Electro Luminescence) display, or other display devices. The display unit 406 may be configured as a touch panel in combination with a touch sensor (not shown). The display unit 406 can display the same image as an image for applying a perceptual stimulus to the subject.

  The storage device 410 is a storage medium such as a hard disk or an external storage device. As an example, the storage device 410 holds teacher data 22 used in the above-described learning process, a learning result 25 in the learning process, an estimation program 412 according to the present embodiment, and the like.

  The estimation program 412 may be provided in a part of an arbitrary program, not as a single program. In this case, imaging according to the present embodiment is realized in cooperation with an arbitrary program. Even such a program that does not include some modules does not depart from the spirit of the estimation system 100 according to the present embodiment. Furthermore, part or all of the functions provided by the estimation program 412 according to the present embodiment may be realized by dedicated hardware. Furthermore, the estimation system 100 and the server may cooperate to realize imaging according to the present embodiment. Furthermore, the estimation system 100 may be configured in the form of a so-called cloud service in which at least one server realizes imaging according to the present embodiment.

[Experiment]
(Experiment contents)
The content of the experiment conducted by the inventors will be described with reference to FIG. The inventors performed a final game (Ultimatum game) on the subjects in order to examine the brain's response to inequality.

  The last game is a game for creating an inequality between two subjects. More specifically, first, one subject (hereinafter also referred to as “first subject”) gives the other subject (hereinafter also referred to as “second subject”) an X circle out of M circles. Propose. The second subject determines whether to accept or reject the first subject's proposal. If the proposal is accepted, the first subject is paid (MX) yen and the second subject is paid M yen. If the proposal is rejected, both the first subject and the second subject cannot get money. The greater the difference in the distribution proposed by the first subject, the more the second subject feels inequality. Such a display of distributions gives a perceptual stimulus with inequality. The distribution difference is digitized as the degree of inequality (see step S2 in FIG. 1).

  FIG. 6 is a conceptual diagram schematically showing the procedure of an ultimate game for examining the brain's response to inequality. In the last game, images 15A to 15E are displayed in order. Images 15A to 15E correspond to the image 15 shown in FIGS.

  As shown in FIG. 6, first, an image 15A indicating the start is displayed. Thereafter, the face and name of the first subject who is the proposer of the distribution money are displayed as an image 15B for one second. The face and name of the proposer is displayed only once for each proposer.

  Next, the distribution to the first subject and the distribution to the second subject are displayed as an image 15C. The total distribution to the first subject and the second subject is, for example, 500 yen. Possible distributions are, for example, “350 yen-150 yen”, “300 yen-200 yen”, “250 yen-250 yen”, “200 yen-300 yen”, “150 yen-350 yen”, “100 yen” Yen-400 Yen "and" 50 Yen-450 Yen ". The distribution selected by the first subject is randomly given noise within ± 25 yen. Thereby, a test subject can be immersed in the last game.

  The image 15C displaying the distribution money is displayed for about 6 seconds. Thereafter, the second subject selects whether to accept or reject the proposal from the first subject by pressing a button. When the second subject accepts the proposal, an image 15D indicating acceptance is displayed. Thereafter, an image 15E indicating the start of the next trial is displayed.

  The time required for one trial is approximately 21 seconds (for example, 19 to 23 seconds), and 56 trials are sequentially performed. That is, the time required for one last game is about 1176 seconds (= 21 seconds × 56 times).

(Experimental result 1)
With reference to FIG. 7 and FIG. 8, the experimental result of estimating the depression tendency from the brain activity pattern obtained when the last game is executed will be described. FIG. 7 is a diagram showing a comparison between an estimated BDI score and an actual BDI score. FIG. 8 is a diagram showing an active region in the brain for inequality.

  The graph (A) in FIG. 7 shows the result of the BDI score estimated from the brain activity pattern obtained when the last game was executed. The X axis of the graph (A) represents an actual measurement value of the change amount of the BDI score. The Y axis of the graph (A) represents the estimated value of the change amount of the BDI score. In the graph (A), the results of the actual measurement value and the estimated value are plotted for each subject. The graph (A) shows a positive correlation between the estimated value and the actually measured value (slope: slope = 0.194, correlation coefficient: R = 0.41, p value: p = 0.00029). This indicates that future depression trends can be estimated from brain activity patterns.

  Such estimation is not possible by presenting the face of the proposer. The result indicating this is shown in graph (B) (slope: slope = 0.046, correlation coefficient: R = 0.24). Also, such an estimation is not possible by presenting the offer. The result indicating this is shown in graph (C) (slope: slope = 0.081, correlation coefficient: R = 0.311).

  These results are robust. This is because the inventors have confirmed that the same result can be obtained by using the Beisian generalized additive model in addition to the kernel Bayesian method.

  As shown in FIG. 8, the inventors have confirmed that the blood flow in the amygdala in the brain is changed due to inequality. A region surrounded by broken lines 701 to 703 is an amygdala. An image (A) in FIG. 8 shows a horizontal cross-sectional view of the subject's brain. An image (B) in FIG. 8 shows a cross-sectional view along the line VIII-VIII of the image (A). As shown in the images (A) and (B), the blood flow volume of the amygdala in the brain changes due to inequality. Therefore, it is preferable to estimate the depression using brain activity information of the amygdala in the brain. In the estimation of the depression, a brain activity pattern of the middle frontal gyrus (MFG) may be used instead of the brain activity information of the amygdala.

(Experimental result 2)
The inventors further examined the part of the amygdala that contributed to predicting changes in depression. Hereinafter, the results will be described with reference to FIG. FIG. 9 is a diagram showing brain activity patterns of the amygdala when inequality is given to the subject.

The inventors performed GLM (Generalized Linear Model) analysis on subjects whose weights w _i output in the learning process are not 0. The inventors of the activity pattern of the amygdala obtained by giving unequal feelings are a subject whose weight w _i is positive (“plus-W” in FIG. 9) and a subject whose weight w _i is negative. ("Minus-W" in FIG. 9). “Plus-W” subjects tend to improve depression. “Minus-W” subjects tend to get worse. In the subject of “plus-W”, the central nucleus 90A of the amygdala and the outer basal nucleus 90B correlate with inequality. On the other hand, in the subject of “plus-W”, the boundary region 90C correlates with inequality.

  The beta values at peak voxels in each region of the amygdala are shown in the experimental results 92A to 92C. As shown in the experimental results 92 </ b> A to 92 </ b> C, the “plus-W” subject and the “minus-W” subject show correlations of patterns opposite to each other with respect to inequality. Our multi-voxel pattern analysis is likely to detect and collect these differences for estimation. The results of this experiment showed that the positive correlation in the central nucleus 90A and the outer basal nucleus 90B and the decrease in the correlation with the unequal feeling in the boundary region 90C increase the depression tendency.

[Summary]
As described above, in the learning process, the estimation system 100 uses the time-series brain activity information measured when a perceptual stimulus that can be correlated with the state of the subject is given to each subject, It acquires about each mutually different area | region. Thereafter, the estimation system 100 determines, for each subject, between each temporal change in brain activity information obtained from each region in the subject's brain and temporal change in sensory stimulation given to the subject. The brain activity pattern X consisting of the respective similarities is calculated. The estimation system 100 learns the correlation between the brain activity pattern X and the state of the subject using the state of each subject that has been quantified in advance and the brain activity pattern X corresponding to each subject.

  Since the brain activity pattern X indicates the degree of interlocking with the perceptual stimulus for each region in the brain, the estimation system 100 can easily identify the portion that reacts to the perceptual stimulus. As a result, the estimation system 100 can accurately identify the active part in the brain even when the type of the perceptual stimulus changes.

  In the estimation process, the estimation system 100 uses the correlation between the brain activity pattern X and the state of the subject (that is, the learning result 25), and the subject from the brain activity pattern X calculated for the subject to be examined. Is estimated.

  Since the estimation system 100 estimates the state of the subject using the brain activity information of the part in the brain that is linked to the perceptual stimulus, the state of the subject can be accurately estimated.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

  1A-1C, 2A-2C, 3A-3C, 5A-5C brain activity image, 11A-11C, 17A-17C brain activity information, 12-14, 18 numerical group, 15, 15A-15E image, 16A, 19A similarity , 20 learning unit, 22 teacher data, 25 learning result, 30 estimation unit, 90A central nucleus, 90B outer basal nucleus, 90C boundary region, 92A to 92C experimental result, 100 estimation system, 101 static magnetic field magnet, 102 gradient magnetic field coil, DESCRIPTION OF SYMBOLS 103 Transmission part, 104 Transmission coil, 105 Reception coil, 106 Reception part, 109,156,406 Display part, 110 Power supply, 115 Bed, 116 Top plate, 150 Data processing part, 151 Sequence control part, 152 Image reconstruction part, 153 storage unit, 154 display control unit, 155 interface, 157 input unit, 2 0 fMRI device 210 processing unit, 400 information processing apparatus, 401 ROM, 402 CPU, 403 RAM, 404 control interface, 405 network interface, 410 storage device, 412 estimation program, 701 broken line.

Claims (4)

An estimation method for estimating a state of a subject,
Obtaining time-series brain activity information measured when a sensory stimulus that can be correlated with the state is given to each subject for each different region in the subject's brain;
For each subject, the degree of similarity between each temporal change in the brain activity information obtained from each region in the subject's brain and the temporal change in the sensory stimulus given to the subject Calculating a similarity group consisting of:
An estimation method comprising: learning a correlation between the similarity group and the state of the subject using the state of each subject that has been quantified in advance and the similarity group corresponding to each subject. .   The estimation method according to claim 1, further comprising estimating a state of the new subject from the similarity group calculated for the new subject using the correlation. Each brain activity information of each subject is represented as a first numerical group of time series,
The sensory stimulus given to each subject is represented as a second numerical group in time series,
The calculating step multiplies each of the numerical values corresponding to the same time in the first numerical value group and the second numerical value group, and adds the multiplied results to calculate the similarity. The estimation method according to claim 1 or 2. An estimation system for estimating a state of a subject,
An acquisition unit that acquires time-series brain activity information measured when a sensory stimulus that can be correlated with the state is given to each subject for each different region in the subject's brain;
For each subject, the degree of similarity between each temporal change in the brain activity information obtained from each region in the subject's brain and the temporal change in the sensory stimulus given to the subject A calculation unit for calculating a similarity group consisting of:
An estimation system comprising a learning unit that uses a state of each subject quantified in advance and the similarity group corresponding to each subject to learn a correlation between the similarity group and the state of the subject.