Estimation system, estimation method, recording medium, estimation model, brain activity training device, brain activity training method, and recording medium storing brain activity training program

Abstract

The estimation system includes: an acquisition unit that acquires measurement data of brain waves and measurement data of functional magnetic resonance imaging simultaneously measured from a subject; a first calculation unit that calculates a first functional connection for each channel combination based on the correlation between channels contained in the measurement data of the brain waves; a second calculation unit that calculates a second functional connection for each intracerebral network based on the correlation between regions of interest contained in the measurement data of functional magnetic resonance imaging; a third calculation unit that uses multiple second functional connections to calculate a score representing the disease tendency of the estimated object to calculate a disease tendency label; and a machine learning unit that determines an estimation model for estimating the disease tendency using a specified first functional connection by performing machine learning using the first functional connection and the disease tendency label of each channel combination.

Description

Technical Field

[0001] The present invention relates to a technique for estimating disease predisposition based on measurement data related to brain activity. Background Art

[0002] Neurofeedback training, which uses functional magnetic resonance imaging (fMRI), a non-invasive method for measuring brain activity, to estimate and regulate brain function, is known. However, neurofeedback training using fMRI alone faces challenges in terms of cost and feasibility.

[0003] Therefore, methods have been proposed that combine electromagnetic field measurement methods such as electroencephalograms (EEGs) or electroencephalograms (EEGs) with fMRI (see, for example, Patent Document 1). In this specification, signal changes (temporal waveforms) measured using EEG are collectively referred to as "brain waves."

[0004] In the methods disclosed in Patent Document 1 and other publications, an estimation model is created using measurement data obtained from simultaneous EEG and fMRI measurements during resting state (hereinafter referred to as "simultaneous EEG / fMRI measurement data"). This estimation model is then used to perform neurofeedback using only the EEG measurement data. EEG offers advantages over other measurement methods in terms of portability, affordability, and widespread availability. Therefore, by adopting the methods disclosed in Patent Document 1 and other publications, costs can be reduced and the feasibility of neurofeedback training can be improved.

[0005] In addition, a technique has been proposed that uses resting-state fMRI measurement data to estimate the activity of individual brain networks and, based on the brain functions expressed by these multiple brain networks, to estimate "disease propensity" (see Non-Patent Document 1, etc.). Estimating "disease propensity" is expected to be applicable to the diagnosis of mental illness, identification of subtypes within the same disease, and selection of treatments.

[0006] Prior art literature

[0007] Patent Literature

[0008] Patent Document 1: Japanese Patent Application Publication No. 2019-093008

[0009] Non-patent literature

[0010] Non-patent literature 1: Andrew T Drysdale et al., “Resting-state connectivity biomarkers define neurophysiological subtypes of depression,” Nature Medicine, Volume 23, Number 1, pp. 28-38 (ISSN: 1546-170X), 2017.1

[0011] Non-Patent Document 2: Takashi Yamada et al., “Resting-State Functional Connectivity-Based Biomarkers and Functional MRI-Based Neurofeedback for Psychiatric Disorders: A Challenge for Developing Theranostic Biomarkers,” International Journal of Neuropsychopharmacology (2017) 20(10), pp. 769-781, July 17, 2017

[0012] Non-Patent Literature 3: Yujiro Yoshihara et al., “Overlapping but asymmetrical relationships between schizophrenia and autism revealed by brain connectivity,” bioRxiv,<URL:https: / / doi.org / 10.1101 / 403212> , September 7, 2018

[0013] Non-patent document 4: Naho Ichikawa et al., “Primary functional brain connections associated with melancholic major depressive disorder and modulation by antidepressants,” Scientific Reports (2020) 10:3542<URL:https: / / doi.org / 10.1038 / s41598-020-60527-z> , 2020

[0014] Non-patent document 5: Eniko Barto'k et al., "Cognitive functions in prepsychotic patients", Progress in Neuro-Psychopharmacology & Biological Psychiatry 29 (2005) 621-625 Summary of the Invention

[0015] Problems to be solved by the invention

[0016] Conventional neurofeedback training targets changes in activity in specific brain regions or specific intra-brain networks (changes in temporal correlations in activity among multiple brain regions) (eg, Non-Patent Document 2).

[0017] A method that can more easily estimate any disease associated with multiple brain networks is desired.

[0018] Solutions for solving problems

[0019] According to a certain embodiment of the present invention, an estimation system includes an acquisition unit that acquires measurement data of brain waves and measurement data of functional magnetic resonance imaging method simultaneously measured from a subject. The measurement data of the brain waves include a time waveform of each of a plurality of channels corresponding to a plurality of sensors arranged on the head of the subject. The estimation system also includes: a first calculation unit that calculates a first functional connection for each channel combination based on the correlation between channels included in the measurement data of the brain waves; a second calculation unit that calculates a second functional connection for each intracerebral network based on the correlation between regions of interest included in the measurement data of functional magnetic resonance imaging method; a third calculation unit that calculates a disease tendency label by calculating a score representing the disease tendency of the estimated object using the plurality of second functional connections; and a machine learning unit that determines an estimation model for estimating the disease tendency using a specified first functional connection by performing machine learning using the first functional connection and the disease tendency label for each channel combination.

[0020] The estimation system may further include an estimation unit configured to input measurement data of the electroencephalogram measured from the subject into the estimation model to estimate the disease tendency of the subject.

[0021] Alternatively, the estimation system further includes a presentation unit that calculates a second score corresponding to the estimated disease tendency of the subject and presents information corresponding to the calculated second score to the subject.

[0022] Alternatively, an estimation model may be prepared for each disease. In this case, an estimation model corresponding to the disease that the subject has may be applied to the subject.

[0023] Changes in the subject's symptoms may be evaluated based on the second score according to the estimated disease tendency of the subject.

[0024] Alternatively, the third calculation unit may calculate the score indicating the disease tendency based on a sum of the plurality of second functional connections corresponding to the disease tendency of the estimation target multiplied by corresponding weighting parameters.

[0025] Alternatively, the third calculation unit may calculate the disease tendency label by normalizing the score indicating the disease tendency and then performing threshold processing.

[0026] The estimation model may include information for selecting a first functional connection used for estimation among the first functional connections of each channel combination, and a weighting parameter corresponding to the selected first functional connection.

[0027] The first calculation unit may calculate the first functional connectivity based on a correlation value between time waveforms within a section included in a window commonly set for the time waveforms of the electroencephalograms of the two target channels.

[0028] The first calculation unit may calculate the first functional connectivity for each frequency band included in the electroencephalogram measurement data and / or for each window size of a set window.

[0029] The estimation system may further include a condition setting unit that determines in advance, based on the subject, a frequency band and / or a window size included in the electroencephalogram measurement data input to the estimation model.

[0030] The second calculation unit may calculate the second functional connectivity based on a correlation value between time waveforms within a section included in a window commonly set for time waveforms representing activity amounts of two target regions of interest.

[0031] According to another embodiment of the present invention, an estimation method includes the steps of acquiring measurement data of brain waves and measurement data of functional magnetic resonance imaging method measured simultaneously from a subject. The measurement data of brain waves include time waveforms of each of a plurality of channels corresponding to a plurality of sensors arranged on the subject's head. The estimation method also includes the following steps: calculating a first functional connection for each channel combination based on the correlation between channels contained in the measurement data of brain waves; calculating a second functional connection for each intra-brain network based on the correlation between regions of interest contained in the measurement data of functional magnetic resonance imaging method; calculating a disease tendency label by calculating a score representing the disease tendency of the estimated object using a plurality of second functional connections; and determining an estimation model for estimating the disease tendency using a prescribed first functional connection by machine learning using the first functional connection and the disease tendency label of each channel combination.

[0032] According to another embodiment of the present invention, a program causes a computer to execute the steps of acquiring measurement data of brain waves and measurement data of functional magnetic resonance imaging method measured simultaneously from a subject. The measurement data of brain waves include time waveforms of each of a plurality of channels corresponding to a plurality of sensors arranged on the subject's head. The program causes the computer to further execute the following steps: calculating a first functional connection for each channel combination based on the correlation between channels included in the measurement data of brain waves; calculating a second functional connection for each intracerebral network based on the correlation between regions of interest included in the measurement data of functional magnetic resonance imaging method; calculating a disease tendency label by calculating a score representing the disease tendency of the estimated object using the plurality of second functional connections; and determining an estimation model for estimating the disease tendency using a specified first functional connection by machine learning using the first functional connection and the disease tendency label for each channel combination.

[0033] According to another embodiment of the present invention, there is provided a learned estimation model for estimating a subject's disease tendency using measurement data of brain waves measured from the subject. The process of constructing the estimation model includes the steps of acquiring measurement data of brain waves and measurement data of functional magnetic resonance imaging method measured simultaneously from the subject. The measurement data of brain waves include the time waveform of each of a plurality of channels corresponding to a plurality of sensors respectively arranged on the subject's head. The process of constructing the estimation model also includes the steps of: calculating a first functional connection for each channel combination based on the correlation between channels contained in the measurement data of brain waves; calculating a second functional connection for each intracerebral network based on the correlation between regions of interest contained in the measurement data of functional magnetic resonance imaging method; calculating a disease tendency label by calculating a score representing the disease tendency of the estimated object using the plurality of second functional connections; and determining the estimation model by machine learning using the first functional connection and the disease tendency label for each channel combination.

[0034] According to another embodiment of the present invention, a brain activity training device for performing neurofeedback training is provided. The brain activity training device includes: a storage device that stores an estimation model for estimating the disease tendency of the subject generated before the neurofeedback training is performed; and an electroencephalogram that is used to measure measurement data of the subject's brain waves during the neurofeedback training. The brain wave measurement data includes a time waveform of each of a plurality of channels corresponding to a plurality of sensors arranged on the subject's head. The brain activity training device also includes: a presentation device; and a processing device that, during the neurofeedback training, uses the estimation model to calculate the disease tendency of the subject based on the measurement data from the electroencephalogram, and outputs a signal for displaying a display corresponding to the disease tendency to the presentation device.

[0035] The estimation model is generated by the following processes: acquiring simultaneously measured EEG data and functional magnetic resonance imaging (fMRI) data from a subject; calculating a first functional connectivity for each channel combination based on inter-channel correlations included in the EEG data; calculating a second functional connectivity for each intracerebral network based on inter-regional correlations included in the fMRI data; calculating a disease propensity label by calculating a score representing the disease propensity of the estimated subject using the plurality of second functional connectivity; and estimating the disease propensity using a predetermined first functional connectivity through machine learning using the first functional connectivity and the disease propensity label for each channel combination, thereby determining the estimation model. The simultaneously measured EEG data includes a time waveform for each channel corresponding to each channel of the EEG data measured during neurofeedback training.

[0036] According to another embodiment of the present invention, a brain activity training method for performing neurofeedback training is provided. The brain activity training method includes the following steps: obtaining an estimation model for estimating the disease tendency of a subject generated before performing neurofeedback training; and measuring measurement data of the subject's brain waves during neurofeedback training. The brain wave measurement data includes a time waveform of each of a plurality of channels corresponding to a plurality of sensors respectively arranged on the subject's head. The brain activity training method also includes the following steps: during neurofeedback training, based on the brain wave measurement data, the estimation model is used to calculate the disease tendency of the subject, and a signal for performing a display corresponding to the disease tendency is output to a presentation device. The step of acquiring an estimation model includes the following steps: acquiring simultaneously measured brain wave data and functional magnetic resonance imaging data from a subject; calculating a first functional connectivity for each channel combination based on inter-channel correlations included in the brain wave measurement data; calculating a second functional connectivity for each intracerebral network based on inter-regional correlations included in the functional magnetic resonance imaging measurement data; calculating a disease propensity label by calculating a score representing the disease propensity of the estimated subject using the plurality of second functional connectivity; and estimating the disease propensity using a predetermined first functional connectivity through machine learning using the first functional connectivity and the disease propensity label for each channel combination, thereby determining an estimation model. The simultaneously measured brain wave measurement data includes a time waveform for each channel corresponding to each channel of the brain wave measurement data measured during neurofeedback training.

[0037] According to another embodiment of the present invention, a brain activity training program for performing neurofeedback training is provided. The brain activity training program causes a computer to execute the following steps: saving an estimation model for estimating a subject's disease tendency generated before performing neurofeedback training; and obtaining measurement data of the subject's brain waves during neurofeedback training. The brain wave measurement data includes a time waveform for each of a plurality of channels corresponding to a plurality of sensors disposed on the subject's head. The brain activity training program causes the computer to further execute the following steps: during neurofeedback training, based on the measurement data of the brain waves, the estimation model is used to calculate the subject's disease tendency, and a signal for displaying a display corresponding to the disease tendency is output to a presentation device. The estimation model is generated by the following processes: acquiring simultaneously measured EEG data and functional magnetic resonance imaging (fMRI) data from a subject; calculating a first functional connectivity for each channel combination based on inter-channel correlations included in the EEG data; calculating a second functional connectivity for each intracerebral network based on inter-regional correlations included in the fMRI data; calculating a disease propensity label by calculating a score representing the disease propensity of the estimated subject using the plurality of second functional connectivity; and estimating the disease propensity using a predetermined first functional connectivity through machine learning using the first functional connectivity and the disease propensity label for each channel combination, thereby determining the estimation model. The simultaneously measured EEG data includes a time waveform for each channel corresponding to each channel of the EEG data measured during neurofeedback training.

[0038] Effects of the Invention

[0039] According to one embodiment of the present invention, it is possible to more easily estimate any disease associated with a plurality of brain networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram showing an outline of the estimation method according to this embodiment.

[0041] Figure 2 It is a schematic diagram showing an outline of the estimation method according to this embodiment.

[0042] Figure 3 Schematic diagram showing a hardware configuration example of a disease tendency estimation system according to the present embodiment.

[0043] Figure 4 1 is a schematic diagram showing a hardware configuration example of a processing device constituting an estimation system that implements the estimation method according to the present embodiment.

[0044] Figure 5 This is a diagram for explaining the process of determining an estimation model in the estimation method according to the present embodiment.

[0045] Figure 6 This is a diagram showing an example of data processing for determining an estimation model in the estimation method according to this embodiment.

[0046] Figure 7 This is a diagram for explaining an overview of an estimation model determined in the estimation method according to the present embodiment.

[0047] Figure 8 : is a flowchart showing the processing procedure of the estimation method according to this embodiment.

[0048] Figure 9 It shows Figure 8 FIG. 1 is a flowchart of a more detailed processing procedure of steps S102 and S104.

[0049] Figure 10 This is a diagram for explaining the outline of preprocessing of EEG measurement data and fMRI measurement data.

[0050] Figure 11 It shows Figure 8 Flowchart of more detailed processing procedures of steps S112 to S118.

[0051] Figure 12 This is a diagram for explaining the outline of the determined estimation model.

[0052] Figure 13 This is a diagram for explaining the outline of neurofeedback training using the estimation method according to this embodiment.

[0053] Figure 14 Schematic diagram showing an implementation example of the estimation method according to this embodiment.

[0054] Figure 15 Schematic diagram showing an example of the functional configuration of a processing device of the estimation system according to the present embodiment.

[0055] Figure 16 Schematic diagram showing an example of the functional configuration of a processing device of the estimation system according to the present embodiment.

[0056] Figure 17 This is a diagram showing an example of evaluation results of feature quantity conditions in the estimation method according to the present embodiment.

[0057] Figure 18 This is a diagram showing an example of the evaluation results of the object specificity of the estimation model determined by the estimation method according to the present embodiment.

[0058] Figure 19 This is a diagram for explaining a method of performing neurofeedback training using an estimation model determined by the estimation method according to this embodiment.

[0059] Figure 20 This is a diagram showing an example of the results of neurofeedback training related to schizophrenia (SCZ).

[0060] Figure 21 : is a figure which shows the example of the result of neurofeedback training related to depression (MDD).

[0061] Figure 22 This is a diagram showing an example of a process for evaluating the long-term effect of neurofeedback training using an estimation model determined by the estimation method according to this embodiment.

[0062] Figure 23 This is a diagram showing an example of the long-term effects of neurofeedback training on major depressive disorder (MDD).

[0063] Figure 24 This is a diagram showing an example of the long-term effects of neurofeedback training on schizophrenia (SCZ).

[0064] Figure 25 This graph shows the effects of neurofeedback training on schizophrenia (SCZ) compared with a comparison group.

[0065] Figure 26 This is another diagram showing the effect of neurofeedback training on schizophrenia (SCZ) compared with a comparison group.

[0066] Figure 27 1 is a diagram showing an experimental example for evaluating the specificity of the effect of neurofeedback training.

[0067] Figure 28 This is a diagram showing another experimental example for evaluating the specificity of the effect of neurofeedback training. DETAILED DESCRIPTION

[0068] The embodiments of the present invention will be described in detail with reference to the accompanying drawings. The same or corresponding parts in the drawings are denoted by the same reference numerals, and their description will not be repeated.

[0069] [A. Summary]

[0070] First, an overview of the estimation method according to this embodiment will be described. Figure 1 and Figure 2 : is a schematic diagram showing an overview of the estimation method according to this embodiment. Figure 1 The outline of the process of determining the estimation model (learning phase) is shown in Figure 2 hereinafter shows an overview of a process for estimating disease tendency (estimation stage) using the determined estimation model.

[0071] In this specification, the term "disease" is a term that includes not only pathological symptoms that occur in humans, but also any mental or physical symptoms that differ from the state exhibited by ordinary people. Symptoms that appear in this situation are also referred to as "disease-like symptoms." "Disease tendency" is a term that includes the possibility (probability) that a subject has symptoms corresponding to the target "disease" as well as the possibility (probability) that a subject will develop symptoms corresponding to the target "disease" as a subject.

[0072] On the other hand, in this specification, the term "estimation model" is not limited to estimating these probabilities; it also includes estimating the probability (the degree of the difference) that a healthy person is in a state of brain activity that differs from a standard "healthy brain activity state" by a predetermined degree or more. In other words, the term "estimation model" may also be used to estimate the relative state of brain activity.

[0073] In this specification, "functional connectivity" is a term encompassing indicators that indicate the degree of functional connectivity between brain regions. "Functional connectivity" can be calculated using any method using data measured using any measurement method. Except where specific measurement data and calculation methods are explicitly stated, the method for calculating "functional connectivity" is not limited in this specification.

[0074] Reference Figure 1 In the estimation method of this embodiment, first, EEG and fMRI are performed simultaneously on the same subject in a resting state to obtain EEG / fMRI simultaneous measurement data ((1) EEG / fMRI simultaneous measurement). At this time, the data obtained by EEG (hereinafter also referred to as "EEG measurement data") and the data obtained by fMRI (hereinafter also referred to as "fMRI measurement data") represent the same brain activity of the same subject. That is, the EEG / fMRI simultaneous measurement data includes measurement data of brain waves measured simultaneously from the subject (EEG measurement data) and measurement data using functional magnetic resonance imaging (fMRI measurement data).

[0075] Each sensor is typically composed of a pair of electrodes. Each sensor is also called a channel, and EEG measurement data is equivalent to multi-channel brain waves. That is, EEG measurement data includes the time waveform of each of the multiple channels corresponding to the multiple sensors configured on the subject's head. Based on the EEG measurement data, the functional connectivity in each frequency band is calculated ((2) Calculation of functional connectivity (FC)). The functional connectivity will also be referred to as "FC" (Functional Connectivity) below.

[0076] The estimation method according to this embodiment is not limited to EEG, which measures the voltage generated by brain electrical activity. Magnetoencephalography (MEG), which measures the fluctuating magnetic field generated by brain electrical activity, can also be used. For ease of explanation, the following description will primarily focus on an example using EEG measurement data.

[0077] fMRI measurement data is used to estimate a subject's disease predisposition based on specific brain networks. Brain networks, also known as resting-state networks (RSNs), are a collective term for characteristic brain activity patterns derived from signal sources belonging to a single brain region or from the collaboration of signal sources belonging to multiple spatially separated brain regions. Resting-state fMRI is primarily used to define brain networks.

[0078] Specifically, as resting-state networks, there are seven known networks: (1) Control Network (CON), (2) Dorsal Attention Network (DAN), (3) Default Mode Network (DMN), (4) Limbic System (LIM), (5) Somatomotor Network (SMN), (6) Ventral Attention Network (VAN), and (7) Visual Network (VIS).

[0079] In addition, (1) the control network (CON) is sometimes also called the frontal parietal network, and (6) the ventral attention network (VAN) is sometimes also called the saliency network.

[0080] Furthermore, the aforementioned resting-state networks are sometimes divided into several subnetworks. More specifically, (1) the control network (CON) is divided into three subnetworks, (3) the default mode network (DMN) is divided into four subnetworks, and all other networks are divided into two subnetworks.

[0081] The disease tendency of the subject is considered to be able to be estimated based on one or more specific brain networks known in advance for each disease. Therefore, in the estimation method according to this embodiment, the disease tendency of the subject is estimated based on one or more specific brain networks ((3) Estimation of disease tendency based on multiple brain networks). In the following description, since the output is the result of threshold processing (binarization as an example) of the disease tendency, the estimation result is also called a "disease tendency label" (label). The disease tendency label takes any one of multiple values ​​(levels).

[0082] Finally, based on the dynamic functional connectivity in each frequency band and the subject's disease tendency, an estimation model for estimating the subject's disease tendency by inputting EEG measurement data is determined ((4) Determining the estimation model). The estimation model is equivalent to a learned model.

[0083] Reference Figure 2 EEG measurement data obtained from the subject via EEG is input into the determined estimation model 10, thereby outputting an estimated result of the subject's disease propensity. The estimated result of the subject's disease propensity can be used for neurofeedback training (hereinafter also referred to as "training"), etc. As described later, the estimation model 10 also includes a function for selecting information suitable for estimating disease propensity from the EEG measurement data obtained from the subject via EEG.

[0084] By using this estimation model 10, the subject's disease propensity can be estimated sequentially, enabling, for example, low-cost neurofeedback. As will be described later, because the estimation model is subject-specific, an estimation model is prepared for each disease. Furthermore, the estimation model corresponding to the disease present in the subject is applied.

[0085] [B. Example of hardware configuration of estimation system]

[0086] Next, a hardware configuration example of an estimation system for realizing the estimation method according to the present embodiment will be described.

[0087] Figure 3 Schematic diagram showing an example of the hardware configuration of the disease tendency estimation system 1 according to this embodiment. Figure 3 , the estimation system 1 includes a processing device 100 , an EEG device 200 and an fMRI device 300 .

[0088] Processing device 100 acquires brainwave measurement data (EEG measurement data) and functional magnetic resonance imaging measurement data (fMRI measurement data) simultaneously measured from the subject. More specifically, processing device 100 receives EEG measurement data measured by EEG device 200 and fMRI measurement data measured by fMRI device 300 to determine an estimation model for estimating disease propensity.

[0089] The EEG device 200 detects signals (electrical signals) representing electroencephalograms generated by a plurality of sensors 220 arranged on the head of a subject S. The EEG device 200 includes a multiplexer 202 , a noise filter 204 , an A / D (Analog to Digital) converter 206 , a storage unit 208 , and an interface 210 .

[0090] The multiplexer 202 sequentially selects a group of cables 222 connected to the plurality of sensors 220, and electrically connects the selected group of cables to the noise filter 204. The noise filter 204 is a filter for removing noise, such as a high-frequency cutoff filter, and removes noise components contained in the signal (electrical signal) representing the electroencephalogram generated between the group of cables corresponding to the selected channel.

[0091] A / D converter 206 samples the electrical signal (analog signal) output from noise filter 204 at a predetermined period and outputs it as a digital signal. Storage unit 208 sequentially stores the time-series data (digital signal) output from A / D converter 206 in association with the selected channel and timing information (e.g., time or count value).

[0092] In response to access from the processing device 100 or the like, the interface 210 outputs the time-series data representing the electroencephalogram stored in the storage unit 208 to the processing device 100 .

[0093] On the other hand, the fMRI device 300 applies a high-frequency electromagnetic field of a resonant frequency to the area of ​​the subject S from which information on brain activity is desired (hereinafter also referred to as the "region of interest"), thereby detecting electromagnetic waves generated from specific atomic nuclei (e.g., hydrogen nuclei) due to resonance, thereby measuring brain activity.

[0094] The fMRI apparatus 300 includes a magnetic field applying mechanism 310 , a receiving coil 302 , a driving unit 320 , and a data processing unit 350 .

[0095] The magnetic field applying mechanism 310 applies controlled magnetic fields (static magnetic field and gradient magnetic field) to the region of interest of the subject S and irradiates RF (Radio Frequency) pulses. More specifically, the magnetic field applying mechanism 310 includes a static magnetic field generating coil 312, a gradient magnetic field generating coil 314, an RF irradiation unit 316, and a bed 318 for placing the subject S within a bore.

[0096] The drive unit 320 is connected to the magnetic field applying mechanism 310 and controls the magnetic field applied to the subject S and the transmission and reception of RF pulse waves. More specifically, the drive unit 320 includes a static magnetic field power supply 322 , a gradient magnetic field power supply 324 , a signal transmitter 326 , a signal receiver 328 , and a bedding driver 330 .

[0097] exist Figure 3 In FIG, the central axis of the cylindrical cavity for placing the subject S is defined as the Z axis, the horizontal direction perpendicular to the Z axis is defined as the X axis, and the vertical direction perpendicular to the Z axis is defined as the Y axis.

[0098] The static magnetic field generating coil 312 is a spiral coil wound around the Z axis that generates a static magnetic field in the Z axis direction within the cavity. The gradient magnetic field generating coil 314 includes an X coil, a Y coil, and a Z coil (not shown) that generate gradient magnetic fields in the X axis direction, the Y axis direction, and the Z axis direction within the cavity, respectively. The RF irradiation unit 316 irradiates the region of interest of the subject S with RF pulses based on the high frequency signal transmitted from the signal transmission unit 326 according to the control sequence. Figure 3 , the RF irradiation unit 316 is shown as an example of a structure in which the RF irradiation unit 316 is built into the magnetic field application mechanism 310 . However, the RF irradiation unit 316 may be provided on the bedding 318 side, or the RF irradiation unit 316 may be integrated with the receiving coil 302 .

[0099] The receiving coil 302 receives electromagnetic waves (NMR signals) emitted from the subject S and outputs an analog signal. The analog signal output from the receiving coil 302 is amplified and A / D converted by the signal receiving unit 328 and then output to the data processing unit 350. The receiving coil 302 is preferably placed close to the subject S to enable high-sensitivity detection of the NMR signal.

[0100] The data processing unit 350 sets a control sequence for the driving unit 320 and outputs a plurality of brain activity pattern images indicating the degree of activity in the brain as information indicating brain activity based on the NMR signals received by the receiving coil 302 .

[0101] The data processing unit 350 includes a control unit 351, an input unit 352, a display unit 353, a storage unit 354, a display control unit 355, an image processing unit 356, a data collection unit 357, and an interface 358. The data processing unit 350 may be a dedicated computer or a general-purpose computer that executes a control program stored in the storage unit 354 or the like to perform predetermined processing.

[0102] The control unit 351 controls the operation of each functional unit, such as generating a control sequence for driving the drive unit 320. The input unit 352 receives various operations and information input by an operator (not shown). The display unit 353 displays various images and various information related to the region of interest of the subject S on the screen. The storage unit 354 stores control programs, parameters, image data (three-dimensional model images, etc.), other electronic data, etc. for executing fMRI-related processing. The image processing unit 356 generates a plurality of brain activity pattern images based on the data of the detected NMR signals. The interface 358 exchanges various signals with the drive unit 320. The data collection unit 357 collects data consisting of a group of NMR signals originating from the region of interest.

[0103] Figure 4 1 is a schematic diagram showing an example of a hardware configuration of a processing device 100 constituting an estimation system 1 that implements the estimation method according to this embodiment. The processing device 100 can typically be a computer that complies with a general-purpose architecture. Figure 4 The processing device 100 includes a processor 102 , a main storage unit 104 , a control interface 106 , a network interface 108 , an input unit 110 , a display unit 112 , and a secondary storage unit 120 as main components.

[0104] The processor 102 is composed of arithmetic processing circuits such as a CPU (Central Processing Unit) and a GPU (Graphical Processing Unit). It implements the various functions described below by executing the code contained in various programs stored in the secondary storage unit 120 in a specified order. The main storage unit 104 is composed of DRAM (Dynamic Random Access Memory) and other memory devices, and stores the code of the programs executed by the processor 102 and various working data required for program execution.

[0105] The processing device 100 has a communication function, which is mainly provided by a control interface 106 and a network interface 108 .

[0106] The control interface 106 exchanges data with the data processing unit 350 of the fMRI apparatus 300. The network interface 108 exchanges data with external devices (e.g., cloud-based data servers). The control interface 106 and the network interface 108 may be implemented using any communication module, such as a wired LAN (Local Area Network), a wireless LAN, a USB (Universal Serial Bus), or Bluetooth (registered trademark).

[0107] The input unit 110 is typically composed of a mouse or keyboard, and receives user operations. The display unit 112 is typically composed of a display, and notifies the user of various information related to the execution status of the processing in the processing device 100 and the operation.

[0108] The secondary storage unit 120 is typically composed of a hard disk or an SSD (Solid State Drive), and stores various programs executed by the processor 102, various data required for processing, set values, and the like. More specifically, the secondary storage unit 120 stores EEG measurement data 20, fMRI measurement data 30, an estimation model determination program 121, an estimation program 122, and estimation model parameters 124.

[0109] [C. Decision Processing of Estimation Model]

[0110] Next, a description will be given of a process for determining an estimation model in the estimation method according to this embodiment.

[0111] Figure 5 This is a diagram for explaining the process of determining an estimation model in the estimation method according to the present embodiment. Figure 6 This is a diagram showing an example of data processing for determining an estimation model in the estimation method according to this embodiment.

[0112] The process for calculating FC as an explanatory variable is performed on the EEG measurement data 20 included in the EEG / fMRI simultaneous measurement data, and the process for calculating a disease tendency label as an explained variable is performed on the fMRI measurement data 30 included in the EEG / fMRI simultaneous measurement data. Figure 5 and Figure 6 These processes will be described below.

[0113] (c1: EEG measurement data 20)

[0114] The EEG measurement data 20 is a collection of signal changes (time waveforms) representing brain waves measured for each channel (sensor). Figure 6(1) Preprocessing corresponds to the conversion into a time waveform 22 of the power of each frequency band. The time waveform 22 of the power is a waveform obtained by sequentially calculating the average value of the square of the amplitude of the corresponding frequency component included in the EEG measurement data 20 for each unit time.

[0115] As a more specific preprocessing, the EEG measurement data 20 (time waveform) is frequency-converted to calculate the amplitude of each frequency. Then, one or more frequencies within a specified frequency band are selected, and the amplitudes of the selected one or more frequencies are squared and averaged to calculate power.

[0116] Furthermore, if the sampling frequency of the EEG measurement data 20 is high, the data may be downsampled to a predetermined sampling frequency before frequency analysis. For example, downsampling may be performed so that the sampling frequency becomes 1 / TR [Hz] to correspond to the irradiation period (TR: repetition time) of the fMRI RF pulse.

[0117] Thus, preprocessing is performed on the EEG measurement data 20 to generate a power time waveform 22 corresponding to the number of channels N × the number of frequency bands M. Examples of frequency bands include theta waves (θ waves: 4 Hz-8 Hz), alpha waves (α waves: 8 Hz-12 Hz), low beta waves (low beta waves: 12 Hz-20 Hz), and high beta waves (high beta waves: 20 Hz-30 Hz). For example, if the number of channels N is 63 and the number of frequency bands M is 4, 252 (= 63 × 4) power time waveforms 22 are generated.

[0118] Next, for the same frequency band, the time correlation of the time waveform 22 of the power between different channels is calculated (with Figure 6 (2) corresponds to the calculation time correlation).

[0119] In this specification, “temporal correlation” refers to correlation values ​​between time waveforms and the temporal waveform of the correlation values ​​within a section included in a window 26 commonly set for a plurality of time waveforms.

[0120] For EEG measurement data 20, the temporal correlation between the two power time waveforms 22 is calculated. Temporal correlation refers to the temporal waveform of the correlation value of interest within the temporal width of a window 26 set for the two power time waveforms 22. Specifically, functional connectivity (FC) is calculated based on the correlation values ​​between the time waveforms within the interval encompassed by the window 26 set for the temporal waveforms of the two target channels.

[0121] Window 26 has a predetermined window size (time width). By sequentially moving the set position of window 26 (the time interval from the start time to the end time) by a step size, and sequentially calculating the correlation corresponding to each set position of window 26, a time waveform 24 of the EEG time correlation can be calculated. The calculated time waveform 24 of the EEG time correlation corresponds to the FC.

[0122] In this manner, the functional connectivity (FC) is calculated for each channel combination (each channel pair) based on the correlation between channels included in the EEG measurement data 20 .

[0123] For example, if the number of EEG channels N is 63, then the time waveform 24 of the EEG time correlation can be calculated for each of 1953 (=N×(N-1) / 2=63×(63-1) / 2) channel combinations (channel pairs). Furthermore, the time waveform 24 of the EEG time correlation has a time length corresponding to the number of shifts (time steps) of the window 26.

[0124] The time waveform 24 of the EEG time correlation is calculated for each frequency band. In other words, the time waveform 24 of the EEG time correlation for M frequency bands is generated.

[0125] Furthermore, the window size (time width) of the set window 26 may be varied to calculate the time waveform 24 of the EEG time correlation for each window.

[0126] As such, the time waveform 24 of the EEG time correlation may be used as a feature quantity for estimating disease tendency by varying the channel pair, frequency band, and window size. In this case, the time waveform 24 of the EEG time correlation is output as a vector having the dimensions of the channel combination (channel pair) x (the number of time steps corresponding to the window 26) for each frequency band and / or window size.

[0127] In this manner, the functional connectivity (FC) may be calculated for each frequency band included in the EEG measurement data 20 and / or each window size of the set window 26 .

[0128] Alternatively, a single vector combining all frequency bands and window sizes can be generated. In this case, the vector is output as {(number of channel combinations (channel pairs)) × (number of window sizes) × (number of frequency bands)} × (number of time steps). That is, in the above example, as the time waveform 24 of the EEG temporal correlation, a vector of 1953 dimensions × (number of time steps) dimensions can be output for each frequency band and / or each window size, or a vector of a higher dimension can be output by combining these.

[0129] (c2: fMRI measurement data 30)

[0130] fMRI measurement data 30 (i.e., brain activity pattern images) is a collection of brain activity pattern images acquired during each RF pulse irradiation cycle. It is known which brain regions correspond to brain activity in each known brain network (resting-state network). These one or more regions corresponding to each brain network are called "regions of interest" (ROIs).

[0131] In the estimation method according to this embodiment, it is assumed that the activity of each brain network is defined by a combination of two ROIs.

[0132] First, the fMRI measurement data 30 are preprocessed (with Figure 6 BOLD signal 32 is calculated for each ROI using the (1′) preprocessing. The BOLD signal refers to the temporal variation in activity of each ROI that depends on blood oxygen concentration. More specifically, during the preprocessing of fMRI measurement data 30, the BOLD signal is calculated based on image features corresponding to the ROI contained in the brain activity pattern image.

[0133] Next, the temporal correlation of the BOLD signal 32 (with Figure 6 (2') corresponds to the calculation of temporal correlation. For fMRI measurement data 30, the temporal correlation between the two BOLD signals 32 is calculated. Temporal correlation refers to the temporal waveform of the correlation value of interest within the temporal width of a window 26 set for the two BOLD signals 32. Specifically, functional connectivity (FC') is calculated based on the correlation values ​​between the temporal waveforms within the interval encompassed by the window 26 commonly set for the temporal waveforms representing the activity levels of the two target ROIs.

[0134] Window 26 has a predetermined window size (time width). By sequentially moving the set position of window 26 (the time interval from the start time to the end time) by a step size, and sequentially calculating the correlation corresponding to each set position of window 26, a time waveform 34 of the BOLD temporal correlation can be calculated. The time waveform 34 of the BOLD temporal correlation represents the temporal variation of the correlation value and corresponds to the functional connectivity (FC').

[0135] The time waveform 34 of the BOLD temporal correlation can be calculated for each ROI combination, that is, the target intracerebral network. In this way, based on the correlation between ROIs included in the fMRI measurement data 30, the functional connectivity (FC') is calculated for each intracerebral network.

[0136] also, Figure 5The word "dynamic" in this context means that a value is calculated for each window being watched, and "static" means that a single value is calculated for the entire period. Figure 5 The “static FC” in the figure refers to the correlation value during the entire period (single functional connectivity).

[0137] The time waveform 34 (FC') of the BOLD temporal correlation calculated in this way is used to estimate the disease tendency (with Figure 6 Previous studies have shown that disease propensity is associated with multiple brain networks (i.e., brain activities in multiple ROIs).

[0138] The estimation method according to this embodiment uses this prior information and the temporal waveforms 34 of BOLD temporal correlations corresponding to multiple intracerebral networks associated with the disease predisposition of the estimated subject to calculate WLS 36, a score representing the disease predisposition of the estimated subject. More specifically, WLS 36 is calculated by multiplying the temporal waveforms 34 (FC') of the multiple BOLD temporal correlations of interest by corresponding weighting parameters and then adding them together. This calculation method is known as the WLS (Weighted Linear Summation) method.

[0139] In this manner, WLS 36 , which is a score indicating disease tendency, is calculated based on the sum of a plurality of functional connections (time waveforms 34 of BOLD temporal correlation) corresponding to the disease tendency of the estimation target multiplied by corresponding weighting parameters.

[0140] In this manner, a score (WLS 36 ) indicating the estimated disease tendency of the subject is calculated using a plurality of BOLD temporal correlation time waveforms 34 (FC′), thereby calculating a disease tendency label 38 .

[0141] Then, the disease tendency score (WLS 36) is normalized and then thresholded to calculate a disease tendency label 38. For example, if binarization is used as the thresholding process, the disease tendency label 38 is represented as "0" indicating health or "1" indicating disease.

[0142] For estimating the disease propensity of the subject, a disease propensity label 38 is output as a vector of one dimension x (the number of windows 26 set) dimensions. The disease propensity label 38 is an explained variable.

[0143] (c3: Estimation Model Decision Process)

[0144] The estimation model specifies the relationship between the time waveform 24 (FC) of the EEG time correlation as an explanatory variable and the disease tendency label 38 (label) as an explained variable. In the estimation method according to this embodiment, a feature quantity suitable for estimating the disease tendency label 38 is selected from the feature quantities contained in the time waveform 24 of the EEG time correlation as a multidimensional vector. In the estimation stage, the disease tendency is estimated using the information of the selected feature quantity (the time correlation calculated in sequence). That is, the estimation model for estimating the disease tendency is determined using the time waveform 24 (FC) of the EEG time correlation for each channel combination (each channel pair) through machine learning using the disease tendency label 38 and the time waveform 24 (FC) of the EEG time correlation.

[0145] By using only a portion of the feature quantities included in the time waveform 24 of the EEG temporal correlation as an explanatory variable in estimation, the dimension can be compressed and reduced, thereby reducing the amount of calculations related to the estimation and speeding up the estimation process.

[0146] Any machine learning algorithm can be used to determine the estimated model, but SLR (Sparse Logistic Regression) can be adopted as an example.

[0147] That is, the time waveform 24 of the EEG time correlation is input as an explanatory variable to the SLR (with Figure 6 (3) corresponds to the SLR input), and the disease tendency label 38 is input as the explained variable to the SLR as the machine learning algorithm (corresponding to Figure 6 (4') corresponds to the SLR input). Then, through machine learning, a feature quantity suitable for estimating the disease tendency label 38 is selected.

[0148] Figure 7 This is a diagram for explaining the outline of the estimation model determined in the estimation method according to this embodiment. Figure 7 The time waveform 24 of the EEG time correlation is a feature quantity group corresponding to the number of EEG channels in each frequency band. Although not shown, the time waveform 24 of the EEG time correlation is also calculated for each window size.

[0149] By machine learning, a predetermined number (for example, 30) of feature quantities F suitable for estimating the disease tendency label 38 are selected from a large number of feature quantities of the multidimensional vector constituting the time waveform 24 of the EEG time correlation. i (i=1, 2,…, x).

[0150] Furthermore, the weighting parameter W may be determined for each selected feature. i(i=1, 2, ..., x). For example, the more suitable the feature quantity F is for estimating the disease tendency label 38, the i The weighting parameter W i Set to a larger value.

[0151] Alternatively, instead of selecting the feature quantity and comparing it with the selected feature quantity F i The corresponding weight parameter W i This method only determines the weighting parameter W i For example, by setting the weighting parameter W to i Setting it to zero will give the same result as not making any selection.

[0152] Thus, the determined estimation model includes the time waveform 24 (feature amount F) of the EEG time correlation used for selection of the time waveform 24 (FC) of the EEG time correlation for each channel combination (each channel pair). i ) and the weighting parameter W corresponding to the time waveform 24 of the selected EEG time correlation. i .

[0153] In the estimation phase, the subject's disease tendency is sequentially estimated based on only the EEG measurement data 20 using the feature quantities and corresponding weighting parameters determined through the above-described process.

[0154] (c4: Processing)

[0155] Figure 8 : is a flowchart showing the processing procedure of the estimation method according to this embodiment. Figure 8 Some of the steps shown may also be implemented by executing a program in the processing device 100 .

[0156] Reference Figure 8 First, EEG and fMRI are measured simultaneously to obtain EEG measurement data 20 and fMRI measurement data 30 (step S100). That is, the processing device 100 obtains brain wave measurement data (EEG measurement data 20) and fMRI measurement data (fMRI measurement data 30) measured simultaneously from the subject.

[0157] The processing device 100 preprocesses the acquired EEG measurement data 20 to calculate the time waveform of the power for each frequency band (step S102). Next, the processing device 100 uses the calculated time waveform of the power to calculate the time waveform of the EEG temporal correlation for each window size (step S104). Specifically, the processing device 100 calculates functional connectivity (FC) for each channel combination based on the inter-channel correlations contained in the EEG measurement data 20.

[0158] In parallel with the processing of steps S102 and S104, or after step S104, the processing device 100 preprocesses the acquired fMRI measurement data 30 to calculate the temporal waveform of the BOLD signal for each ROI constituting the intracerebral network (step S112). Next, the processing device 100 uses the calculated temporal waveform of the BOLD signal to calculate the temporal waveform of the BOLD temporal correlation for each window size (step S114). Specifically, the processing device 100 calculates functional connectivity (FC') for each intracerebral network based on the correlation between the ROIs included in the fMRI measurement data 30.

[0159] Next, the processing device 100 selects a BOLD temporal correlation waveform corresponding to the estimated subject's disease propensity from the calculated BOLD temporal correlation waveforms, multiplies each selected BOLD temporal correlation waveform by the corresponding weighting parameter, and adds them together to calculate the WLS (step S116). The processing device 100 then normalizes the calculated WLS and binarizes it to calculate a disease propensity signature representing the disease propensity (step S118). Specifically, the processing device 100 calculates a score (WLS) representing the estimated subject's disease propensity using multiple functional connectivity (FC') lines, thereby calculating the disease propensity signature.

[0160] Finally, processing device 100 performs machine learning using the temporal waveform of the EEG temporal correlation and the disease propensity label to determine feature quantities and weighting parameters for estimating the disease propensity label (step S120). Specifically, processing device 100 performs machine learning using functional connectivity (FC) and the disease propensity label for each channel combination to determine an estimation model for estimating the disease propensity using a predetermined functional connectivity (FC).

[0161] An estimation model can be determined through such a process.

[0162] [D.EEG / fMRI simultaneous measurement]

[0163] Then, Figure 1 "(1) EEG / fMRI simultaneous measurement" and Figure 8 The subject S is placed in the chamber of the fMRI apparatus 300 with the sensor worn on his head. Figure 3 The illustrated estimation system 1 performs EEG and fMRI in parallel.

[0164] The processing device 100 stores the measurement data from the EEG device 200 and the fMRI device 300 in association with each other based on the same time. By associating the measurement data based on the same time, the EEG measurement data 20 and the fMRI measurement data 30 can be acquired with the same time axis.

[0165] [E. Calculation of functional connectivity (FC) based on EEG measurement data 20]

[0166] Next, describe in detail Figure 1 "(2) Computational Functional Connection (FC)" shown and Figure 8 Steps S102 to S104 are shown.

[0167] First, as preprocessing for the EEG measurement data 20 (time waveform), the time waveform is subjected to frequency conversion. For example, a fast Fourier transform can be used as the frequency conversion process. Furthermore, the fast Fourier transform is not limited to the fast Fourier transform; a Hilbert transform, a discrete Fourier transform, or the like can also be used.

[0168] As Figure 8 In the pre-processing of step S102, frequency domain data (the relationship between frequency and amplitude) is calculated by frequency transforming the EEG measurement data 20. The power of each target frequency band is calculated by calculating the average of the squared amplitude values ​​of the frequencies included in the frequency band.

[0169] exist Figure 8 In step S104, any two channels are selected, the windows are moved sequentially along the time axis, and the correlation values ​​between the time waveforms of the powers within the windows are calculated sequentially.

[0170] Through these processes, the time waveform 24 (FC) of the EEG time correlation can be calculated.

[0171] Figure 9 It shows Figure 8 Refer to the flowchart of the more detailed processing procedures of steps S102 and S104. Figure 9 The processing device 100 selects a channel included in the acquired EEG measurement data 20 (step S1021), selects a moment as the object for power calculation (step S1022), and performs a fast Fourier transform on the time waveform included in the window with the selected moment as the reference position (step S1023).

[0172] Alternatively, the time waveform included in the window may be moved and averaged along the time axis before being subjected to a fast Fourier transform. Applying such a moving average can reduce high-frequency noise components.

[0173] Next, the processing device 100 selects a frequency band for power calculation (step S1024), calculates the average of the squared amplitude values ​​of the frequencies included in the selected frequency band (step S1025), and then stores the average of the squared amplitude values ​​in association with the selected time and the selected frequency band (step S1026).

[0174] The processing device 100 determines whether all frequency bands have been selected (step S1027). If all frequency bands have not been selected ("No" in step S1027), the processing from step S1024 onward is repeated.

[0175] If all frequency bands have been selected ("Yes" in step S1027), the processing device 100 determines whether all times have been selected (step S1028). If not all times have been selected ("No" in step S1028), the processing from step S1022 onwards is repeated.

[0176] If all time points have been selected ("Yes" in step S1028), the processing device 100 determines whether all channels have been selected (step S1029). If not all channels have been selected ("No" in step S1029), the processing from step S1021 onwards is repeated.

[0177] If all channels have been selected ("YES" in step S1029), the calculation process of the time waveform 22 of the power of each frequency band is completed at this stage. Then, the calculation process of the time waveform 24 (FC) of the EEG time correlation is performed next.

[0178] The processing device 100 selects a window setting (window size and step size) for calculating the EEG time correlation (step S1041 ), and selects a frequency band for calculating the EEG time correlation (step S1042 ).

[0179] Furthermore, regarding the window setting (window size and step length), a plurality of combinations may be prepared in advance, or only one combination may be used.

[0180] Furthermore, the processing device 100 selects a channel combination for which the EEG temporal correlation is to be calculated (step S1043 ).

[0181] Next, the processing device 100 selects a time for calculating the EEG temporal correlation (step S1044), extracts the time waveform of the power contained in a window with the selected time as the reference position for the two channels corresponding to the selected channel combination (step S1045), and calculates the correlation value of the extracted time waveform of the power (step S1046). The processing device 100 then stores the calculated correlation value in association with the selected time, channel combination, frequency band, and window settings (step S1047).

[0182] The processing device 100 determines whether all times have been selected (step S1048). If all times have not been selected ("No" in step S1048), the processing from step S1044 onward is repeated.

[0183] If all time points have been selected ("Yes" in step S1048), the processing device 100 determines whether all channel combinations have been selected (step S1049). If not, the processing from step S1043 onwards is repeated.

[0184] If all channel combinations have been selected ("Yes" in step S1049), the processing device 100 determines whether all frequency bands have been selected (step S1050). If not all frequency bands have been selected ("No" in step S1050), the processing from step S1042 onwards is repeated.

[0185] If all frequency bands have been selected ("Yes" in step S1050), processing device 100 determines whether all window settings have been selected (step S1051). If not, processing from step S1041 onwards is repeated.

[0186] If all window settings have been selected ("YES" in step S1051), the calculation process of the time waveform 24 (FC) of the EEG time correlation is completed at this stage.

[0187] [F. Processing of Estimating Disease Tendency Based on fMRI Measurement Data 30]

[0188] Next, describe in detail Figure 1 "(3) Estimating disease tendency based on multiple brain networks" and Figure 8 Steps S112 to S118 are shown.

[0189] First, in preprocessing (step S112) of fMRI measurement data 30 (brain activity pattern image), BOLD signal 32 for each ROI is calculated from the brain activity pattern image. In extracting this BOLD signal 32, processing is performed to compensate for time delays occurring in fMRI.

[0190] More specifically, when the BOLD signal 32 representing the neural state of the ROI being studied is set to s(t) and the hemodynamic response function (HRF) is set to h(t), the transformation y(t) of the BOLD signal representing the activity level of each ROI is equivalent to adding an error e(t) to the convolution of s(t) and h(t) as shown in the following equation (1).

[0191] [Number 1]

[0192]

[0193] Here, HRF(t) depends on the irradiation period TR of the RF pulse of fMRI.

[0194] The estimated value of the brain state s^(t) can be expressed as the following equation (2) using the Wiener filter d(t).

[0195] [Number 2]

[0196]

[0197] Here, when H(x), Y(x), E(x), and D(x) are set to the Fourier transform of h(t), y(t), e(t), and d(t), the estimated value of the brain state s^(t) can be expressed as the following formula (3).

[0198] [Number 3]

[0199]

[0200] The estimated brain state value s^(t) shown in equation (3) above corresponds to the BOLD signal. Specifically, the estimated brain state value s^(t) is estimated by deconvolving the observed y(t) with the HRF. Deconvolution with the HRF compensates for the time delay (measurement point offset) between the EEG measurement data 20 and the fMRI measurement data 30.

[0201] Figure 10 This is a diagram for explaining the outline of pre-processing of the EEG measurement data 20 and the fMRI measurement data 30 .

[0202] Reference Figure 10The window 26 is sequentially moved and set relative to the EEG measurement data 20 in steps, and the EEG time correlation is calculated for each window 26 , thereby calculating the time waveform 24 of the EEG time correlation.

[0203] Meanwhile, fMRI measurement data 30 is deconvolved using the HRF to compensate for the time delay associated with RF pulse irradiation, and then a BOLD signal 32 is calculated. Specifically, deconvolution using the HRF allows the time axis of EEG measurement data 20 to be substantially aligned with the time axis of BOLD signal 32. A disease propensity signature 38 is then calculated using BOLD signal 32.

[0204] exist Figure 8 In step S114, for each combination of two ROIs, the window is sequentially moved along the time axis, and the correlation values ​​between the time waveforms of the BOLD signals within the window are sequentially calculated. In addition, the combination of the two ROIs may also be a combination of the same ROI.

[0205] exist Figure 8 In steps S116 and S118, the WLS is calculated by multiplying the time waveforms of the multiple BOLD time correlations corresponding to the disease tendency of the estimated object by the corresponding weighting parameters and then adding them. Then, the calculated WLS is normalized and binarized to calculate the disease tendency label representing the disease tendency. More specifically, the k-th BOLD time correlation time waveform 34 (FC'(k)) and the corresponding weighting parameter W can be used. FC (k) to calculate WLS 36 as shown in equation (4) below.

[0206] WLS=∑FC'(k)×W FC (k)…(4)

[0207] WLS is a score that is centered around 0 and shows a larger value as the degree of disease tendency increases. WLS can be normalized to a probability p according to the following formula (5).

[0208] p=1 / (1+exp(-WLS))…(5)

[0209] The probability p (0≤p≤1) is bounded by 0.5 and is closer to 1 as the degree of disease tendency increases.

[0210] The functional connectivity (FC') of interest is selected based on the estimated disease propensity of the subject. For example, Non-Patent Document 3 discloses a disease discriminator for schizophrenia (SCZ) that uses 16 functional connectivity (FC'). Furthermore, Non-Patent Document 4 discloses a disease discriminator for melancholic depression (MDD) that uses 10 functional connectivity (FC').

[0211] By referring to these prior arts, a plurality of functional connections (FC') can be selected according to the estimated disease tendency of the subject, and weighting parameters W corresponding to the selected functional connections (FC') can be multiplied by calculation. FC The sum of the obtained values ​​determines WLS.

[0212] Finally, by binarizing the probability p, the disease tendency label can be calculated.

[0213] Figure 11 It shows Figure 8 Detailed processing flow chart of steps S112 to S118 is shown in FIG. Figure 11 The processing device 100 selects a ROI for BOLD signal calculation (step S1121), extracts activity from the image feature values ​​of the regions corresponding to the selected ROIs from the fMRI measurement data 30 (step S1122). The temporal changes in the extracted activity are deconvolved with the HRF to calculate the temporal waveform of the BOLD signal (step S1123), which is then associated with the selected ROI and stored (step S1124).

[0214] The processing device 100 determines whether all ROIs have been selected (step S1125 ). If not all ROIs have been selected (“No” in step S1125 ), the processing from step S1121 onward is repeated.

[0215] If all ROIs have been selected ("Yes" in step S1125), the calculation process of the BOLD signal for each ROI is completed at this stage. Then, the calculation process of the time waveform 34 of the BOLD time correlation is performed next.

[0216] The processing device 100 selects a ROI combination as a target for calculating the BOLD temporal correlation (step S1141 ).

[0217] Next, the processing device 100 selects a time for calculating the BOLD temporal correlation (step S1142), extracts the temporal waveform of the BOLD signal 32 within a window with the selected time as the reference position for the two ROIs corresponding to the selected ROI combination (step S1143), and calculates the correlation value of the extracted BOLD signal temporal waveform (step S1144). The processing device 100 then associates the calculated correlation value with the selected time and ROI combination and stores it (step S1145).

[0218] The processing device 100 determines whether all times have been selected (step S1146 ). If all times have not been selected (“No” in step S1146 ), the processing from step S1142 onward is repeated.

[0219] If all time points have been selected ("Yes" in step S1146), the processing device 100 determines whether all ROI combinations have been selected (step S1147). If not all ROI combinations have been selected ("No" in step S1147), the processing from step S1141 onwards is repeated.

[0220] If all ROI combinations have been selected ("Yes" in step S1147), the calculation process of the time waveform 34 of the BOLD temporal correlation is completed at this stage. Then, the calculation process of WLS is performed next.

[0221] The processing device 100 selects the disease predisposition of the estimated subject (step S1161) and determines multiple intracerebral networks (ROI combinations) associated with the selected disease predisposition (step S1162). Furthermore, the processing device 100 determines weighting parameters corresponding to each of the multiple determined intracerebral networks (step S1163). The processing device 100 then multiplies the time waveform 34 of the BOLD temporal correlation of each of the multiple determined intracerebral networks by the corresponding weighting parameters and then calculates the sum (step S1164). The calculated sum is the WLS corresponding to the estimated disease predisposition of the subject.

[0222] The processing device 100 determines whether all disease tendencies have been selected (step S1165). If all disease tendencies have not been selected ("No" in step S1165), the processing from step S1161 onward is repeated.

[0223] If all disease tendencies have been selected ("Yes" in step S1165), the calculation process of the WLS for each disease tendency is completed at this stage. Then, the calculation process of the disease tendency label is performed next.

[0224] The processing device normalizes the calculated WLS to calculate probability p (step S1181), performs threshold processing on the calculated probability p, and outputs a column of values ​​0 or 1 (step S1182). The output column of values ​​0 or 1 is a disease tendency label indicating the disease tendency.

[0225] [G. Modeling Process]

[0226] Next, describe in detail Figure 1 "(4) Determine the estimation model" shown and Figure 8 Step S120 is shown.

[0227] In the decision process of the estimation model, by performing machine learning on the relationship between the time waveform 24 (FC) of the EEG time correlation as the explanatory variable and the disease tendency label 38 as the explained variable, a specified number of feature quantities (for example, 30) suitable for estimating the disease tendency label 38 and the corresponding weighting parameters are determined from the large number of feature quantities of the multidimensional vector contained in the time waveform 24 of the EEG time correlation.

[0228] As such a machine learning method, any machine learning algorithm can be used, and the case where SLR is used as an example will be described. The specific processing procedure of SLR will be described below.

[0229] As a linear discriminant function that discriminates between two classes S1 and S2 by the weighted sum of each feature quantity, the following equation (6) is assumed.

[0230] [Number 4]

[0231]

[0232] Here, x is a feature quantity in a D-dimensional space (x=(x1, x2, ..., x D ) t ∈R D ), θ is a weighted vector containing bias terms (θ=(θ1,θ2,…,θ D ) t ). The hyperplane corresponding to f(x;θ)=0 determines the boundary between class S1 and class S2.

[0233] In SLR, the logistic function shown in formula (7) is used to calculate the probability that each feature value belongs to class S2 for the hyperplane that determines the boundary between class S1 and class S2.

[0234] [Number 5]

[0235]

[0236] Here, probability p ranges from 0 to 1. When f(x; θ) = 0 (on the hyperplane), probability p is 0. When f(x; θ) is at positive or negative infinity (a position far from the hyperplane), probability p is 1. In other words, probability p represents the likelihood that any feature value x belongs to class S2.

[0237] Here, when an arbitrary binary output variable y is introduced (y=0 corresponds to class S1, y=1 corresponds to class S2), it is possible to generate a data sequence {(x1, y1), (x2, y2), ..., (x N ,y N )}Define the probability function shown in the following formula (8).

[0238] [Number 6]

[0239]

[0240]

[0241] Each item in formula (8) represents the probability p of the nth sample n (In y n =1, p n , in y n =0, 1-p n )(Refer to formula (9)), therefore, the product of the terms shown in formula (8) means the probability of all samples included in the data sequence.

[0242] For the purpose of machine learning, a probability function l(θ) as shown in equation (10) is introduced, and a weight vector θ that maximizes the value of the probability function l(θ) defined by equation (10) is searched.

[0243] [Number 7]

[0244]

[0245] The probability function l(θ) contains the probability p of a nonlinear element that depends on the weight vector θ n Therefore, in the solution of the probability function l(θ), the gradient and the Hessian matrix can be used. If the weight vector θ(=(θ1, θ2, ..., θ) that maximizes the probability function l(θ) can be determined, D ) t ), it is possible to select a portion of the feature quantity as a feature quantity suitable for estimation based on the value of the element (weighting parameter) included in the weighting vector θ.

[0246] For example, a predetermined number (for example, 30) of weighting parameters are selected in descending order from the weighting parameter having the largest magnitude, and feature quantities corresponding to the selected weighting parameters are selected.

[0247] In this way, the feature amount (specification of the channel pair, frequency band, and window size used in the EEG time correlation) used to estimate the disease tendency label 38 and the corresponding weighting parameter can be determined.

[0248] Figure 12 This is a diagram for explaining the outline of the determined estimation model. Figure 12 In the estimation stage, the time waveform 24 of the EEG temporal correlation is input to the estimation model 10. More specifically, the amount of information of a predetermined window size (e.g., 30 seconds) is input at each step (e.g., 30 seconds).

[0249] Estimation model 10 includes multiple combinations of feature quantity information 11 and weighting parameters 12. Only the information corresponding to feature quantity information 11 included in estimation model 10 (EEG time correlation selected as the feature quantity) in the input EEG time correlation time waveform 24 is used. The used EEG time correlation is then multiplied by the corresponding weighting parameter 12, and the sum of the results is calculated in adder 13. The sum is then binarized to 0 or 1 by binarizer 14. The binarized result is output as a disease tendency.

[0250] Furthermore, in determining the estimation model, the channel pair, frequency band, and window size are all considered as variable factors, but the frequency band and window size may be determined in advance as feature quantity conditions.

[0251] More specifically, as a method for determining feature quantity conditions (frequency band and window size), simultaneous EEG / fMRI measurement data can be acquired across multiple sessions and, through methods such as cross-validation, the frequency band and window size that yield the highest discrimination performance (e.g., an indicator represented by the AUC (Area Under the Curve)) can be determined. By predetermining the feature quantity conditions in this way, the amount of computation required to determine the estimation model can be reduced.

[0252] As described above, the frequency band and / or window size included in the EEG measurement data 20 input to the estimation model may be determined in advance according to the subject.

[0253] [H. Estimation phase]

[0254] Next, a description will be given of a processing example in the estimation phase using the estimation model determined by the processing in the learning phase as described above.

[0255] In the estimation phase, EEG measurement data measured from the subject is input into the estimation model to estimate the subject's disease tendency. A typical application example of this estimation phase is neurofeedback training.

[0256] Figure 13 : is a diagram for explaining the outline of neurofeedback training using the estimation method according to this embodiment. Figure 13 The brain activity training device 2 for performing neurofeedback training includes an EEG device 200 , a storage device 502 , a display device 510 , and a processing device 500 .

[0257] The storage device 502 stores the estimated model. The estimated model stored in the storage device 502 is generated before the neurofeedback training is performed. In addition, the storage device 502 can also be implemented using the memory included in the processing device 500, or can also be implemented using Figure 14 The server device 400 shown is implemented.

[0258] The display device 510 is an example of a presentation device, and provides visual and / or auditory information to the user.

[0259] The EEG device 200 is equivalent to an electroencephalogram, and measures the brainwave data of the subject S during neurofeedback training. Figure 1 The illustrated EEG device 200 similarly includes time waveforms for each of the multiple channels corresponding to the multiple sensors disposed on the head of the subject S. That is, substantially the same EEG device 200 is used when generating the estimation model and when performing neurofeedback training. Therefore, the EEG measurement data included in the EEG / fMRI simultaneous measurement data used when generating the estimation model includes time waveforms for each channel corresponding to the respective channels of the EEG measurement data measured during neurofeedback training.

[0260] The processing device 500 acquires EEG measurement data from the subject S via EEG and estimates disease propensity using a predetermined estimation model. The disease propensity is estimated for each cycle (typically, for each step length). The processing device 500 calculates a score corresponding to the estimated disease propensity and provides a score display 520 corresponding to the calculated score on the display device 510. In this way, the processing device 500 calculates a score corresponding to the disease propensity of the subject S using the estimation model based on the measurement data from the EEG device 200, and presents information corresponding to the calculated score to the subject. That is, the processing device 500 outputs a signal for displaying a display corresponding to the disease propensity to the display device 510.

[0261] In addition, the processing device 500 may also be implemented by executing a brain activity training program on a general-purpose computer.

[0262] For example, the score display 520 includes a reference circle 522 and a score circle 524 whose size varies according to the score. The size of the score circle 524 is sequentially updated according to the disease tendency estimated based on the EEG measurement data measured from the subject S.

[0263] Subject S is informed in advance that a reward will be earned by moving score circle 524 closer to or further from reference circle 522. Subject S, either on their own or in response to external instructions, consciously uses their brain to perform calculations, associations, or meditation to adjust the size of score circle 524 in the designated direction. This conscious use of their brain allows for alleviation or treatment of the target disease.

[0264] In neurofeedback training using the estimation method according to this embodiment, disease predisposition can be estimated at any location as long as the estimation model 10 and EEG measurement data 20 are available. Taking advantage of this, for example, neurofeedback training can be conducted at any location after simultaneous EEG and fMRI measurements have been performed using dedicated equipment.

[0265] Figure 14 Schematic diagram showing an example of implementation of the estimation method according to this embodiment. Figure 14 For example, at a dedicated measurement station, EEG and fMRI measurements are performed simultaneously on each subject, and the processing device 100 determines an estimation model 10 for each subject. The determined estimation model 10 is transmitted from the measurement station to the server device 400 .

[0266] The server device 400 stores subject data 402 including an estimation model for each subject.

[0267] In addition to the measurement station, the server device 400 is accessed from the treatment center desired by each subject in one or more treatment centers to obtain the estimation model corresponding to each subject. Furthermore, a processing device 500 as described later is provided in each treatment center to perform the following operations based on the obtained estimation model. Figure 13 Neurofeedback training shown.

[0268] By using Figure 14 The system shown can reduce the implementation cost of neurofeedback training.

[0269] [I. Functional structure]

[0270] Next, an example of the functional configuration of devices included in the estimation system 1 that implements the estimation method according to the present embodiment will be described.

[0271] (i1: processing device 100)

[0272] Figure 15 1 is a schematic diagram showing an example of the functional configuration of the processing device 100 of the estimation system 1 according to this embodiment. Figure 15 Each of the functions shown is implemented by the processor 102 of the processing device 100 executing the estimation model determination program.

[0273] Furthermore, the estimation model determination program 121 may be executed by one or more processors included in the processing device 100, or by a plurality of processing devices in cooperation with each other. In the latter case, a plurality of computers arranged on a network, known as a so-called cloud system, may be used. Furthermore, instead of a structure implemented by a processor executing a program (software implementation), all or part of the structure may be implemented using a hard-wired structure such as an FPGA (Field-Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit).

[0274] Each program according to the present embodiment may be installed in a manner that utilizes functions provided by an OS (Operating System). Even in such a case, the program is included in the scope of protection of the present invention.

[0275] Reference Figure 15 The processing device 100 includes pre-processing modules 150 and 160 , time correlation calculation modules 152 and 162 , a WLS calculation module 164 , a binarization module 166 and a model estimation module 168 .

[0276] The pre-processing module 150 transforms the EEG measurement data 20 into a power time waveform 22. The power time waveform 22 can also be calculated for each frequency band and / or each window size.

[0277] The time correlation calculation module 152 calculates the time waveform 24 of the EEG time correlation from the time waveform 22 of the power for each channel combination (channel pair).

[0278] The pre-processing module 160 calculates the BOLD signal 32 of each ROI based on the fMRI measurement data 30 .

[0279] The time correlation calculation module 162 calculates a time waveform 34 (FC′) of the BOLD time correlation based on the BOLD signal 32 of each ROI.

[0280] The WLS calculation module 164 calculates a WLS 36 as a score indicating the disease tendency of the estimated subject using the time waveform 34 of the BOLD temporal correlation corresponding to a plurality of intracerebral networks associated with the disease tendency of the estimated subject.

[0281] After normalizing the WLS 36 , the binarization module 166 calculates a disease tendency label 38 (label) as a result of binarizing the disease tendency.

[0282] The model estimation module 168 determines feature values ​​and weighting parameters for estimating the disease propensity label 38 based on the EEG temporal correlation time waveform 24 and the disease propensity label 38. The determined feature value and weighting parameter set is output as the estimation model 10.

[0283] (i2: processing device 500)

[0284] Then, Figure 13 and Figure 14 The hardware structure of the processing device 500 is similar to that of the above-mentioned Figure 4 The hardware structure of the processing device 100 shown is the same, so the detailed description will not be repeated.

[0285] Figure 16 1 is a schematic diagram illustrating an example of a functional configuration of the processing device 500 of the estimation system 1 according to the present embodiment. Figure 16 The functions shown are executed by the processor of the processing device 500 (with Figure 4 The estimation procedure 122 shown is also implemented).

[0286] The estimation program may be executed using one or more processors included in the processing device 500, or it may be executed by multiple processing devices in a collaborative manner. In the latter case, multiple computers configured on a network, known as a so-called cloud system, may be used. Furthermore, instead of implementing the program by having a processor execute the program (software implementation), all or part of the program may be implemented using a hardwired structure such as an FPGA or ASIC.

[0287] Each program according to the present embodiment may be installed in a manner that utilizes functions provided by the OS. Even in such a case, the program is included in the scope of protection of the present invention.

[0288] Reference Figure 16 The processing device 500 includes a pre-processing module 550 , a time correlation calculation module 552 , a weighted sum calculation module 554 , a binarization module 556 , an estimation model acquisition module 558 and a display control module 560 .

[0289] The pre-processing module 550 transforms the EEG measurement data 20 into a power time waveform 22. The power time waveform 22 can also be calculated for each frequency band and / or each window size.

[0290] The time correlation calculation module 552 calculates the time waveform 24 of the EEG time correlation from the time waveform 22 of the power for each channel combination (channel pair).

[0291] The estimation model acquisition module 558 acquires the estimation model 10 corresponding to the subject from the server device 400 or the like. The estimation model 10 includes a set of feature quantities and weighting parameters for estimating the disease tendency label 38 .

[0292] The weighted sum calculation module 554 selects one or more feature quantities (EEG time correlation) as the object in the time waveform 24 of the EEG time correlation according to the estimation model 10 acquired by the estimation model acquisition module 558, and calculates the sum of the values ​​obtained by multiplying by the corresponding weighting parameters as WLS 36.

[0293] The binarization module 556 calculates a disease tendency (0 or 1) as a result of binarizing the disease tendency after normalizing the WLS 36 .

[0294] The display control module 560 calculates a score based on the disease tendency values ​​sequentially output from the binarization module 556, and calculates a score display 520 for display on the display device 510. In this way, changes in the subject's symptoms are evaluated based on the score corresponding to the estimated disease tendency of the subject.

[0295] [J. Example]

[0296] Next, some results obtained by applying the estimation method according to this embodiment to actual subjects will be described.

[0297] In the following examples, healthy subjects or subjects judged to be subclinical are used as subjects. Subclinical refers to a state in which the subject is judged to be highly prone to exhibit at least some of the symptoms based on the responses to inquiries for evaluating the severity of the symptoms of the disease in question.

[0298] (j1: Feature quantity condition)

[0299] First, an example of the result of evaluating the estimation accuracy of the feature quantity conditions for determining the estimation model will be described.

[0300] For each subject, EEG and fMRI were performed simultaneously during resting state to acquire EEG / fMRI simultaneous measurement data. EEG / fMRI simultaneous measurement data were acquired for each subject for at least eight sessions (each session lasting less than 5 minutes). Schizophrenia (SCZ) (see Non-Patent Document 3) and major depressive disorder (MDD) (see Non-Patent Document 4) were considered as target diseases.

[0301] The EEG / fMRI data from seven of the eight simultaneous EEG / fMRI sessions were used to determine the estimation model. The EEG / fMRI data from the remaining session was then used as validation data to evaluate the estimation performance using leave-one-out cross-validation (LOOCV). The mean AUC was used as the evaluation metric for the estimation performance.

[0302] Figure 17 FIG. 1 is a diagram showing an example of the evaluation results of the feature quantity conditions in the estimation method according to this embodiment. Figure 17 , the change in estimation performance when the window size for calculating the temporal correlation is changed to different values ​​(8TR, 12TR, 16TR, 20TR, and 24TR). TR represents the irradiation period of the RF pulse.

[0303] exist Figure 17 (A) shows an example of the evaluation results when schizophrenia (SCZ) is set as the target disease. Figure 17 (B) shows an example of the evaluation result when depression (MDD) is set as the target disease.

[0304] like Figure 17 (A) and Figure 17 As shown in (B), by changing the window size, the average value and the degree of variation of the average AUC also change.

[0305] like Figure 17 As shown in (A), for schizophrenia (SCZ), when the window size is 20TR as the feature quantity condition, the average AUC is good overall. However, when the window size is 24TR, some subjects show the highest average AUC.

[0306] In addition, if Figure 17As shown in (B), the deviation is smaller for depression (MDD) than for schizophrenia (SCZ). In addition, for depression (MDD), it can be seen that when the window size as a feature quantity condition is 20TR or 24TR, the overall average of the average AUC is good.

[0307] like Figure 17 As shown, it can be seen that it is preferable to select the optimal feature value condition for each subject.

[0308] (j2: Estimate the subject specificity of the model)

[0309] Next, the subject specificity of the estimation model will be described.

[0310] In the above Figure 17 In this report, evaluation results of the performance of estimating a score indicating disease propensity for schizophrenia (SCZ) using an estimation model determined based on EEG measurement data for schizophrenia (SCZ) (hereinafter referred to as the "schizophrenia estimation model") are presented. Furthermore, evaluation results of the performance of estimating a score indicating disease propensity for major depressive disorder (MDD) using an estimation model determined based on EEG measurement data for major depressive disorder (MDD) (hereinafter referred to as the "depression estimation model") are presented. Below, examples of the results of a cross-evaluation of each model are presented.

[0311] Figure 18 This is a diagram showing an example of the evaluation results of the object specificity of the estimation model determined by the estimation method according to the present embodiment.

[0312] exist Figure 18 (A) shows the estimation performance (average AUC) when the schizophrenia estimation model is used to estimate the disease tendency of schizophrenia (SCZ) and the estimation performance when the disease tendency of major depressive disorder (MDD) is estimated. Figure 18 (B) shows the estimation performance when the depression estimation model is used to estimate the disease tendency of major depressive disorder (MDD) and the estimation performance when the disease tendency of schizophrenia (SCZ) is estimated.

[0313] like Figure 18 As shown in (A), the schizophrenia estimation model shows specific estimation performance for the estimation of disease tendency of schizophrenia (SCZ). Figure 18 As shown in (B), the depression estimation model shows specific estimation performance for the estimation of disease tendency of major depressive disorder (MDD).

[0314] according to Figure 18 (A) and Figure 18The verification results of the estimation accuracy of the cross-method for each disease tendency shown in (B) show that the estimation model determined by the estimation method according to this embodiment has subject specificity.

[0315] (j3: Neurofeedback training)

[0316] Next, an example of performing neurofeedback training using an estimation model determined by the estimation method according to this embodiment will be described.

[0317] Figure 19 : is a diagram for explaining a method of performing neurofeedback training using an estimation model determined by the estimation method according to this embodiment. Figure 19 As a training schedule (one day's worth), a group 614 consisting of a plurality of blocks 612 is performed over multiple consecutive days. Each block 612 includes a plurality of loops 600. Each loop 600 includes a series of processes consisting of an interval 602 (time T1), an induction period 604 (time T2), and a display period 606 (time T3).

[0318] Interval 602 corresponds to the rest period between the previous cycle 600. Guidance period 604 corresponds to the period during which the subject, either by himself or in accordance with external instructions, consciously uses his brain to perform calculations, associations, meditation, etc., in order to achieve a higher evaluation score. Display period 606 corresponds to the period during which the score calculated for the subject during guidance period 604 is displayed.

[0319] During the induction period 604, the EEG measurement data obtained from the subject is used to estimate the subject's disease propensity. The estimation of the subject's disease propensity may be repeated multiple times during the induction period 604. Since the disease propensity estimation results (0 or 1) are calculated multiple times, these results can be averaged to calculate a score representing the degree of the subject's disease propensity during the induction period 604.

[0320] For example, if a model is used where "0" represents health, a smaller score representing the subject's disease propensity is considered preferable. A score display 520 corresponding to the calculated score is provided to the subject. In score display 520, the smaller the score, the closer score circle 524 is to reference circle 522.

[0321] The subjects are rewarded with money or other rewards according to their scores. Motivated by these rewards, they try to get higher scores.

[0322] The time T1 of the interval 602 is set to about 5 seconds, for example. The time T2 of the guidance period 604 is set to about 50 to 70 seconds, for example. The time T3 of the display period 606 is set to about 5 seconds, for example.

[0323] In the following examples, two target diseases are assumed: schizophrenia (SCZ) (see Non-Patent Document 3) and major depressive disorder (MDD) (see Non-Patent Document 4).

[0324] The sampling frequency of EEG was set to 500 Hz, and the EEG measurement data was processed to remove artifacts (subject-specific independent components extracted in advance).

[0325] The length (window size) of EEG measurement data used to estimate disease propensity is set to an integer multiple of the fMRI RF pulse irradiation period TR. More specifically, for schizophrenia (SCZ), it is set to 16 TR (2.45 seconds × 16 = 39.2 seconds), and for major depressive disorder (MDD), it is set to 20 TR (2.45 seconds × 20 = 49 seconds).

[0326] Accordingly, the time T2 of the induction period 604 for schizophrenia (SCZ) is set to 70 seconds, and the time T2 of the induction period 604 for depression (MDD) is set to 85 seconds.

[0327] Figure 20 : is a diagram showing an example of the results of neurofeedback training related to schizophrenia (SCZ). Figure 20 (A) shows an experimental example of WLS as a score representing the disease tendency of the estimated object before and after training. Figure 20 (B) shows an experimental example of the Schizotypal Personality Questionnaire (SPQ) before and after training. SPQ is an example of a schizophrenia-like score. Figure 20 (C) shows an experimental example of the N-back task before and after training.

[0328] Figure 20 (A) and Figure 20 "A" to "I" in (B) represent subjects. Figure 20 (A) shows the WLS and Figure 20 The SPQ values ​​shown in (B) all indicate that smaller values ​​mean greater improvement in symptoms. Figure 20 The WLS shown in (A) showed no significant results, but Figure 20 The SPQ shown in (B) shows a tendency to improve through training.

[0329] about Figure 20 The N-back task shown in (C) is a test that evaluates the ability (cognitive function) to remember information presented N times before. The results of the N-back task are expressed as a score called "d prime". The larger the d prime value, the greater the improvement in cognitive function. Figure 20(C) shows the results for N=2, 3, and 4. Figure 20 In the N-back task (C), training showed an improvement trend. In particular, a significant change was confirmed in the paired t-test for the 4-back task (4-backtest).

[0330] Figure 21 : is a diagram showing an example of the results of neurofeedback training related to depression (MDD). Figure 21 (A) shows an experimental example of WLS as a score representing the disease tendency of the estimated object before and after training. Figure 21 (B) shows an experimental example of the Beck Depression Inventory (BDI) and the Self-Rating Depressive Scale (SDS) before and after training. BDI and SDS are examples of depression-like symptom scores. Figure 21 (C) shows an experimental example of the results of the N-back task before and after training.

[0331] Figure 21 (A) and Figure 21 "A" to "G" in (B) represent subjects. Figure 21 (A) shows the WLS and Figure 21 The BDI and SDS shown in (B) indicate that the smaller the value, the greater the improvement in symptoms. Figure 21 (A) and Figure 21 The results shown in (B) show that there is a tendency for improvement through training. Figure 21 In the result example shown in (C), regarding the N-back task, an improvement trend was also observed through training.

[0332] (j4: Long-term effects of neurofeedback training)

[0333] Next, an example of evaluation of long-term effects including the results of a follow-up survey (follow-up: FU) one to two months after the completion of neurofeedback training will be described.

[0334] Figure 22 This is a diagram showing an example of a process for evaluating the long-term effect of neurofeedback training using an estimation model determined by the estimation method according to this embodiment.

[0335] Reference Figure 22 Training was performed over 3 days, and measurements (score calculation) were performed on the day before training (Pre-training), the day after training (Post-training), and the follow-up survey day 1 to 2 months after training (Follow-up survey: FU).

[0336] Figure 23This is a diagram showing an example of the long-term effects of neurofeedback training on major depressive disorder (MDD). Figure 23 (A) shows an experimental example of WLS. Figure 23 (B) shows an experimental example of BDI. Figure 23 (C) shows an experimental example of RRS as a score indicating the frequency of ruminative thinking. The smaller the value shown by RRS, the better the state can be determined.

[0337] exist Figure 23 (A)~ Figure 23 In (C), the change in the score of each subject is shown in the form of a line graph, and the change in the average score of the entire subject is shown in the form of a bar graph.

[0338] about Figure 23 The WLS shown in (A) has large individual differences among subjects, but overall it can be said to be meaningful as a score indicating disease tendency.

[0339] about Figure 23 The BDI shown in (B) maintained a reduced state immediately after training (Post) and 1 to 2 months later (FU), indicating that the effect of training lasted for a long time. Figure 23 (B) also shows the sub-scores used to calculate BDI. The sub-scores also show the same tendency as BDI.

[0340] about Figure 23 The RRS shown in (C) maintained a reduced state both immediately after training (Post) and 1 to 2 months later (FU), indicating that the effect of training lasted for a long time. Figure 23 (C) also shows the sub-scores used to calculate the RRS. The sub-scores also show the same tendency as the RRS.

[0341] Figure 24 This is a diagram showing an example of the long-term effects of neurofeedback training on schizophrenia (SCZ). Figure 24 In FIG, the change of the WLS of each subject is shown in the form of a line graph, and the change of the score obtained by averaging the WLS of all subjects is shown in the form of a bar graph.

[0342] exist Figure 24 In the example, "CTRL" indicates the results of the comparison group. The comparison group refers to a group of subjects who were trained using pre-prepared information from other people as feedback, rather than information from the subject. In other words, the comparison group experiment shows that the subjects were trained based on their own brain activity, even though their own brain activity was not referenced. This also applies to the following experimental examples.

[0343] about Figure 24 The WLS shown shows a significant difference from the comparison group, although there are large individual differences among the subjects, and there is a tendency for improvement through training.

[0344] (j5: Effects of neurofeedback training)

[0345] Next, an example of the effect of neurofeedback training using the comparison group as a reference will be described.

[0346] Figure 25 This is a graph showing the effects of neurofeedback training on schizophrenia (SCZ) in comparison with a comparison group. Figure 25 The vertical axis of the graph shown represents the change in value before and after neurofeedback (Post-Pre).

[0347] It can be seen that the comparison group (CTRL) is distributed around the point where the value before and after training does not change (the value of the vertical axis is zero), while the group that has been properly trained is distributed around the point on the negative side (i.e., the value of SPQ becomes smaller after training). Figure 25 The sub-item scores used to calculate the SPQ are also shown in . The sub-item scores also showed the same tendency as the SPQ.

[0348] about Figure 25 The SPQ shown also showed significant differences compared to the comparison group, showing a tendency to improve with training.

[0349] Figure 26 This is another graph showing the effect of neurofeedback training on schizophrenia (SCZ) compared with a control group. Figure 26 (A)~ Figure 26 The vertical axis of the graph shown in each figure of (D) represents the change in the value before and after neurofeedback (Post-Pre). Figure 26 (A)~ Figure 26 (D) shows experimental examples of scores related to cognitive functions.

[0350] More specifically, in Figure 26 (A) shows an experimental example of the N-back task (N=2). Figure 26 (B) shows an experimental example of the N-back task (N=4). In each experimental example, it can be seen that the comparison group (CTRL) is distributed centered around the point where the value before and after training did not change (the value on the vertical axis is zero), while the appropriately trained group is distributed centered around the point on the positive side (i.e., the value of d prime increased after training).

[0351] exist Figure 26(C) and Figure 26 (D) shows an example of cognitive function evaluation using the CANTAB (Cambridge Neuropsychological Test Automated Battery) (see Non-Patent Document 5, etc.). More specifically, sustained attention (RVP: rapid visual information processing) is evaluated. Sustained attention has been reported to be impaired in patients with schizophrenia.

[0352] As scores, A' and p(Hit) are output. Larger values ​​for both A' and p(Hit) are preferred. These scores were calculated using CANTAB(R) [Cognitive Assessment Software]. Cambridge Cognition (2019). All rights reserved. www.cantab.com.

[0353] It can be seen from any experimental example that the comparison object group (CTRL) is distributed centered around the point where the value before and after training does not change (the value of the vertical axis is zero). In contrast, the group that has been properly trained is distributed centered around the point on the positive side (that is, the values ​​of A' and p(Hit) both increase after training).

[0354] like Figure 25 and Figure 26 As shown, there is a strong tendency indicating the possibility of improving symptoms through training.

[0355] (j6: Specificity of the effects of neurofeedback training)

[0356] Neurofeedback training itself produces nonspecific effects such as learning effects, but neurofeedback training using an estimation model determined using the estimation method according to this embodiment produces specific effects that overcome these nonspecific effects. This is explained by presenting an experimental example.

[0357] Figure 27 : is a diagram showing an experimental example for evaluating the specificity of the effect of neurofeedback training. Figure 27 In (A), as a psychological index related to depression (MDD), an example of changes in RRS and sub-scores is shown. Figure 27 (B) shows examples of changes in SPQ and sub-scores as psychological indicators related to schizophrenia (SCZ).

[0358] exist Figure 27(A) and Figure 27 In (B), "MDD" refers to a group trained using an estimation model (depression estimation model) determined based on EEG measurement data for depression (MDD), and "SCZ" refers to a group trained using an estimation model (schizophrenia estimation model) determined based on EEG measurement data for schizophrenia (SCZ). Furthermore, "CTRL" refers to a comparison group.

[0359] about Figure 27 As shown in (A) of the psychological indicators related to depression (MDD), it can be seen that the RRS (total score) and sub-item scores of the group (MDD) trained using the depression estimation model have undergone specific changes.

[0360] about Figure 27 As shown in (B) of the psychological indicators related to schizophrenia (SCZ), it can be seen that the SPQ (total score) and sub-scores of the group (SCZ) trained using the schizophrenia estimation model have undergone specific changes.

[0361] Figure 28 : is a diagram showing another experimental example for evaluating the specificity of the effect of neurofeedback training. Figure 28 Examples of changes in cognitive function are shown in . Figure 28 (A) shows an experimental example of the N-back task (N=3). Figure 28 (B) shows an experimental example of the N-back task (N=4).

[0362] according to Figure 28 In the experimental example shown in (A), whether the training was performed using the depression estimation model or the schizophrenia estimation model, the cognitive function showed an improvement trend. Figure 28 In the experimental example shown in (B), training using the schizophrenia estimation model showed a significant trend of improvement in cognitive function.

[0363] According to these experimental examples, a tendency toward improvement in cognitive function was shown by neurofeedback training using the estimation model determined by the estimation method according to the present embodiment, and a higher tendency toward improvement was confirmed by using the schizophrenia estimation model.

[0364] [K. Advantages]

[0365] According to the estimation system according to this embodiment, brain functions expressed by multiple intrabrain networks and arbitrary diseases related to the multiple intrabrain networks can be estimated more simply using EEG measurement data.

[0366] In addition, according to the estimation system according to this embodiment, since only the feature quantities in the EEG measurement data that are effective for estimating disease tendencies are used in the estimation model, the dimension of the estimation model can be compressed and reduced, thereby reducing the amount of calculation related to the estimation of disease tendencies and making the estimation of disease tendencies faster.

[0367] Furthermore, according to the estimation system according to this embodiment, since disease tendency can be estimated for any disease associated with a plurality of brain networks, neurofeedback training can be applied to various diseases.

[0368] In addition, according to the estimation system according to this embodiment, the estimation model can be determined using the measurement data obtained by simultaneously performing EEG and fMRI in a resting state. Therefore, when performing simultaneous EEG and fMRI measurements, there is no need to assign tasks to the subject, thereby reducing the burden on the subject when constructing the estimation model.

[0369] Furthermore, the neurofeedback training provided by the estimation system according to this embodiment can provide improvement trends for several diseases and can maintain the improvement trends over a long period of time.

[0370] Furthermore, the estimation model used in the neurofeedback training provided by the estimation system according to the present embodiment exhibits subject specificity, and thus the estimation model is generated according to the disease.

[0371] The embodiments disclosed herein are to be considered in all respects as illustrative and non-restrictive. The scope of the present invention is indicated not by the above description of the embodiments but by the claims, and is intended to include all modifications within the meaning and scope equivalent to the claims.

[0372] Description of Reference Numerals

[0373] 1: Estimation system; 2: Brain activity training device; 11: Feature quantity information; 12: Weighting parameter; 13: Adder; 14: Binarizer; 20: EEG measurement data; 22: Time waveform of power; 24: Time waveform of EEG time correlation; 26: Window; 30: fMRI measurement data; 32: BOLD signal; 34: Time waveform of BOLD time correlation; 38: Disease tendency label; 100, 500: Processing device; 102: Processor; 104: Main storage unit; 106: Control interface 108: Network interface; 110, 352: Input unit; 112, 353: Display unit; 120: Secondary storage unit; 121: Estimation model determination program; 122: Estimation program; 124: Estimation model parameters; 150, 160, 550: Preprocessing module; 152, 162, 552: Time correlation calculation module; 164: WLS calculation module; 166, 556: Binarization module; 168: Model estimation module; 200: EEG device; 202: Multiplexer; 204: Noise filter Filter; 206: A / D converter; 208, 354: Storage unit; 210, 358: Interface; 220: Sensor; 222: Cable; 300: fMRI device; 302: Receiving coil; 310: Magnetic field applying mechanism; 312: Static magnetic field generating coil; 314: Gradient magnetic field generating coil; 316: Irradiation unit; 318: Bedding; 320: Driving unit; 322: Static magnetic field power supply; 324: Gradient magnetic field power supply; 326: Signal transmitting unit; 328: Signal receiving unit; 330: Bedding driving unit; 350: Data processing unit; 351: Control unit; 356: Image processing unit; 357: Data collection unit; 400: Server device; 402: Subject data; 502: Storage device; 510: Display device; 520: Score display; 522: Reference circle; 524: Score circle; 554: Weighted sum calculation module; 558: Estimation model acquisition module; 560: Display control module; 600: Loop; 602: Interval; 604: Guidance period; 606: Display period; 612: Block; 614: Group.

Claims

1. An estimation system comprising: an acquisition unit for acquiring measurement data of electroencephalograms and functional magnetic resonance imaging (FMRI) simultaneously measured from a subject, wherein the electroencephalogram measurement data includes a time waveform for each of a plurality of channels corresponding to a plurality of sensors disposed on the subject's head; The estimation system further comprises: a first calculation unit that calculates a first functional connection for each channel combination based on the correlation between channels included in the brain wave measurement data; a second calculation unit that calculates a second functional connection for each intra-brain network based on correlations between regions of interest included in the functional magnetic resonance imaging measurement data; a third calculation unit that calculates a disease propensity label by calculating a score representing the estimated disease propensity of the subject using a plurality of the second functional connections; as well as A machine learning unit that determines an estimation model for estimating the disease propensity using the prescribed first functional connectivity by performing machine learning using the first functional connectivity of each channel combination and the disease propensity label.

2. The estimation system according to claim 1, wherein The estimation system further includes an estimation unit that inputs measurement data of the electroencephalogram measured from the subject into the estimation model to estimate the disease tendency of the subject.

3. The estimation system according to claim 2, wherein: The estimation system further includes a presentation unit that calculates a second score corresponding to the estimated disease tendency of the subject and presents information corresponding to the calculated second score to the subject.

4. The estimation system according to claim 3, wherein: The estimation model is prepared for each disease. An estimation model corresponding to the disease present in the subject is applied to the subject.

5. The estimation system according to any one of claims 1 to 4, wherein: Based on a second score corresponding to the estimated disease tendency of the subject, a change in the subject's symptoms is evaluated.

6. The estimation system according to any one of claims 1 to 4, wherein: The third calculation unit calculates a score indicating the disease tendency based on a sum of a plurality of second functional connections corresponding to the disease tendency of the estimation subject multiplied by corresponding weighting parameters.

7. The estimation system according to claim 6, wherein: The third calculation unit calculates the disease tendency label by performing threshold processing after normalizing the score representing the disease tendency.

8. The estimation system according to any one of claims 1 to 4, wherein: The estimation model includes information for selecting a first functional connection used in the estimation among the first functional connections of each channel combination, and a weighting parameter corresponding to the selected first functional connection.

9. The estimation system according to any one of claims 1 to 4, wherein: The first calculation unit calculates the first functional connectivity based on a correlation value between time waveforms within a section included in a window commonly set for time waveforms of electroencephalograms of two target channels.

10. The estimation system according to any one of claims 1 to 4, wherein: The first calculation unit calculates the first functional connectivity for each frequency band included in the brain wave measurement data and / or for each window size of a set window.

11. The estimation system according to claim 10, wherein: The estimation system further includes a condition setting unit that determines in advance, based on the subject, a frequency band and / or a window size included in the electroencephalogram measurement data input to the estimation model.

12. The estimation system according to any one of claims 1 to 4, wherein: The second calculation unit calculates the second functional connectivity based on a correlation value between time waveforms within a section included in a window commonly set for time waveforms representing activity amounts of two target regions of interest.

13. An estimation method comprising the steps of acquiring measurement data of electroencephalograms and functional magnetic resonance imaging (FMRI) simultaneously measured from a subject, wherein the electroencephalogram measurement data includes a time waveform for each of a plurality of channels corresponding to a plurality of sensors disposed on the subject's head. The estimation method further comprises the following steps: calculating a first functional connection for each channel combination based on the correlation between channels included in the brain wave measurement data; calculating a second functional connection for each intra-brain network based on correlations between regions of interest included in the functional magnetic resonance imaging measurement data; calculating a disease propensity label by calculating a score representing an estimated disease propensity of the subject using a plurality of the second functional connections; as well as An estimation model for estimating the disease propensity is determined using the prescribed first functional connectivity through machine learning using the first functional connectivity of each channel combination and the disease propensity label.

14. A recording medium storing a program causing a computer to execute the steps of acquiring measurement data of electroencephalograms and functional magnetic resonance imaging (FMRI) simultaneously measured from a subject, the electroencephalogram measurement data including time waveforms of each of a plurality of channels corresponding to a plurality of sensors disposed on the subject's head. The program causes the computer to further perform the following steps: calculating a first functional connection for each channel combination based on the correlation between channels included in the brain wave measurement data; calculating a second functional connection for each intra-brain network based on correlations between regions of interest included in the functional magnetic resonance imaging measurement data; calculating a disease propensity label by calculating a score representing an estimated disease propensity of the subject using a plurality of the second functional connections; as well as An estimation model for estimating the disease propensity is determined using the prescribed first functional connectivity through machine learning using the first functional connectivity of each channel combination and the disease propensity label.

15. A brain activity training device for performing neurofeedback training, comprising: a storage device storing an estimation model for estimating the disease tendency of the subject generated before performing the neurofeedback training; and an electroencephalogram (EEG) device for measuring brainwave data of the subject during the neurofeedback training; The electroencephalogram measurement data includes a time waveform of each of a plurality of channels corresponding to a plurality of sensors disposed on the subject's head. The brain activity training device further comprises: Presentation device; as well as a processing device that calculates the subject's disease tendency using the estimation model based on the measurement data from the electroencephalogram during the neurofeedback training, and outputs a signal for displaying a display corresponding to the disease tendency to the presentation device; The estimation model is generated by the following process: acquiring simultaneously measured brain wave data and functional magnetic resonance imaging data from the subject, wherein the simultaneously measured brain wave data includes a time waveform for each channel corresponding to each channel of the brain wave data measured during the neurofeedback training; calculating a first functional connection for each channel combination based on the correlation between channels included in the brain wave measurement data; calculating a second functional connection for each intra-brain network based on correlations between regions of interest included in the functional magnetic resonance imaging measurement data; calculating a disease propensity label by calculating a score representing an estimated disease propensity of the subject using a plurality of the second functional connections; as well as The disease propensity is estimated using a prescribed first functional connectivity by machine learning using the first functional connectivity of each channel combination and the disease propensity label, thereby determining the estimation model.

16. A brain activity training method for performing neurofeedback training, the brain activity training method comprising the following steps: obtaining an estimation model for estimating the disease tendency of the subject generated before performing the neurofeedback training; as well as measurement data of the brain waves of the subject measured during the neurofeedback training, The electroencephalogram measurement data includes a time waveform of each of a plurality of channels corresponding to a plurality of sensors disposed on the subject's head. The brain activity training method further includes the following steps: in the neurofeedback training, based on the brain wave measurement data, using the estimation model to calculate the disease tendency of the subject, and outputting a signal for displaying a display corresponding to the disease tendency to a presentation device; The step of obtaining the estimation model comprises the following steps: acquiring simultaneously measured brain wave data and functional magnetic resonance imaging data from the subject, wherein the simultaneously measured brain wave data includes a time waveform for each channel corresponding to each channel of the brain wave data measured during the neurofeedback training; calculating a first functional connection for each channel combination based on the correlation between channels included in the brain wave measurement data; calculating a second functional connection for each intra-brain network based on correlations between regions of interest included in the functional magnetic resonance imaging measurement data; calculating a disease propensity label by calculating a score representing an estimated disease propensity of the subject using the plurality of second functional connections; and The disease propensity is estimated using a prescribed first functional connectivity by machine learning using the first functional connectivity of each channel combination and the disease propensity label, thereby determining the estimation model.

17. A recording medium storing a brain activity training program, wherein the brain activity training program is used to perform neurofeedback training, wherein the brain activity training program causes a computer to execute the following steps: saving an estimation model for estimating the disease tendency of the subject generated before performing the neurofeedback training; and obtaining measurement data of the subject's brain waves during the neurofeedback training, The electroencephalogram measurement data includes a time waveform of each of a plurality of channels corresponding to a plurality of sensors disposed on the subject's head. The brain activity training program causes the computer to further perform the following steps: in the neurofeedback training, based on the brain wave measurement data, using the estimation model to calculate the disease tendency of the subject, and outputting a signal for displaying a display corresponding to the disease tendency to a presentation device; The estimation model is generated by the following process: acquiring simultaneously measured brain wave data and functional magnetic resonance imaging data from the subject, wherein the simultaneously measured brain wave data includes a time waveform for each channel corresponding to each channel of the brain wave data measured during the neurofeedback training; calculating a first functional connection for each channel combination based on the correlation between channels included in the brain wave measurement data; calculating a second functional connection for each intra-brain network based on correlations between regions of interest included in the functional magnetic resonance imaging measurement data; calculating a disease propensity label by calculating a score representing an estimated disease propensity of the subject using a plurality of the second functional connections; as well as The disease propensity is estimated using a prescribed first functional connectivity by machine learning using the first functional connectivity of each channel combination and the disease propensity label, thereby determining the estimation model.

Images