Discrimination device, method for discriminating depressive symptoms, method for judging the level of depressive symptoms, method for classifying patients with depression, method for judging the treatment effect of depressive symptoms, and brain activity training device

Abstract

The aim is to achieve a disease label for discriminating depressive symptoms based on the activity state of the brain. A means for solving the problems of the present invention is to provide a discrimination device for assisting in judging whether a subject has depressive symptoms. The discrimination device includes a storage device for storing information on an identification classifier, which is generated based on signals obtained by using a brain activity detection device to measure in advance and chronologically the brain activities of multiple predetermined regions of the brains of multiple participants including healthy individuals and depressive patients in a resting state, and through classifier generation processing. The classifier is generated to discriminate the disease label of depressive symptoms based on the weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of multiple predetermined regions. The discrimination device further includes a processor that performs discrimination processing, and the discrimination processing generates a classification result for the depressive symptoms of the subject by using the classifier.

Description

Technical Field

[0001] The present invention relates to a discrimination device, a method for discriminating depressive symptoms, a method for judging the level of depressive symptoms, a method for classifying patients with depression, a method for judging the treatment effect of depressive symptoms, and a brain activity training device. Background Art

[0002] (Biomarker)

[0003] An index for quantifying biological information to quantitatively grasp biological changes in a living body is called a "biomarker".

[0004] The US Food and Drug Administration (FDA) defines a biomarker as "a characteristic of an indicator that is measured as a normal biological process, a pathogenic process, or a response to an exposure or intervention (including a therapeutic intervention)". In addition, a biomarker characterized by the state or change or degree of cure of a disease is used as a surrogate marker (substitute marker) for verifying the effectiveness of a new drug in a clinical trial. Blood glucose level or cholesterol level, etc. are representative biomarkers as indicators of lifestyle-related diseases. Among biomarkers, not only substances contained in urine or blood from a living body are included, but also electrocardiogram, blood pressure, PET images, bone density, and lung function, etc. are included. In addition, due to the development of genomic analysis or proteomic analysis, various biomarkers related to, for example, DNA, RNA, and living body proteins have been discovered.

[0005] Biomarkers are expected to be used not only for measuring the treatment effect after a patient gets sick, but also for preventing diseases as an indicator for disease prevention in daily life, or for personalized medicine for selecting an effective treatment to avoid side effects.

[0006] However, in the case of neuro / psychiatric disorders, molecular markers, etc. that can be used as objective indicators in terms of biochemistry or molecular genetics are still under research and development.

[0007] On the other hand, for example, the following disease discrimination system has also been reported. The disease discrimination system is configured to classify schizophrenia, depression, or other mental disorders according to characteristics obtained from hemoglobin signals measured by in vivo optical measurement by using the technology of near-infrared spectroscopy (NIRS) (Patent Document 1).

[0008] Biomarkers are used not only for general diagnosis but also for drug development.

[0009] In the case of general drug development, a pharmacodynamic marker is used in the exploratory research stage, and a toxicity marker for examining the toxicity of a candidate compound is also used in the non-clinical stage.

[0010] In the clinical trial phase, surrogate markers are used as evaluation items, or various markers such as diagnostic markers, prognostic markers, predictive markers, monitoring markers, and safety markers are used.

[0011] The term "surrogate marker" refers to a marker whose scientific relationship with the true endpoint has been verified.

[0012] The term "diagnostic marker" refers to two types of markers, namely, diagnostic markers that identify the presence or type of a disease, and diagnostic markers that judge the severity or degree of progression of a disease. The term "predictive marker" refers to a biomarker that identifies the population for which a specific drug is effective. The term "prognostic marker" refers to a biomarker that predicts the progression or recovery of a disease regardless of a specific treatment. A "monitoring marker", also known as an "efficacy marker", is a marker that verifies the efficacy or therapeutic effect of a drug. When this marker is monitored over a long period, it can also be used as a hint for detecting the cause of drug resistance.

[0013] For example, when the target molecule is clear, for example, the target molecule can be used as a classification marker in clinical trials to identify and select a group of patients for which a specific drug is effective. In this case, when a drug using a classification marker is approved, a diagnostic reagent for measuring the classification marker can be called an "accompanying diagnostic reagent". In other cases, a diagnosis performed to select a treatment method for a patient through such a classification marker can be called an "accompanying diagnosis".

[0014] The costs of late-stage trials in clinical trials are very high, so it is necessary to consider developing biomarkers suitable for late-stage trials in clinical trials to improve the efficiency of clinical trials. However, the development of such biomarkers is generally not easy.

[0015] In addition, for example, when considering biological indicators (biomarkers), in the pathophysiology of a disease, so-called "trait markers" reflect behavioral or biological processes related to the cause of the disease, and so-called "state markers" reflect the clinical condition of a patient. Trait markers are also called "genetic indicators", and state markers are also called "state-dependent indicators".

[0016] (Real-time neurofeedback)

[0017] For example, so far, for obsessive-compulsive disorder (OCD), which is a type of neurosis, drug treatment methods and behavioral treatment methods are known as treatment methods. For example, selective serotonin reuptake inhibitors are used for drug treatment, and for example, exposure and response prevention therapy, which combines exposure therapy and response prevention therapy, is called a behavioral treatment method.

[0018] On the other hand, real-time neurofeedback is considered a possible treatment for neurological / psychiatric disorders (Patent Document 6).

[0019] Brain functional imaging, which includes, for example, functional magnetic resonance imaging (fMRI) as a method of visualizing the hemodynamic response related to the activity of the human brain by using magnetic resonance imaging (MRI), has been used to detect the difference between brain activity caused by sensory stimulation or the execution of a cognitive task and brain activity caused by the resting state or the execution of a control task, thereby identifying the brain activation region corresponding to the component of the brain function of interest, that is, finding the localization of the brain function.

[0020] In recent years, real-time neurofeedback techniques using brain functional imaging (e.g., functional magnetic resonance imaging (fMRI)) have been reported (Non-Patent Document 1). Real-time neurofeedback techniques have begun to attract attention as a treatment method for neurological disorders and psychiatric disorders.

[0021] Neurofeedback is a type of biofeedback, and the subject receives feedback on his or her own brain activity to learn a method of manipulating brain activity.

[0022] For example, it has been reported that when the activity of the anterior cingulate cortex is measured by fMRI and the patient receives real-time feedback on this activity as the magnitude of inflammation (fire) in an attempt to reduce the inflammation, central chronic pain has been improved in real time and for a long time (see Non-Patent Document 2).

[0023] (Resting-state fMRI)

[0024] Furthermore, recent studies have shown that the brain is active even in the resting state. Specifically, there are populations of nerve cells in the brain that are quiescent during human activities and actively excited during the resting period. Anatomically, the medial surface of the combined left and right cerebral hemispheres mainly corresponds to such a region and includes, for example, the medial surface of the frontal lobe, the posterior cingulate cortex, the precuneus, the posterior part of the parietal association area, and the middle temporal gyrus. The region representing the baseline of the brain activity in this resting state is named the "default mode network (DMN)" and acts as a network for synchronization (see Non-Patent Document 3).

[0025] For example, brain activity in the default mode network is given as an example of the difference between the brain activity of healthy individuals and that of patients with mental disorders. The default mode network represents regions where brain activity is more active in the resting state compared to during the execution of goal-based tasks. It has been reported that patients with diseases such as schizophrenia or Alzheimer's disease show abnormalities in the default mode network compared to healthy individuals. For example, it has been reported that patients with schizophrenia have a decreased correlation in activity between the posterior cingulate cortex belonging to the default mode network and the lateral parietal cortex, the medial frontal cortex, or the cerebellar cortex in the resting state.

[0026] However, for example, it is not necessarily revealed how this default mode network and cognitive function are related to each other, and how the correlation between the functional connections of brain regions is related to the above-mentioned neurofeedback.

[0027] On the other hand, the correlation between activities caused by tasks in multiple brain regions, for example, is examined to evaluate the functional connections of these multiple brain regions. In particular, the evaluation of functional connections by fMRI in the resting state is also called "resting-state functional connectivity MRI (rs-fc MRI)", and clinical studies on various neuro / psychiatric disorders have been gradually carried out more and more widely. However, the existing rs-fc MRI methods examine the activities of global neural networks (such as the above-mentioned default mode network), and more detailed functional connections are not currently fully considered.

[0028] (Magnetic Resonance Imaging)

[0029] Now, this magnetic resonance imaging will be briefly described below.

[0030] Specifically, to date, as a method for imaging cross-sections of the living brain or the whole body, magnetic resonance imaging using the magnetic resonance phenomenon of atoms in the living body (especially the atomic nuclei of hydrogen atoms) has been used, for example, in human clinical diagnostic imaging.

[0031] When magnetic resonance imaging is applied to the human body, magnetic resonance imaging has the following characteristics, for example, when compared with "X-ray CT", which is a similar human tomography.

[0032] (1) An image with a concentration corresponding to the distribution of hydrogen atoms and the signal relaxation time (which reflects the bonding strength between atoms) is obtained. Thus, shades showing differences depending on the nature of the tissue are exhibited, facilitating the observation of tissue differences.

[0033] (2) The magnetic field is not absorbed by bones. Thus, it is easy to observe the regions surrounded by bones (such as the intracranial or spinal cord).

[0034] (3) Different from X-rays, magnetic resonance imaging does not harm the human body and can thus be used in a wide range of applications.

[0035] This magnetic resonance imaging uses the magnetism of hydrogen atomic nuclei (protons) that are most abundant and have the greatest magnetism in the various cells of the human body. Classically, the movement of the spin angular momentum in the magnetic field responsible for the magnetism of the hydrogen atomic nucleus is compared with the precession of a top.

[0036] Now, to explain the background of the present invention, the principle of nuclear magnetic resonance is briefly outlined through an intuitive classical model.

[0037] The direction of the spin angular momentum of the above hydrogen atomic nucleus (the direction of the axis of rotation of the top) is random in an environment without a magnetic field, but is the same as the direction of the magnetic field lines when a static magnetic field is applied.

[0038] In this state, when an oscillating magnetic field is additionally applied and the frequency of this oscillating magnetic field is the resonance frequency f0 = γB0 / 2π (γ: a coefficient specific to the substance) determined by the intensity of the static magnetic field, resonance causes energy to move towards the atomic nucleus and the direction of the magnetization vector changes (the amplitude of precession increases). In this state, when the oscillating magnetic field is cancelled, precession returns the tilt angle to its original angle while returning to the direction under the static magnetic field. This process is detected externally through an antenna coil, and thus an NMR signal can be obtained.

[0039] When the static magnetic field intensity is B0 (T), this resonance frequency f0 of hydrogen atoms is 42.6 × B0 (MHz).

[0040] In addition, magnetic resonance imaging can also visualize the parts of the brain that become active in response to, for example, external stimuli, based on the fact that the detected signal changes according to the change in blood flow. This magnetic resonance imaging is particularly called "functional magnetic resonance imaging".

[0041] Regarding the equipment for fMRI, a general MRI device additionally installed with the hardware and software required for fMRI measurement is used.

[0042] Since oxygenated hemoglobin and deoxygenated hemoglobin in the blood have different magnetisms, fMRI uses the fact that the change in blood flow causes a change in the intensity of the NMR signal. Oxygenated hemoglobin has the property of a diamagnetic substance and does not affect the relaxation time of hydrogen atoms in the surrounding water, while deoxygenated hemoglobin is a paramagnetic substance and changes the surrounding magnetic field. Thus, when the brain is stimulated, local blood flow increases, and oxygenated hemoglobin increases, the corresponding changes can be detected as an MRI signal. As a stimulus to the subject, for example, visual stimuli, auditory stimuli, or the execution of a predetermined task (for example, Patent Document 2) are used.

[0043] On the other hand, in the study of brain function, the activity of the brain is measured by measuring the increase in the nuclear magnetic resonance signal (MRI signal) of hydrogen atoms corresponding to the phenomenon of a decrease in the concentration of deoxyhemoglobin in red blood cells in small veins or capillaries (BOLD effect).

[0044] In particular, when studying the motor function of a person, the subject performs some movements, and at the same time, the above-mentioned MRI device measures the activity of the brain.

[0045] On the other hand, in the case of humans, non-invasive measurement of brain activity is required, and in this case, decoding techniques that can extract more detailed information from fMRI data are being developed (for example, Non-Patent Document 4). In particular, fMRI analyzes brain activity in units of voxels in the brain, and thus can estimate stimulus input or recognition status based on the spatial pattern of brain activity.

[0046] In addition, as a technique that has developed such a decoding technique, Patent Document 3 discloses a method for analyzing brain activity of a biomarker by brain functional imaging for neuro / psychiatric disorders. This method derives a correlation matrix of the activity levels of predetermined brain regions for each subject based on the data of resting-state functional connectivity MRI measured for a group of healthy individuals and a group of patients. Features are extracted by performing regularized canonical correlation analysis on the correlation matrix and the attributes of the subject (including the disease / healthy individual label of the subject). Based on the results of regularized canonical correlation analysis (sparse canonical correlation analysis), a classifier used as a biomarker is generated by discriminant analysis using sparse logistic regression (SLR). It has been shown that using such machine learning techniques, the diagnostic results of autism can be predicted based on the connections (functional connections) between brain regions derived from resting-state fMRI data (Patent Documents 4 and 5). In addition, it has been shown that the verification of prediction performance is not limited to brain activity measured at one site, but can also be generalized to some extent to brain activity measured at other sites.

[0047] In addition, in recent years, it has been pointed out that fMRI can be used for a large multi-site sample group (n = 1188) obtained from a large number of research sites to classify patients with depression into four neurophysiological subtypes (biotypes), and these biotypes are clearly represented by different patterns indicating dysfunctional connections in the limbic and fronto-striatal networks (see Non-Patent Document 5). In Non-Patent Document 5, these biotypes cannot be distinguished based on clinical characteristics alone, but are associated with different clinical symptom scales. In addition, the response to transcranial magnetic stimulation treatment (n = 154) is predicted.

[0048] In Non-Patent Document 5, the above subtypes are classified according to the functional connectivity in the brain based on the BOLD signals obtained from 258 regions in the brain.

[0049] Prior Art Documents

[0050] Patent Documents

[0051] Patent Document 1: JP2006 / 132313

[0052] Patent Document 2: JP2011-000184A

[0053] Patent Document 3: JP2015-62817A

[0054] Patent Document 4: JP6195329 B1

[0055] Patent Document 5: WO2017 / 090590A1

[0056] Patent Document 6: JP5641531 B1

[0057] Non-Patent Document 1: Nikolaus Weiskopf, "Real-time fMRI and its application to neurofeedback", NeuroImage 62(2012)682-692

[0058] Non-Patent Document 2: deCharms RC, Maeda F, Glover GH et al, "Control over brain activation and pain learned by using real-time functional MRI", Proc Natl Acad Sci USA 102(51), 18626-18631, 2005

[0059] Non-Patent Document 3: Raichle ME, Macleod AM, Snyder AZ.et al. "A default mode of brain function", Proc Natl Acad Sci USA 98(2), 676-682, 2001

[0060] Non-Patent Document 4: Kamitani Y, Tong F. Decoding the visual and subjective contents of the human brain. Nat Neurosci.2005;8:679-85

[0061] Non-Patent Document 5: Drysdale AT, Grosenick L, Downar J, Dunlop K, Mansouri F, Meng Y, Fetcho RN, Zebley B, Oathes DJ, Etkin A, Schatzberg AF, Sudheimer K, Keller J, Mayberg HS, Gunning FM, Alexopoulos GS, Fox MD, Pascual-Leone A, Voss HU, Casey BJ, Dubin MJ, Liston C. Resting-state connectivity biomarkers define neurophysiological subtypes of depression. Nat Med. 2017 23(1):28-38 Summary of the Invention

[0062] Problems to be Solved by the Invention

[0063] As described above, when considering applying the analysis of brain activity by brain functional imaging (e.g., functional magnetic resonance imaging) to the treatment of neurological / psychiatric disorders, for example, as the above biomarker, it is also expected to apply the analysis of brain activity by brain functional imaging as a non-invasive functional marker to, for example, the development of diagnostic methods and the search for and identification of target molecules used in the development of drugs for achieving radical treatment.

[0064] For example, so far, for depression, the development of practical biomarkers using genes has not been completed. Therefore, it is not easy to objectively judge the effect of drugs and the development of therapeutic drugs is also difficult.

[0065] In order to generate a classifier by machine learning based on measurement data of brain activity and put this classifier into practical use as a biomarker, it is necessary to improve the prediction accuracy of the biomarker generated by machine learning for brain activity measured at one site. In addition, it is required that the biomarker generated in this way can also be generalized to brain activity measured at other sites.

[0066] That is to say, when constructing a classifier by machine learning based on measurement data of brain activity, there are two main problems.

[0067] The first problem is the problem of small sample size.

[0068] The data volume N of the number of participants is much smaller than the dimension M of the brain activity measurement data obtained through measurement, so the parameters of the classifier are prone to overfitting to the training data (referred to as "overfitting" or "overtraining").

[0069] Due to this overfitting, the constructed classifier exhibits extremely poor performance on newly sampled test data. This is because such test data has not been used to train the classifier.

[0070] Therefore, it is necessary to appropriately introduce regularization to identify and use only the essential features for the desired generalization of the classifier.

[0071] For example, in the above-mentioned non-patent document 5, this regularization may not have been fully considered.

[0072] The second problem is that the constructed classifier is clinically useful and scientifically reliable only when it maintains good performance on MRI data scanned at a camera location different from the location where the training data was collected.

[0073] This is the so-called generalization ability with respect to the camera location.

[0074] However, in clinical applications, it is often observed that a classifier trained using data obtained at a specific location cannot generalize to data scanned at different locations.

[0075] The present invention is made to solve the above problems, and the object of the present invention is to objectively discriminate the disease label of depressive symptoms for the activity state of the brain. In addition, another object of the present invention is to discriminate information representing the degree of treatment effect on depressive symptoms. Another object of the present invention is to provide a classifier configured to output an index value serving as a biomarker for objectively discriminating the state of depressive symptoms. Another object of the present invention is to discriminate the level of depressive symptoms. Another object of the present invention is to classify patients with depression.

[0076] Solutions for Solving the Problems

[0077] One embodiment of the present invention relates to a discrimination device for assisting in determining whether a subject has depressive symptoms. The discrimination device according to this embodiment includes a storage device for storing information for identifying a classifier, the classifier being generated based on signals obtained by measuring in advance and in chronological order signals representing brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and depression patients in a resting state by using a brain activity detection device, through classifier generation processing. The classifier is generated to discriminate the disease label of depressive symptoms based on the weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through machine learning and feature selection from the functional connections of the multiple predetermined regions. The multiple functional connections selected include at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex; and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The discrimination device according to this embodiment further includes a processor configured to execute discrimination processing for generating a classification result of the depressive symptoms of the subject by using the classifier.

[0078] One embodiment of the present invention relates to a discrimination device for the level of depressive symptoms. The discrimination device according to this embodiment includes a storage device for storing information for identifying a classifier, the classifier being generated based on signals obtained by measuring in advance and in chronological order signals representing brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and depression patients in a resting state by using a brain activity detection device, through classifier generation processing. The classifier is generated to discriminate the disease label of depressive symptoms based on the weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through machine learning and feature selection from the functional connections of the multiple predetermined regions. The multiple functional connections selected include at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex; and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The discrimination device according to this embodiment further includes a processor configured to generate, by the classifier, an index value for evaluating depressive symptoms for elements of a correlation matrix of functional connections measured for the subject in a resting state. The processor is configured to compare the index value with a reference range of index values set in advance for each of the multiple functional connections according to the level of depressive symptoms. The processor is configured to determine that the subject has a level of depressive symptoms corresponding to the reference range including the index value.

[0079] One embodiment of the present invention relates to a discrimination device for judging the treatment effect on a subject. The discrimination device according to the present embodiment includes a storage device for storing information for identifying a classifier, the classifier being generated based on signals obtained by measuring in advance and chronologically the brain activities of a plurality of predetermined regions of each brain of a plurality of participants including healthy individuals and depression patients in a resting state by using a brain activity detection device, and through a classifier generation process. The classifier is generated to discriminate the disease label of depressive symptoms based on the weighted sum of a plurality of functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the plurality of predetermined regions. The plurality of selected functional connections includes at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex; and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The discrimination device according to the present embodiment further includes a processor configured to generate a first value for evaluating depressive symptoms for an element of a correlation matrix of functional connections measured for the subject in a resting state at a first time point by using the classifier. The processor is configured to generate a second value for evaluating depressive symptoms for an element of a correlation matrix of the same functional connections inside the brain measured for the same subject in a resting state at a second time point, the second time point being a time point after the start of treatment and later than the first time point, by using the classifier. The processor is configured to compare the first value with the second value. The discrimination device according to the present embodiment is configured to determine that the treatment is effective for improving the depressive symptoms of the subject when the second value is improved compared with the first value, and / or determine that the treatment is ineffective for improving the depressive symptoms of the subject when the second value is not improved compared with the first value.

[0080] One embodiment of the present invention relates to a discrimination device for classifying depression patients when classifying depressive symptoms into a plurality of preset subcategories. The discrimination device according to the present embodiment includes a processor configured to generate an index value for evaluating depressive symptoms for an element of a correlation matrix of functional connections measured for a subject in a resting state. The processor is configured to compare the index value with a reference range of index values preset for each of the plurality of functional connections according to each of the plurality of subcategories. The processor is configured to determine that the subject has one of the plurality of subcategories corresponding to the reference range including the index value.

[0081] One embodiment of the present invention relates to a computer program for causing the discrimination device to perform the above processing.

[0082] One embodiment of the present invention relates to a discrimination method for assisting in judging a subject with depressive symptoms. The discrimination method according to the present embodiment includes the following steps: generating an index value for evaluating depressive symptoms for elements of a correlation matrix of functional connections measured for a subject in a resting state. The functional connection includes at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The discrimination method according to the present embodiment further includes the following steps: determining that the subject has depressive symptoms when the index value exceeds a reference value.

[0083] One embodiment of the present invention relates to a discrimination method for assisting in judging the level of depressive symptoms of a subject. The discrimination method according to the present embodiment includes the following steps: generating an index value for evaluating depressive symptoms for elements of a correlation matrix of functional connections measured for a subject in a resting state. The functional connection includes at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The discrimination method according to the present embodiment further includes the following steps: comparing the index value with a reference range of index values set in advance for each of the functional connections according to the level of depressive symptoms; and determining that the subject has a level of depressive symptoms corresponding to the reference range including the index value.

[0084] One embodiment of the present invention relates to a discrimination method for assisting in judging the treatment effect on a subject. The discrimination method according to this embodiment includes the following steps: generating a first value for evaluating depressive symptoms for elements of the correlation matrix of the functional connections measured for the subject at a first time point in a resting state. The functional connections include at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The discrimination method according to this embodiment further includes the following steps: generating a second value for evaluating depressive symptoms for elements of the correlation matrix of the same functional connections inside the brain measured for the same subject at a second time point in a resting state, where the second time point is a time point after the start of treatment and later than the first time point. The discrimination method according to this embodiment further includes the following steps: comparing the first value with the second value; and determining that the treatment is effective in improving the depressive symptoms of the subject when the second value is improved compared to the first value.

[0085] One embodiment of the present invention relates to a discrimination method for assisting in judging the treatment effect on a subject. The discrimination method according to this embodiment includes the following steps: generating a first value for evaluating depressive symptoms for elements of the correlation matrix of the functional connections measured for the subject at a first time point in a resting state. The functional connections include at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The discrimination method according to this embodiment further includes the following steps: generating a second value for evaluating depressive symptoms for elements of the correlation matrix of the same functional connections inside the brain measured for the same subject at a second time point in a resting state, where the second time point is a time point after the start of treatment and later than the first time point. The discrimination method according to this embodiment further includes the following steps: comparing the first value with the second value; and determining that the treatment is ineffective in improving the depressive symptoms of the subject when the second value is not improved compared to the first value.

[0086] One embodiment of the present invention relates to a discrimination method for assisting in classifying patients with depression. The discrimination method according to this embodiment includes the following steps: When classifying depressive symptoms into a plurality of preset subcategories, for the elements of the correlation matrix of the functional connectivity measured for the subject in the resting state, an index value for evaluating the depressive symptoms is generated. The functional connectivity includes at least one selected from the following: a first functional connectivity between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and a second functional connectivity between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The discrimination method according to this embodiment further includes the following steps: comparing the index value with a reference range of index values preset for each of the functional connectivities according to each of the subcategories among the plurality of subcategories; and determining that the subject has one of the plurality of subcategories corresponding to the reference range including the index value.

[0087] One embodiment of the present invention relates to a discrimination device for assisting in classifying patients with depression. The discrimination device according to this embodiment includes: a processor configured to perform classification processing; and a storage device configured to store information for identifying a classifier, the classifier being generated based on signals obtained by measuring in advance and in chronological order the brain activities of a plurality of predetermined regions of the brains of a plurality of participants including healthy individuals and patients with depression in the resting state by using a brain activity detection device, through a classifier generation process. In the discrimination device according to this embodiment, when classifying depressive symptoms into a plurality of preset subcategories, the classifier is generated to discriminate the label of the subcategory of depression for the elements of the correlation matrix to be discriminated corresponding to the plurality of functional connectivities, based on the weighted sum of a plurality of functional connectivities selected as being related to the label of the subcategory of depression through feature selection via machine learning from the functional connectivities of the plurality of predetermined regions. The plurality of selected functional connectivities includes at least one selected from the following: a first functional connectivity between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex; and a second functional connectivity between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The processor is configured to perform processing for discriminating the subcategory based on the weighted sum and the elements of the correlation matrix to be discriminated.

[0088] One embodiment of the present invention relates to a discrimination device for assisting in judging the treatment effect on a depression patient. The discrimination device according to this embodiment includes a processor, and the processor is configured to: perform a classifier generation process for measuring, for a plurality of subjects, the correlation at a first time point of a plurality of functional connections selected from functional connection identification numbers 1 to 12 shown in Table 1 and the correlation at a second time point of the plurality of functional connections, thereby pre-generating a classifier configured to distinguish, in a correlation state space expanded by the difference between the correlation at the first time point and the correlation at the second time point of the plurality of functional connections, a group of subjects showing a treatment effect among the plurality of subjects from a group of subjects not showing a treatment effect among the plurality of subjects, and the second time point is set to be after the start of treatment and later than the first time point. The processor is configured to measure the first correlation of the plurality of functional connections inside the brain of the subject at rest at the first time point, and measure the second correlation of the plurality of functional connections inside the brain of the same subject at rest at the second time point. The processor is configured to discriminate the treatment effect on the subject by using the classifier based on the difference between the first correlation and the second correlation of the plurality of functional connections of the subject.

[0089] One embodiment of the present invention relates to a discrimination method for assisting in judging the treatment effect on a depression patient. The discrimination method according to this embodiment includes the following steps: measuring, for a plurality of subjects, the correlation at a first time point of a plurality of functional connections selected from functional connection identification numbers 1 to 12 shown in Table 1 and the correlation at a second time point of the plurality of functional connections, thereby pre-generating a classifier configured to distinguish, in a correlation state space expanded by the difference between the correlation at the first time point and the correlation at the second time point of the plurality of functional connections, a group of subjects showing a treatment effect among the plurality of subjects from a group of subjects not showing a treatment effect among the plurality of subjects, and the second time point is set to be after the start of treatment and later than the first time point. The discrimination method according to this embodiment further includes the following steps: measuring the first correlation of the plurality of functional connections inside the brain of the subject at rest at the first time point; and measuring the second correlation of the plurality of functional connections inside the brain of the same subject at rest at the second time point. The discrimination method according to this embodiment further includes the following step: discriminating the treatment effect on the subject by using the classifier based on the difference between the first correlation and the second correlation of the plurality of functional connections of the subject.

[0090] One embodiment of the present invention relates to a method for using a classifier to assist in determining whether a subject has depressive symptoms or the level of depressive symptoms. The classifier is generated based on signals obtained by pre - measuring, in chronological order, signals representing brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and patients with depression in a resting state using a brain activity detection device, through classifier generation processing. The method includes the following steps: generating the classifier, which is configured to discriminate the disease label of depressive symptoms based on the weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the multiple predetermined regions. The multiple selected functional connections include at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex; and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The method further includes the following step: inputting an index value generated based on elements of the correlation matrix of the multiple functional connections of the subject into the classifier.

[0091] One embodiment of the present invention relates to a method for using a classifier to assist in determining the treatment effect of depressive symptoms. The classifier is generated based on signals obtained by pre - measuring, in chronological order, signals representing brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and patients with depression in a resting state using a brain activity detection device, through classifier generation processing. The method includes the following steps: generating the classifier, which is configured to discriminate the disease label of depressive symptoms based on the weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the multiple predetermined regions. The multiple selected functional connections include at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The method further includes the following step: inputting an index value generated based on elements of the correlation matrix of the multiple functional connections of the subject into the classifier.

[0092] One embodiment of the present invention relates to a method for using a classifier to assist in classifying a subject into a plurality of pre-set subcategories when classifying depressive symptoms into the plurality of pre-set subcategories, where the classifier is generated by classifier generation processing based on signals obtained by pre-temporally measuring signals representing brain activities of a plurality of predetermined regions of each brain of a plurality of participants including healthy individuals and patients with depression in a resting state using a brain activity detection device. The classifier is generated to discriminate a disease label of depressive symptoms based on a weighted sum of a plurality of functional connections selected as being related to the disease label of depressive symptoms by feature selection via machine learning from the functional connections of the plurality of predetermined regions. The plurality of selected functional connections includes at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The method includes the step of inputting an index value generated based on elements of a correlation matrix of the plurality of functional connections of the subject into the classifier.

[0093] One embodiment of the present invention relates to a discrimination device for assisting in judging the treatment effect on a subject. The discrimination device includes a classifier generation device, and the classifier generation device includes a first processor configured to generate a first classifier based on signals obtained by measuring in advance and in chronological order signals representing brain activities of multiple predetermined regions of the brains of multiple participants including healthy individuals and depression patients in a resting state. The first classifier is generated to discriminate the disease label of depressive symptoms based on the weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the multiple predetermined regions. The multiple selected functional connections include at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex; and a second functional connection between the left pars opercularis of the inferior frontal gyrus and the right dorsomedial prefrontal cortex and the right supplementary motor area. The first processor is configured to: for multiple subjects, measure the correlation at a first time point and the correlation at a second time point of multiple functional connections selected from functional connection identification numbers 1 to 12 shown in Table 5, thereby generating a second classifier configured to distinguish, in a correlation state space expanded by the difference between the correlation at the first time point and the correlation at the second time point of the multiple functional connections, a group of subjects showing a treatment effect among the multiple subjects from a group of subjects not showing a treatment effect among the multiple subjects, where the second time point is set after the start of treatment and later than the first time point. The discrimination device further includes a classification device including a second processor and a storage device. The storage device is configured to store information related to the first classifier and the second classifier, and information for classifying depressive symptoms into multiple preset subcategories. The second processor is configured to perform a process of classifying a subject into multiple subcategories by using the first classifier. The second classifier is configured to perform a process of measuring the first correlation of the multiple functional connections in the resting state at the first time point for a subject classified into a specific subcategory through the classification process. The second processor is configured to perform a process of measuring the second correlation of the multiple functional connections in the resting state at the second time point for the same subject. The second processor is configured to discriminate the treatment effect on the subject by using the second classifier based on the difference between the first correlation and the second correlation of the multiple functional connections of the subject.

[0094] One embodiment of the present invention relates to a discrimination method for assisting in judging the treatment effect on a subject. The discrimination method according to this embodiment includes a classification step. The classification step is used to classify the subject into a plurality of preset subcategories when classifying depressive symptoms into the plurality of preset subcategories. By using a first classifier generated by processing signals obtained by measuring in advance and in chronological order the brain activities of a plurality of predetermined regions of each brain of a plurality of participants including healthy individuals and patients with depression in a resting state by using a brain activity detection device, the subject is classified into the plurality of subcategories. In the discrimination method according to this embodiment, the first classifier is generated to discriminate the disease label of depressive symptoms based on the weighted sum of a plurality of functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the plurality of predetermined regions. The plurality of selected functional connections includes at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The discrimination method according to this embodiment includes the following steps: a first correlation measurement step for measuring the first correlation of the plurality of functional connections in the resting state before administration for the subject classified into a specific subcategory in the classification step; and a second correlation measurement step for measuring the second correlation of the plurality of functional connections in the resting state after a predetermined period of time has elapsed since the start of administration for the same subject. The discrimination method according to this embodiment further includes a discrimination step for discriminating the treatment effect on the subject classified into the specific subcategory by using a second classifier. The second classifier is configured to measure the correlation at a first time point and the correlation at a second time point of a plurality of functional connections selected from the functional connection identification numbers 1 to 12 shown in Table 1 for a plurality of subjects, and the second time point is set to be after administration and later than the first time point. The discrimination method further includes the following steps: in the correlation state space expanded by the difference between the correlations of the plurality of functional connections at the first time point and the second time point, based on the difference between the first correlation and the second correlation of the plurality of functional connections of the subject pre-generated by the second classifier generation process that distinguishes a group of subjects showing a treatment effect among the plurality of subjects from a group of subjects not showing a treatment effect among the plurality of subjects, the treatment effect on this subject is discriminated.

[0095] One embodiment of the present invention relates to a first classifier generation device, comprising a processor and a storage device. The processor is configured to generate information for identifying a classifier based on signals obtained by pre-measuring in chronological order signals representing brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and patients with depression in a resting state by using a brain activity detection device. The classifier is generated to discriminate a disease label of depressive symptoms based on a weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the multiple predetermined regions. The multiple functional connections selected include at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The storage device is configured to store information for identifying the first classifier generated by the processor.

[0096] One embodiment of the present invention relates to a method for generating a first classifier. In this embodiment, the first classifier is generated through a classifier generation process based on signals obtained by pre-measuring in chronological order signals representing brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and patients with depression in a resting state by using a brain activity detection device. The first classifier is generated to discriminate a disease label of depressive symptoms based on a weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the multiple predetermined regions. The multiple functional connections selected include at least one functional connection selected from the following: a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex; and a second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area.

[0097] One embodiment of the present invention relates to a second classifier generation device, including a processor and a storage device. The processor is configured to measure, for a plurality of subjects, the correlation at a first time point of a plurality of functional connections selected from functional connection identification numbers 1 to 12 shown in Table 3 and the correlation at a second time point of the plurality of functional connections, where the second time point is set after the start of treatment and later than the first time point. In this embodiment, the processor is configured to generate a second classifier, which is configured to distinguish, in a correlation state space expanded by the difference between the correlation at the first time point and the correlation at the second time point of the plurality of functional connections, a group of subjects showing a treatment effect among the plurality of subjects from a group of subjects not showing a treatment effect among the plurality of subjects. The storage device is configured to store information for identifying the second classifier generated by the processor.

[0098] One embodiment of the present invention relates to a method for generating a second classifier. In this embodiment, for a plurality of subjects, the correlation at a first time point of a plurality of functional connections selected from functional connection identification numbers 1 to 12 shown in Table 4 and the correlation at a second time point of the plurality of functional connections are measured, where the second time point is set after the start of treatment and later than the first time point. In this embodiment, the method includes the following steps: pre-generating a classifier, which is configured to distinguish, in a correlation state space expanded by the difference between the correlation at the first time point and the correlation at the second time point of the plurality of functional connections, a group of subjects showing a treatment effect among the plurality of subjects from a group of subjects not showing a treatment effect among the plurality of subjects.

[0099] One embodiment of the present invention relates to a brain activity training device, including a brain activity detection device, a prompting device, a processor, and a storage device. In this embodiment, the brain activity detection device is configured to detect signals representing the brain activities of multiple predetermined regions of the brain of a trainee undergoing neurofeedback training in chronological order. The storage device is configured to store: information for identifying, via machine learning, among the functional connections of the multiple predetermined regions of each brain of the multiple participants, the functional connections to be trained through feature selection for the disease label for discriminating depressive symptoms, based on signals obtained by measuring in advance and in chronological order the signals representing the brain activities of the multiple predetermined regions of each brain of multiple participants including healthy individuals and depressive patients in a resting state; and the target pattern of the functional connections to be trained in the neurofeedback training. The processor is configured to execute the processing of the neurofeedback training, which includes: calculating the temporal correlation of the functional connections to be trained within a predetermined time period based on the signals detected by the brain activity detection device; calculating a reward value according to the similarity to the target pattern based on the calculated temporal correlation; and prompting the trainee via the prompting device with information indicating the magnitude of the reward value. In this case, the functional connections to be trained include a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and each predetermined region among the multiple predetermined regions of the trainee's brain corresponds to each predetermined region among the multiple predetermined regions of each brain of the multiple participants.

[0100] One embodiment of the present invention relates to a method for controlling a brain activity training device, the brain activity training device including a brain activity detection device, a prompting device, a processor, and a storage device. In this embodiment, the brain activity detection device is configured to detect signals representing the brain activities of multiple predetermined regions of the brain of a trainee in a neurofeedback training in chronological order, and the storage device is configured to store: information for identifying, via machine learning, a functional connection to be trained among multiple functional connections selected by feature selection for a disease label for discriminating depressive symptoms from the functional connections of the multiple predetermined regions of the brains of the multiple participants, based on signals obtained by measuring in advance and in chronological order signals representing the brain activities of the multiple predetermined regions of the brains of multiple participants including healthy individuals and depressive patients in a resting state; and a target pattern of the functional connection to be trained in the neurofeedback training. The method includes the following steps: using the brain activity detection device to detect, in chronological order, signals representing the brain activities of multiple predetermined regions of the brain of the trainee in the neurofeedback training; using the processor to calculate, within a predetermined time period, the temporal correlation of the functional connection to be trained based on the signals detected by the brain activity detection device; using the processor to calculate a reward value based on the calculated temporal correlation according to the similarity to the target pattern; and using the processor to prompt, via the prompting device, information indicating the magnitude of the reward value to the trainee. In this case, the functional connection to be trained includes a first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and each predetermined region among the multiple predetermined regions of the trainee's brain corresponds to each predetermined region among the multiple predetermined regions of the brains of the multiple participants.

[0101] Effects of the Invention

[0102] According to the present invention, a discrimination device and a discrimination method for an index for objectively discriminating the state of depressive symptoms with respect to the activity state of the brain can be provided. According to the present invention, information indicating the degree of the treatment effect on depressive symptoms can be objectively discriminated. According to the present invention, a classifier can be provided that is configured to output an index value serving as a biomarker for objectively discriminating the state of depressive symptoms. According to the present invention, the level of depressive symptoms can be discriminated. According to the present invention, depressive patients can be objectively classified.

[0103] According to the present invention, neurofeedback training can be performed. BRIEF DESCRIPTION OF THE DRAWINGS

[0104] Figure 1 is a schematic diagram for showing the overall structure of the MRI device 10.

[0105] Figure 2 It is a hardware block diagram of the data processing unit 32.

[0106] Figure 3 It is a conceptual diagram for showing the process of extracting a correlation matrix representing the correlation between functional connections in the resting state of the region of interest.

[0107] Figure 4 It is a conceptual diagram for explaining the process of generating a first classifier based on the correlation matrix.

[0108] Figure 5 It is for performing as Figure 4 shown, the processing of generating a first classifier and the discrimination processing using the generated first classifier, and is a functional block diagram.

[0109] Figure 6 It is a flowchart for explaining the processing that the data processing unit 32 needs to perform in order to generate a first classifier.

[0110] Figure 7 It is a conceptual diagram for explaining inner-loop feature extraction.

[0111] Figure 8 It is a diagram for explaining the concept of inner-loop feature extraction processing.

[0112] Figure 9 It is a diagram for showing the result of the iterative processing when performing inner-loop feature extraction for specific hyperparameters λ1 and λ2.

[0113] Figure 10 It is a flowchart for explaining the inner-loop feature extraction processing in more detail.

[0114] Figure 11 It is a conceptual diagram for explaining outer-loop feature extraction.

[0115] Figure 12 It is a flowchart for explaining the outer-loop feature extraction processing in more detail.

[0116] Figure 13 It is a diagram for showing the concept of the processing of generating a classifier in step S108.

[0117] Figure 14 It is a flowchart for discriminating the depressive symptoms of the subject.

[0118] Figure 15 It is a functional block diagram for showing an exemplary situation of processing data collection, estimation processing, and measurement of the brain activity of the subject in a distributed manner.

[0119] Figure 16 It is a flowchart for discriminating the level of depressive symptoms of the subject.

[0120] Figure 17-1 is a flowchart for discriminating the treatment effect of the subject to be examined.

[0121] Figure 17-2 is a flowchart for discriminating the treatment effect of the subject to be examined.

[0122] Figure 18 is a flowchart for discriminating sub - types of depression.

[0123] Figure 19 is a functional block diagram for performing the process of generating a classifier and the classification process using the generated classifier.

[0124] Figure 20 is for showing the processing in a distributed manner Figure 19 an exemplary situation of the data collection, estimation processing, and measurement of the brain activity of the subject to be examined shown.

[0125] Figure 21 is a flowchart for generating a second classifier.

[0126] Figure 22 is a flowchart for judging the treatment effect by using the second classifier.

[0127] Figure 23 is a functional block diagram for performing the process of generating a second classifier and the treatment effect discrimination process using the generated second classifier.

[0128] Figure 24 is a flowchart for judging the treatment effect for the subject who has been determined to have a sub - type of depression by using the second classifier.

[0129] Figure 25 shows the 12 pairs of functional connections selected.

[0130] Figure 26 is for showing the distribution of the magnitudes of 54 non - zero c i k of the graph.

[0131] Figure 27 of a shows the distribution of the relevant weighted sum of the Hiroshima city cohort. Figure 27 of c shows the distribution of the relevant weighted sum of the Chiba city cohort. Figure 27 of c shows the result of considering the discrimination record by replacing the training data and the test data, where: *p < 0.05, and **p < 0.01.

[0132] Figure 28 of d shows the smoothed histogram of the relevant weighted sum of the melancholic MDD group, and Figure 28e shows the smoothed histogram of the relevant weighted sum for the non - melancholic MDD group, where: *p < 0.05, and **p < 0.01.

[0133] Figure 29 f shows the smoothed histogram of the relevant weighted sum for the treatment - resistant MDD group, Figure 29 g shows the smoothed histogram of the relevant weighted sum for the ASD group, and Figure 29 h shows the smoothed histogram of the relevant weighted sum for the SSD group, where: *p < 0.05, and **p < 0.01.

[0134] Figure 30 shows the results of the permutation test. Figure 30 a shows the histogram of the permutation test (repeated 1000 times) by LOOCV for the training data. Figure 30 b shows the accuracy of the out - of - sample test data set, and the binomial distribution is shown as a curve. The accuracy of the melancholic MDD classifier trained and validated without permutation is represented by a vertical line. The results of the permutation test are significant for LOOCV (p = 0.049) and out - of - sample validation (p = 0.036), where: *p < 0.05.

[0135] Figure 31 a shows the correlation between the BDI score and the discrimination result using the first classifier (all MDD and healthy individuals). Figure 31 b shows the correlation between the BDI and the discrimination result using the first classifier (all MDD). Figure 31 c shows the smoothed histogram of the relevant weighted sum before and after treatment. Figure 31 d shows the correlation between the change in the BDI score after drug administration and the change in the judgment result using the first classifier.

[0136] Figure 32 a shows the differences in the correlations of various functional connections between the healthy control group and the MDD group, as well as the differences in the correlations before and after drug administration. Figure 32 b shows the differences in the correlations in terms of FC1 and FC2 before and after drug administration for the Hiroshima cohort. Figure 32 c shows the differences in the correlations in terms of FC1 and FC2 before and after drug administration for the Chiba cohort. Figure 32 d shows the differences in the correlations in terms of FC1 and FC2 before and after drug administration for the healthy individual cohort.

[0137] Figure 33e shows the difference in the correlation of FC1 and FC2 before and after administration between the remission group and the non-remission group after administration. Figure 33 f shows the distribution of signed ΔFC1 and ΔFC2 in the administration group.

[0138] Figure 34 is a diagram for showing the concept of the structure of the brain activity training device.

[0139] Figure 35 shows an example of the display on the monitor for representing the proximity between the index value and the target value.

[0140] Figure 36 is a diagram for showing an example of the training sequence in neurofeedback.

[0141] Figure 37 a shows the neurofeedback score during the training period of MDD participants. Figure 37 b shows the HDRS score of MDD participants before and after training. Figure 37 c shows the neurofeedback score during the training period of subclinical depression participants. Figure 37 d shows the correlation between the change amount of the BDI score after training and the change amount of rs-fc MRI.

[0142] Figure 38 is a flowchart for showing neurofeedback training. Detailed Description

[0143] Now, an explanation of the embodiments of the present invention will be given with reference to the drawings. In the following description, components and processing steps having the same reference numerals represent the same or corresponding components and processing steps, and the description of these components and processing steps will not be repeated unless necessary.

[0144] 1. Explanation of Terms

[0145] In the present invention, depressive symptoms include at least one selected from the following: depressive mood; decreased interest; decreased willpower; impatience; depression; weakened thinking ability, concentration, or decision-making ability; sense of worthlessness or guilt; suicidal thoughts, suicidal ideation, or suicidal attempts; pathological thinking; delusions; physical symptoms (such as general discomfort, headache, dizziness, pain in various parts of the body such as back pain, palpitations, shortness of breath, loss of appetite, and weight loss); and sleep disorders. Depressive symptoms preferably include symptoms associated with major depressive disorder (MDD) based on the Diagnostic and Statistical Manual of Mental Disorders (DSM)-IV.

[0146] In the present invention, depression is not restricted, as long as it is accompanied by depressive symptoms, but preferably it is MDD. MDD includes melancholic MDD, non-melancholic MDD, and treatment-resistant MDD. Here, MDD is sometimes simply referred to as "depression".

[0147] As an evaluation of the degree of depression, for example, hitherto, for screening during medical examinations or auxiliary materials during consultations, an evaluation using the Beck Depression Inventory has been used. In addition, the Hamilton Depression Rating Scale (HDRS), which is used as a rating scale for doctors, is a multi-item questionnaire that provides an index for depression and is used as an index for evaluating recovery. Its abbreviation is "HAM-D".

[0148] In the present invention, although the subject is not restricted, the subject is preferably a person with depressive symptoms. Age or gender is not restricted. The subject can be a person who has not received treatment for improving depressive symptoms, or can be a person who has received such treatment.

[0149] Treatment for improving depressive symptoms includes at least one selected from drug therapy, neurofeedback therapy, electroconvulsive therapy without convulsions, and repetitive transcranial magnetic stimulation. Examples of drugs to be administered can be at least one selected from the following: tricyclic antidepressants (e.g., imipramine, trimipramine, clomipramine, amitriptyline, nortriptyline, amoxapine, lofepramine, and dosulepin), tetracyclic antidepressants (e.g., maprotiline, mianserin, and setiptiline), trazodone, selective serotonin reuptake inhibitors (e.g., escitalopram, fluvoxamine, paroxetine, and sertraline), serotonin-norepinephrine reuptake inhibitors (e.g., milnacipran and duloxetine), noradrenergic and specific serotonergic antidepressants (e.g., mirtazapine), and benzamide drugs (e.g., sulpiride).

[0150] There is no concept of "cure" for the treatment of depressive symptoms. Therefore, the improvement of depression means, for example, improvement in clinical findings such as the BDI, improvement in depressive symptoms compared to the past state, or being in a "remission" state in terms of clinical findings.

[0151] Embodiments of the present invention include: assisting in judging the level of depressive symptoms of a subject, assisting in judging whether a subject has depressive symptoms, assisting in judging the treatment effect of a subject, and assisting in classifying a subject into a subclass of a disease.

[0152] In the present invention, drug re-profiling refers to: detecting the efficacy on depressive symptoms for a drug that has shown efficacy on another disease or symptom in humans, or a drug that is weakly toxic to humans but has not shown obvious efficacy and the effect on depressive symptoms is unknown.

[0153] 2. Camera Conditions for Resting-State fMRI and Extraction of Correlation Matrix

[0154] The camera equipment for resting-state fMRI is not restricted. The camera conditions are also not restricted as long as fMRI images can be acquired. For example, the magnetic field is about 3.0T, the field of view is from about 192 mm to about 256 mm, the matrix is about 64×64, the number of slices is from about 30 to about 40, the number of volumes is from about 112 to about 244, the slice thickness is from about 3.0 mm to about 4.0 mm, the slice gap is from about 0 mm to about 0.8 mm, the TR is from about 2000 ms to about 2700 ms, the TE is from about 25 ms to about 31 ms, the total camera time is from about 5 minutes to about 10 minutes, the flip angle is from about 75 degrees to about 90 degrees, and the slice acquisition order is ascending (interleaved). The camera is preferably performed under dark illumination. In addition, the subject is preferably kept awake without thinking about anything during the camera. In addition, the subject is preferably kept looking at the cross mark in the center of the monitor screen during the camera.

[0155] The captured fMRI images can be processed by the method described by Yahata et al. in the literature (Nature Communications|7:11254|DOI:10.1038 / ncomms11254).

[0156] The captured fMRI image data is not particularly restricted. However, for example, SPM8 (Wellcome Trust Center for Neuroimaging, University College London, UK) of Matlab R2014a (Mathworks Inc., USA) can be used for standard preprocessing of T1-weighted structural images and resting-state functional images. For example, the captured images can be realigned, slice timing corrected, co-registered, normalized, and smoothed (FWHM = 8 mm). In addition, for all image data of each subject, images that are judged to have moved more than 0.5 mm relative to the previous image can be excluded from the analysis.

[0157] Based on the preprocessed image data, functional connectivity (FC) (i.e., connection strength) is calculated for multiple predetermined regions of interest (ROIs) in the brain. Functional connectivity is a feature (element of the correlation matrix) commonly used in resting-state brain activity analysis and is defined by the Pearson correlation coefficient between the time series signals of different regions of interest. For example, the connection strength can be represented by the average value of the Pearson correlation coefficient based on the values measured within a predetermined time period, although the representation method is not particularly restricted. Optionally, the connection strength can be represented by other statistics for the Pearson correlation coefficient within a predetermined time period.

[0158] As described later, in the present embodiment, in order to determine the state of depressive symptoms, for example, functional connectivity (connection strength) is calculated for all or a part of the regions of interest having functional connectivity identification numbers (the "ID" in Table 1) 1 to 12 shown in Table 1.

[0159] [Table 1]

[0160]

[0161] In Table 1, "L" and "R" in "Lat." represent the left and right brains in a distinguishable manner. "BSA" represents the Brodmann area, and "BA" represents the number of Brodmann areas. "Weight" represents the weight of the correlation weighted sum (hereinafter sometimes simply referred to as the "weighted sum") described later.

[0162] The extraction of the elements of the correlation matrix is not particularly limited, but for example, the following process can be adopted.

[0163] First, the time series average signals of all the voxels included in each region of interest are extracted. Then, the time series average signals are passed through a band-pass filter (0.008 Hz to 0.1 Hz) to remove the noise of these signal values. After that, regression is performed using 9 explanatory variables (the average signals of the whole brain, white matter, cerebrospinal fluid, and 6 body movement correction parameters). The residual sequence after regression is regarded as the time series signal value related to the functional connectivity, and this time series signal value is set as the element of each region of interest, and the Pearson correlation coefficient of the time series is calculated for each element of each correlation matrix of each pair of regions of interest. This correlation coefficient is a value representing the connection strength of the functional connectivity, and the connection strength corresponding to each pair is obtained.

[0164] The correlation coefficient of each functional connectivity is input as input data to the classifier described later, and based on the correlation coefficient and the coefficient representing the weight (contribution degree) pre-calculated for each functional connectivity for performing the classification process using the classifier, an index value for determining the disease label of depressive symptoms for this functional connectivity is calculated. In other cases, based on this correlation coefficient and this coefficient, the correlation weighted sum of multiple functional connectivities is calculated as the index value for determining the disease label of depressive symptoms. The term "correlation weighted sum" here refers to the value obtained by multiplying multiple functional connectivities by the corresponding weights and taking their sum.

[0165] Therefore, the index value is not the data obtained by simply measuring the brain activity of the subject, but a value artificially calculated in consideration of the weights of each functional connectivity. The index value is used for the determination of the label of depressive symptoms, the determination of the level of depressive symptoms, the determination of the treatment effect, or the classification of patients with depression.

[0166] Among the 12 pairs of functional connections, the functional connection with the functional connection identification number 1 contributes the most to depressive symptoms. The contribution of the functional connection with the functional connection identification number 2 is the second largest. Therefore, in each of the embodiments described below, at least one or both of the functional connection identification number 1 and the functional connection identification number 2 can be selected for use.

[0167] The 12 pairs of functional connections are particularly suitable for discriminating between the melancholic MDD patient group and the healthy control group.

[0168] 3. Generation of the first classifier and discrimination device

[0169] The first classifier is generated through a classifier generation process based on signals obtained by using a brain activity detection device to measure in advance and in a time series manner the brain activities of multiple predetermined regions of each brain of multiple participants in a resting state. These multiple participants include healthy individuals and patients with depression. The term "patient with depression" refers to a participant who has been diagnosed with depression in a doctor's diagnosis in advance and is associated with the "disease label" of depression. The classifier is generated to discriminate the disease label of depressive symptoms based on the weights of the functional connections selected as being related to the disease label of depressive symptoms through feature selection from the functional connections of multiple predetermined regions through machine learning.

[0170] Specifically, as described later, from the functional connections of multiple regions of interest set in advance, the functional connections specifically related to the disease label of depressive symptoms extracted through sparse canonical correlation analysis, the functional connections to be used for discriminating the disease label of depression are selected through feature selection using sparse logistic regression. Then, the above-mentioned relevant weighted sum is calculated based on the functional connections selected in this way.

[0171] 3-1. Discrimination device 1

[0172] For example, the first classifier is generated based on the fMRI image data obtained from the Figure 1 MRI device 10 shown.

[0173] Figure 1 is a schematic diagram showing the overall structure of the MRI device 10 as the discrimination device 1 according to the first embodiment of the present invention.

[0174] As Figure 1As shown, the MRI apparatus 10 includes: an MRI imaging unit 25 configured to perform MRI imaging; and a data processing unit 32 configured to set a control sequence for the MRI imaging unit 25 and process various data signals to generate an image. The MRI imaging unit 25 includes: a magnetic field applying mechanism 11 configured to apply a magnetic field targeted at a region of interest of the subject 2 to irradiate an RF wave; a receiving coil 20 configured to receive a response wave (NMR signal) from the subject (trainee) 2 and output an analog signal; and a driving unit 21 configured to control the magnetic field to be applied to the subject 2 and control the transmission or reception of the RF wave.

[0175] The central axis of the cylindrical cavity in which the subject 2 is to be placed is set as the Z-axis, and the horizontal and vertical directions orthogonal to the Z-axis are set as the X-axis and Y-axis, respectively.

[0176] The MRI apparatus 10 has such a structure that the nuclear spins of the nuclei of the subject 2 are oriented in the magnetic field direction (Z-axis) by the static magnetic field applied by the magnetic field applying mechanism 11 and precess at the Larmor frequency specific to these nuclei in a state with the magnetic field direction as the axis.

[0177] Then, when an RF pulse having the same frequency as the Larmor frequency is irradiated, the atoms resonate with each other to absorb energy and are excited, resulting in the occurrence of the nuclear magnetic resonance (NMR) phenomenon. When the irradiation of the RF pulse stops after this resonance, the atoms output an electromagnetic wave (NMR) signal having the same frequency as the Larmor frequency during the relaxation process of emitting energy and returning to their original steady state.

[0178] The receiving coil 20 receives the output NMR signal as a response wave from the subject 2, and the data processing unit 32 processes the region of interest of the subject 2 to form an image.

[0179] The magnetic field applying mechanism 11 includes a static magnetic field generating coil 12, a gradient magnetic field generating coil 14, an RF irradiation unit 16, and a bed 18 for placing the subject 2 in the cavity.

[0180] For example, the subject 2 lies supine on the bed 18. Although not particularly limited, the subject 2 can, for example, use prism glasses 4 to view the screen displayed on a cue device 6 (e.g., a display) vertically mounted with respect to the Z-axis. The subject 2 receives visual stimuli through the image of the cue device 6. The subject 2 can receive visual stimuli through a structure in which an image is projected by a projector in front of the subject 2.

[0181] Such visual stimuli correspond to the cue of the feedback information in the above-described neurofeedback.

[0182] The drive unit 21 includes a static magnetic field power supply 22, a gradient magnetic field power supply 24, a signal transmission unit 26, a signal reception unit 28, and a bed drive unit 30, and the bed drive unit 30 is configured to move the bed 18 to any position in the Z-axis direction.

[0183] The data processing unit 32 includes: an input unit 40 configured to receive various operation or information inputs from an operator (not shown); a display unit 38 configured to display various images or various information related to the region of interest of the subject 2 on a screen; a storage unit 36 configured to store programs for performing various processes, control parameters, image data (e.g., structural images), and other electronic data; a control unit 42 configured to control the operations of the respective functional units, such as generating a control sequence for driving the drive unit 21; an interface unit 44 configured to perform transmission / reception of various signals with respect to the drive unit 21; a data collection unit 46 configured to collect data including a set of NMR signals originating from the region of interest; an image processing unit 48 configured to form an image based on the NMR signal data; and a network interface unit 50 configured to perform communication via a network.

[0184] In addition, except in the case where the data processing unit 32 is a dedicated computer, the data processing unit 32 can also be a general-purpose computer configured to execute functions for operating the respective functional units, and can perform specified calculations, data processing, or generation of a control sequence based on the programs installed in the storage unit 36. Now, an explanation will be given based on the assumption that the data processing unit 32 is a general-purpose computer.

[0185] The static magnetic field generating coil 12 is configured such that a current supplied from the static magnetic field power supply 22 flows through a helical coil wound around the Z axis to generate an induced field, thereby generating a static magnetic field in the Z-axis direction in the cavity. The region of interest of the subject 2 is set in a region where the static magnetic field formed in the cavity is highly constant. More specifically, the static magnetic field generating coil 12 includes, for example, four air-core coils, and the combination of these four air-core coils forms a constant magnetic field inside, and the spins of predetermined atomic nuclei (more specifically, the spins of hydrogen atomic nuclei) in the body of the subject 2 are given an orientation.

[0186] The gradient magnetic field generating coil 14 includes an X coil, a Y coil, and a Z coil (not shown), and is provided on the inner peripheral surface of the static magnetic field generating coil 12 having a cylindrical shape.

[0187] These X, Y, and Z coils sequentially switch the X-axis direction, Y-axis direction, and Z-axis direction respectively, thereby superimposing a gradient magnetic field on the constant magnetic field in the cavity and applying an intensity gradient to the static magnetic field. The Z coil applies a gradient to the magnetic field intensity in the Z-axis direction during excitation to limit the resonance plane. The Y coil applies a gradient within a short time period immediately after applying the magnetic field in the Z-axis direction to add phase modulation proportional to the Y coordinate to the detection signal (phase encoding), and the X coil then applies a gradient during data acquisition to add frequency modulation proportional to the X coordinate to the detection signal (frequency encoding).

[0188] The switching of the gradient magnetic field to be superimposed is achieved by the transmission unit 24 outputting different pulse signals for each of the X, Y, and Z coils according to the control sequence. In this way, the position of the subject 2 exhibiting the NMR phenomenon can be identified, and the position information of the three-dimensional coordinates required to form an image of the subject 2 can be given.

[0189] As described above, the three orthogonal gradient magnetic fields can be respectively assigned the slice direction, phase encoding direction, and frequency encoding direction, thereby performing imaging from various angles through the combination of these three gradient magnetic fields. For example, in addition to transverse slices in the same direction as the imaging direction using an X-ray CT device, for example, sagittal slices and coronal slices orthogonal to the transverse slice, as well as oblique slices perpendicular to the plane and not parallel to the axes of the three orthogonal gradient magnetic fields, can also be imaged.

[0190] The RF irradiation unit 16 is configured to irradiate a radio frequency (RF) pulse to the region of interest of the subject 2 based on the high-frequency signal transmitted from the signal transmission unit 33 according to the control sequence.

[0191] In Figure 1 the RF irradiation unit 16 is incorporated into the magnetic field application mechanism 11, but it can be provided in the bed 18 or integrated with the receiving coil 20.

[0192] The receiving coil 20 is configured to detect the response wave (NMR signal) from the subject 2 and is arranged near the subject 2 to detect the NMR signal with high sensitivity.

[0193] In the receiving coil 20, a tiny current is generated based on electromagnetic induction when the electromagnetic wave of the NMR signal cuts the coil wire. This tiny current is amplified in the signal receiving unit 28 and converted from an analog signal to a digital signal for transmission to the data processing unit 32.

[0194] That is to say, the following structure is adopted. When the RF irradiation unit 16 applies a high-frequency electromagnetic field with a resonance frequency to the subject 2 in a state where a Z-axis gradient magnetic field is applied to the static magnetic field, predetermined atomic nuclei (e.g., hydrogen atomic nuclei) in a part that satisfies the resonance condition in terms of the intensity of the magnetic field are selectively excited and start to resonate. The predetermined atomic nuclei in the part that satisfies the resonance condition (e.g., a cross-section with a predetermined thickness of the subject 2) are excited, and the spins rotate simultaneously. When the excitation pulse stops, in the receiving coil 20, the electromagnetic wave emitted by the rotating spins then excites a signal, which is detected and lasts for a period of time. Using this signal, tissues containing predetermined atoms in the body of the subject 2 are observed. Then, by applying X and Y gradient magnetic fields, the signal is detected to determine the emission position of the signal.

[0195] The image processing unit 48 measures the detection signal while repeatedly generating the excitation signal based on the data constructed in the storage unit 36, reduces the resonance frequency to the X coordinate through the first Fourier transform calculation, restores the Y coordinate through the second Fourier transform to obtain an image, and displays the corresponding image on the display unit 38.

[0196] For example, using such an MRI system, the above-mentioned BOLD signal is captured in real time, and the control unit 42 performs the analysis processing described later on the images captured in chronological order, thereby enabling the capture of functional connectivity MRI (rs-fc MRI) in the resting state.

[0197] Figure 2 It is a hardware block diagram of the data processing unit 32.

[0198] As described above, the hardware of the data processing unit 32 is not particularly limited, and a general-purpose computer can be used as the hardware.

[0199] In Figure 2In addition to the memory drive 2020 and the disk drive 2030, the computer main body 2010 of the data processing unit 32 further includes: a processor (central processing unit: CPU) 2040; a bus (also referred to as an “interface”) 2050 connected to the disk drive 2030 and the memory drive 2020; a read-only memory (ROM) 2060 for storing programs (e.g., a boot program); a random access memory (RAM) 2070 for temporarily storing commands of application programs and providing a temporary storage space; a non-volatile storage device 2080 for storing application programs, system programs, and data; and a communication interface 2090. The communication interface 2090 corresponds to the interface unit 44 for transmitting / receiving signals with respect to, for example, the drive unit 21, and the network interface 50 for communicating with other computers via a network (not shown). As the non-volatile storage device 2080, a hard disk drive (HDD) or a solid state drive (SSD) or the like can be used. The non-volatile storage device 2080 corresponds to the storage unit 36. The computer main body 2010 of the data processing unit 32 also serves as a first classifier generation device and / or a second classifier generation device.

[0200] The CPU 2040 executes arithmetic processing based on programs to implement the respective functions of the data processing unit 32, such as the respective functions of the control unit 42, the data collection unit 46, and the image processing unit 48.

[0201] A program for causing the data processing unit 32 to execute the functions of the above-described embodiments is stored in a CD-ROM 2200 or a storage medium 2210, and inserted into the disk drive 2030 or the memory drive 2020. Further, the program can be transferred to the non-volatile storage device 2080. The program is loaded onto the RAM 2070 when it is executed.

[0202] The data processing unit 32 further includes a keyboard 2100 and a mouse 2110 serving as input devices, and a display 2120 serving as an output device. The keyboard 2100 and the mouse 2110 correspond to the input unit 40, and the display 2120 corresponds to the display unit 38.

[0203] A program for implementing such functions of the data processing unit 32 as described above may not always include an operating system (OS) for executing the functions of, for example, the information processing device in the computer main body 2010. The program may only include commands for calling appropriate functions (modules) in a controlled mode to obtain a desired result. The operation mode of the data processing unit 32 is well known, and thus a detailed description of the operation mode is omitted here.

[0204] In addition, there may be one or more computers for executing the program. That is, either centralized processing or distributed processing can be performed.

[0205] Figure 3 It is a conceptual diagram showing a process of extracting a correlation matrix representing the correlation between the connection functions in the resting state related to the disease label of depressive symptoms.

[0206] As Figure 3 shown, based on the fMRI data corresponding to n (n: natural number) time points in the resting state measured in real time, the average "activity level" of the region of interest is calculated, and the correlation value of the activity level of the brain region (region of interest) is calculated.

[0207] In this case, 137 regions other than the cerebellar region are regarded as the regions of interest. Therefore, considering symmetry, the number of independent non - diagonal components in the correlation matrix is:

[0208] (137×137 - 137) / 2 = 9316.

[0209] In Figure 3 it, the correlation only represents 34×34 components.

[0210] The calculation of the elements of this correlation matrix is not particularly limited. For example, it can be performed as follows.

[0211] Based on the resting - state brain activity data, the functional connectivity between different regions of interest is calculated for each participant. Functional connectivity is a feature commonly used in resting - state brain activity analysis and is defined by the Pearson correlation coefficient between the time - series signals of different regions of interest.

[0212] First, the time - series average signal of all voxels included in each region of interest is extracted.

[0213] Next, the time - series average signal is passed through a band - pass filter to remove the noise of these signal values. Then, regression is performed using 9 explanatory variables (the average signals of the whole brain, white matter, and cerebrospinal fluid, and 6 body motion correction parameters).

[0214] The residual sequence after regression is regarded as the time - series signal value related to functional connectivity, and the Pearson correlation coefficient of the time - series is calculated for different ROIs.

[0215] As the regions of interest, 137 regions of interest included in the brain sulcus atlas (BAL) are used. The functional connectivity FC between these 137 regions of interest is used as a feature.

[0216] Regarding the brain sulcus atlas, the following published literature is known.

[0217] Literature: Perrot et al., Med Image Anal, 15(4), 2011

[0218] Literature: Tzourio-Mazoyer et al., Neuroimage, 15(1), 2002

[0219] Figure 4 is a conceptual diagram for explaining the process of generating a first classifier as a biomarker based on such a correlation matrix as Figure 3 shown.

[0220] As described below, for example, the first classifier serves as an indicator for a diagnostic marker of "depression" and can be used to assist doctors in diagnosing the severity or degree of progression of a disease. Thus, a device configured to calculate the first classifier and output the result can also be referred to as a "diagnostic assistance device".

[0221] As Figure 4 shown, the data processing unit 32 derives elements of a correlation matrix of the activity levels of brain regions (regions of interest) for each participant through a process described later, based on fMRI data measured for participants including a group of healthy individuals and a group of depression patients captured by the MRI imaging unit 25.

[0222] Next, the data processing unit 32 extracts features for the correlation matrix and the attributes of the participants (including the disease / healthy individual label of the participant) through regularized canonical correlation analysis. In machine learning or statistics, the term "regularization" generally refers to the following method: adding a regularization term weighted by a hyperparameter to the error function and suppressing the complexity or degrees of freedom of the model, thereby preventing overfitting. As a result of regularized canonical correlation analysis, when the explanatory variables are also sparsified, regularized canonical correlation analysis is specifically referred to as "sparse canonical correlation analysis (SCCA)". Now, a specific example of performing sparse canonical correlation analysis will be described.

[0223] Then, in this sparse canonical correlation analysis, as described later, the value of the hyperparameter is adjusted to generate a canonical variable that is only connected to the "disease label", and the functional connectivity FC connected to the corresponding canonical variable is extracted. The union of the functional connectivities FC extracted within the range where there is a canonical variable satisfying this condition when the hyperparameter is changed within a predetermined range is called the "first union".

[0224] In addition, the "first union set" obtained as a result of the sparse canonical correlation analysis using the data processing unit 32 is set as an explanatory variable, and for example, in each cross-validation step, discriminant analysis by sparse logistic regression is performed while performing leave-one-out cross-validation (LOOCV). The union set of the functional connections FC extracted as explanatory variables in all cross-validations is called the "second union set".

[0225] Finally, for the data related to all participants, discriminant analysis by sparse logistic regression is performed on the "disease label" as the target variable with the "second union set" used as the explanatory variable, thereby generating a first classifier.

[0226] Figure 5 is a functional block diagram for performing the process of generating the first classifier as shown in Figure 4 and the discriminant process using the generated first classifier.

[0227] First, the non-volatile storage device 2080 stores the participant rs-fc MRI measurement data 3102 and a plurality of personal attribute information 3104 associated with each participant who has measured the MRI measurement data, where the participant rs-fc MRI measurement data 3102 is information obtained by pre-measuring signals representing the brain activities of a plurality of predetermined regions of the brains of a plurality of participants including healthy individuals and patients with depression in chronological order using an MRI device.

[0228] The term "personal attribute information" here includes "personal characteristic information" for identifying the participant and "measurement condition information" for identifying the measurement conditions of each participant.

[0229] The term "personal characteristic information" refers to information related to the participant, such as disease label, age, gender, or medication history, etc.

[0230] The term "measurement condition information" refers to conditions for identifying measurement conditions such as information on the measurement location where the participant is measured (including information for identifying the measurement site and / or measurement device), whether the measurement is performed in an open-eye or closed-eye state during the measurement, or the measurement magnetic field strength, etc.

[0231] The processor 2040 performs the process of generating a classifier for the disease label based on the participant rs-fc MRI measurement data 3102 and the corresponding personal attribute information 3104.

[0232] The correlation matrix calculation unit 3002 calculates a correlation matrix of the functional connectivity of the brain activities in a plurality of predetermined regions for each participant based on the rs-fc MRI measurement data 3102 of the participants. For each participant, data related to the calculated correlation matrix of the functional connectivity is stored as the correlation matrix data 3106 of the functional connectivity in the non-volatile storage device 2080.

[0233] The first feature selection unit 3004 sequentially selects one subset from different K subsets (K: a natural number of 2 or more) extracted from a plurality of participants, and performs sparse canonical correlation analysis on a plurality of attribute information and the elements of the correlation matrix for the (K - 1) subsets other than the selected subset, thereby extracting specific attribute information in the plurality of personal attribute information, for example, the elements of the correlation matrix connected to the canonical variables corresponding only to the disease labels. In addition, the first feature selection unit 3004 obtains a first union set that is the union set of the elements of the extracted correlation matrix for the sequentially selected subsets, and stores the first union set as the first functional connectivity union set data 3108 in the non-volatile storage device 2080. The "first functional connectivity union set data" may also be the index of the elements corresponding to the first union set in the correlation matrix data 3106 for identifying the functional connectivity.

[0234] When the remaining participants other than the K subsets among the plurality of participants are set as a test set and the test set is divided into N different groups, the second feature selection unit 3006 calculates, by sparse logistic regression, a test classifier for estimating specific attribute information (for example, disease labels) for the set of participants other than the one selected group among the N groups based on the first union set, and extracts the elements of the correlation matrix that are the explanatory variables of the test classifier due to sparsification. The second feature selection unit 3006 further sequentially selects one group from the N groups, repeats feature extraction to obtain a second union set (which is the union set of the elements of the correlation matrix extracted as the explanatory variables of the test classifier), and stores the second union set as the second functional connectivity union set data 3110 in the non-volatile storage device 2080. The "second functional connectivity union set data" may also be the index of the elements corresponding to the second union set in the correlation matrix data 3106 for identifying the functional connectivity.

[0235] The number of elements in each of the N groups may be one.

[0236] The classifier generation unit 3008 calculates a first classifier for estimating specific attribute information (for example, disease labels) by sparse logistic regression with the second union set used as the explanatory variable. The classifier generation unit 3008 stores information for identifying the generated first classifier as the classifier data 3112 in the non-volatile storage device 2080.

[0237] The discrimination processing unit 3010 performs discrimination processing on the input data based on the classifier identified by the classifier data 3112.

[0238] In the above description, the second union set is set as the explanatory variable, and the classifier generation unit 3008 generates a classifier. However, for example, the classifier generation unit 3008 can directly generate a first classifier with the first union set used as the explanatory variable. However, as will be described later, from the perspective of dimensionality reduction and generalization performance, it is desirable to set the second union set as the explanatory variable.

[0239] In the generation of the first classifier, regularized logistic regression using a regularization method (e.g., L1 regularization or L2 regularization) can be used, and more specifically, for example, the above-mentioned "sparse logistic regression" can also be used. In addition, in the generation of the first classifier, for example, a support vector machine or linear discriminant analysis (LDA) can be used. The following gives an explanation by taking sparse logistic regression as an example.

[0240] As will be described later, in parallel with the feature selection process using the second feature selection unit 3006, one group excluded from the N groups can be sequentially selected, and the test classifier calculated by the second feature selection unit 3006 can be used to calculate the discrimination result with the excluded group used as the test sample, thereby performing cross-validation.

[0241] In this way, the dimensionality of the explanatory variable is reduced through the process of feature selection in the nested structure, thereby enabling efficient reduction of the dimension in terms of time while using almost all the participant data in the process executed by the second feature selection unit 3006.

[0242] In addition, the test set in the process of the second feature selection unit 3006 is set independently of the data set used for dimensionality reduction, thereby being able to avoid extremely optimistic results.

[0243] Figure 6 It is a flowchart for explaining the processing that the data processing unit 32 needs to perform to generate a classifier as a biomarker.

[0244] Now, refer to Figure 6 to describe in more detail Figure 4 the processing shown.

[0245] The most critical issue in creating biomarkers based on the connections between brain regions derived from resting-state fMRI data and the discriminative labels (disease labels) of the diseases of the participants is that the dimensionality of the data is much larger than the number of data. Thus, when using the data set for a classifier to perform learning for predicting disease labels (labels indicating whether a participant has a disease are called "disease labels") without regularization, overfitting occurs, and the performance of predicting unknown data drops significantly.

[0246] On the other hand, in general machine learning, the process of explaining measurement data with a smaller number of explanatory variables is called "feature selection (or feature extraction)". In the present embodiment, in the "multiple correlation values (multiple connections) of the activity levels of brain regions (regions of interest)", this kind of feature selection (or feature extraction) that can construct a first classifier with a smaller number of correlation values (that is, selecting more important correlation coefficients as explanatory variables) can be called "extraction of a simplified representation" in the machine learning of the first classifier for predicting target disease labels.

[0247] In addition, in the present embodiment, a regularization method is used as the feature extraction method. In this way, in canonical correlation analysis, considering the execution of regularization, sparsification, and the process of selecting more important explanatory variables, this process is called "sparse canonical correlation analysis". More specifically, for example, as a regularization method for combining to achieve sparsification, a method of imposing a penalty on the magnitude of the absolute value of the parameters of canonical correlation analysis can be used, that is, "L1 regularization" as described later.

[0248] Specifically, referring to Figure 6 , when the data processing unit 32 starts the process of generating a classifier in response to an input of, for example, the start of processing from the input unit 40 (step S100), the data processing unit 32 reads the MRI measurement data for each participant from the storage unit 36 (step S102), and performs the process of extracting the first feature by sparse canonical correlation analysis (SCCA) (step S104).

[0249] Hereinafter, the process of step S104 will be called "inner-loop feature extraction".

[0250] Now, the following gives an explanation of "sparse canonical correlation analysis" and "inner-loop feature extraction" respectively to illustrate the process of inner-loop feature extraction in step S104.

[0251] (Sparse Canonical Correlation Analysis)

[0252] Now, L1-regularized canonical correlation analysis is described as sparse canonical correlation analysis. This L1-regularized canonical correlation analysis is disclosed in the following literature.

[0253] Literature: Witten DM, Tibshirani R, and T Hastie. A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis. Biostatistics, Vol. 10, No. 3, pp. 515 - 534, 2009。

[0254] First, in general canonical correlation analysis (CCA), for data pairs x1 and x2, the variables x1 and x2 are normalized to have a mean of 0 and a standard deviation of 1.

[0255] Generally speaking, CCA can be used to identify potential relationships between a pair of measured quantities.

[0256] Specifically, in CCA, projection vectors are retrieved to find the maximum correlation between a pair of projected variables (canonical variables).

[0257] In contrast, when applying L1 regularization to canonical correlation analysis, that is, when using sparse CCA (i.e., L1 - SCCA) caused by L1 - norm regularization, the following optimization problem is solved.

[0258] Now, consider the following variable combinations.

[0259] [Equation 1]

[0260] In the presence of a combination of N - element measurements each including variables x1 ∈ R p1 and x2 ∈ R p2 , assume X1 = [x1 1 , x1 2 , … x1 N T represents an N×p1 matrix as a combination of the first variable, and X2 = [x2 1 , x2 2 , … x2 N T represents an N×p2 matrix as a combination of the second variable.

[0261] In addition, in this case, assume that the columns forming matrices X1 and X2 are normalized to have a mean of 0 and a variance of 1.

[0262] Then, L1 - SCCA can be formulated as the following expression (1).

[0263] [Equation 2]

[0264] ​​ It satisfies

[0265] In the expressions given above, the hyperparameters λ1 and λ2 respectively represent the sparsity levels of the weight parameters w1 and w2 (referred to as "sparse projection vectors" because the corresponding variables are sparse due to w1 and w2).

[0266] Specifically, in the example of the present embodiment, it is assumed that two data matrices corresponding to the above variables are constructed to identify the potential relationship between human attribute information and functional connectivity FC.

[0267] Below, the participants include a group of patients with depression (disease label "major depressive disorder (MDD)") and a healthy control group (disease label "healthy control (HC)"), and the first row of the data matrix X1 represents the human attribute information (human characteristic information and measurement condition information) related to a participant, and for example, the characteristic information and measurement condition information include the following items.

[0268] i) Disease label (MDD or HC)

[0269] ii) Location information (indicating where the brain activity of the participant was measured, i.e., locations A, B, and C)

[0270] iii) Age

[0271] iv) Gender

[0272] v) Imaging condition (eyes open or closed)

[0273] vi) Medication status 1 (antipsychotic)

[0274] vii) Medication status 2 (antidepressant)

[0275] viii) Medication status 3 (sedative)

[0276] Specifically, the number of columns of the human attribute information data matrix X1 is 10, i.e., p1 = 10.

[0277] The first column includes 1 (= MDD) or 0 (= HC).

[0278] The next three columns represent the facilities (locations) where the brain activity was measured, and in this case, depending on the measurement at three different locations, it includes any one of

[100] (location A),

[010] (location B), and

[001] (location C).

[0279] The fifth column includes the value of the participant's age. The sixth column is information related to the participant's gender, and includes a value indicating either "1" (male) or "0" (female). The seventh column shows whether the participant opened or closed his eyes during the measurement in the measurement conditions, and includes a value of "1" (eyes open) or "0" (eyes closed). In addition, the last three columns include three pieces of condition information related to medication history, and each column includes a value of "1" (with a history of medication) or "0" (without a history of medication).

[0280] The human characteristic information and measurement condition information are not limited to the above-mentioned items, and may include other characteristics, such as information related to the participant's medication history of other drugs, or information related to other measurement conditions (e.g., the size of the applied magnetic field of an MRI device), etc., for example.

[0281] Assume that in the second data matrix X2, the elements of the non-diagonal lower triangular part of the correlation matrix representing the correlation (FC) between the functional connections of the participants are represented in a row vector format.

[0282] Furthermore, L1-SCCS is applied to the matrix pair X1 and X2, and then sparse projection vectors w1 and w2 are derived.

[0283] In addition, as described above, the first data matrix X1 is set to satisfy such a condition: when the hyperparameters λ1 and λ2 are set to predetermined values, the first data matrix X1 is projected to a typical variable of specific human attribute information (i.e., a typical variable having only a "disease label" in this case) through a sparse projection vector w1 due to sparsification. In this case, in the corresponding second data matrix X2, the sparse projection vector w2 is used to identify the index (element) of the correlation matrix element for identifying the functional connection associated only with the disease label.

[0284] That is, in the inner-loop feature extraction, the hyperparameters λ1 and λ2 of L1-SCCA are changed independently of each other, for example, between 0.1 and 0.9 in units of 0.1.

[0285] However, the range in which the hyperparameters λ1 and λ2 are set to be variable and the change units of the hyperparameters λ1 and λ2 are not limited to this example.

[0286] Regarding the processing of L1-SCCA, a range of hyperparameters λ1 and λ2 is found for which there are canonical variables connected only to the "disease" label.

[0287] The projection of the original correlation matrix elements onto the subspace defined by the non-zero elements of the derived sparse projection vector w2 is represented in the following manner.

[0288] Define variable i kto represent the index of the k-th non-zero element of the projection vector w2. In this case, 1 ≤ k ≤ m is satisfied, and m represents the number of non-zero elements.

[0289] Then, consider the projection matrix E obtained by projecting onto the following subspace.

[0290] [Equation 3]

[0291] E = [e i1 , e i2 , …, e im T

[0292] where: e ik ∈ R p2 is the standard basis vector that contains "1" as the i k -th element and contains "0" as the other elements.

[0293] Finally, project the original correlation matrix element vector x2 in the following manner to derive a vector in the subspace z ∈ R m .

[0294] [Equation 4]

[0295] z = Ex2 (2)

[0296] As a result, a specific number of features (correlation matrix elements) associated with the disease label (MDD / HC) can be selected.

[0297] Select the correlation matrix elements crucial for classification by choosing the correlation matrix elements corresponding to the canonical variables associated only with the diagnostic label.

[0298] (Inner loop feature extraction)

[0299] Figure 7 is a conceptual diagram for explaining the inner loop feature extraction.

[0300] In the inner loop feature extraction of step S104 in Figure 6 , as Figure 7 shown, a group of participants is divided into K sets (K: a natural number of 2 or more), and the above L1-SCCA is performed (K - 1) times on the remaining subsets except for one set in (K - 1) subsets.

[0301] That is, as Figure 7 ​As shown in (a) of, a subset corresponding to (K - 1) / K (e.g., 8 / 9 when K = 9) in the set of participants (hereinafter referred to as "inner - loop training data") is used for this inner - loop feature extraction. The remaining one of the K subsets is used as a "test pool" including test data to be used for training in the outer - loop feature extraction described later, and thus is not used in the inner - loop feature extraction.

[0302] As Figure 7 shown in (b) of, for the remaining subsets of the (K - 1) subsets except for one set (indicated by diagonal lines in (b) of), L1 - SCCA is performed by changing hyperparameters λ1 and λ2 in a predetermined range at a predetermined step size. Within a specific range of hyperparameters λ1 and λ2, functional connection elements FC related to canonical variables associated only with the "diagnosis" label are extracted as features. Figure 7 This extraction process is repeated for subsets of the (K - 1) subsets by sequentially changing the set to be excluded.

[0303] In each repetition, the union of the elements FC of the correlation matrix of the extracted functional connections is set as the "union of functional connection elements FC selected in the inner loop" (the first union).

[0304] Through this process, the inappropriate influence caused by differences in person - attribute information at different camera locations or camera conditions at different camera locations corresponding to the nuisance variable NV is reduced.

[0305] As described later, this process is used to construct a robust first classifier generalized to a foreign country (e.g., the United States) based on a first classifier generated from MRI measurement data obtained at multiple camera locations in, for example, Japan.

[0306] Next, referring back to

[0307] , the data processing unit 32 performs a second feature extraction process (step S106) using sparse logistic regression based on the results of the inner - loop feature extraction. Figure 6 The process in step S106 is hereinafter referred to as "outer - loop feature extraction".

[0308] Now, descriptions of "sparse logistic regression" and "outer - loop feature extraction" are given to explain the process of the outer - loop feature extraction in step S106.

[0309] (Sparse Logistic Regression)

[0310] (Sparse Logistic Regression)

[0311] Sparse logistic regression is a method that extends logistic regression analysis to the framework of Bayesian estimation and is a method for simultaneously performing dimensional compression of feature vectors and estimating the weights used for discrimination. This method is useful when the dimensionality of the feature vectors of the data is extremely large and contains a large number of unnecessary features. For unnecessary features, the weight parameters in linear discriminant analysis are set to 0 (i.e., feature extraction is performed), and only a small number of features related to discrimination are retrieved (sparsity).

[0312] In sparse logistic regression, for each class, the probability p of obtaining the acquired feature data belonging to the class to be classified is obtained, and the feature data is assigned to the class that outputs the maximum value. The probability p is output through the logistic regression expression. The estimation of the weights is performed by automatic relevance determination (ARD), and since the weights are close to 0, features that contribute less to class discrimination are excluded from the calculation.

[0313] Specifically, the first union of the features extracted by using the above-mentioned L1-regularized CCA is input, and a first classifier is used based on the next-level hierarchical Bayesian estimation to predict the disease label.

[0314] At this time, logistic regression is used as the classifier to predict the probability that the diagnostic label is a disease (diagnosed as autism in this case) based on the feature input z (the selected FC) extracted in the expression (2) given above.

[0315] [Mathematical formula 5]

[0316]

[0317] In the expression given above, y represents the diagnostic class / label. That is, y = 1 represents the MDD class, and y = 0 represents the HC class.

[0318] In addition, the following hat added to z (the " ^ " assigned to the top of the letter is called a " hat ") is a feature vector including the extended input.

[0319] [Mathematical formula 6]

[0320]

[0321] Based on the connectivity correlation matrix of the MRI samples of a participant in the resting state, the feature vector z is extracted according to the expression (2).

[0322] Using the extended input " 1 " is a standard method for introducing a certain (bias) input into the first classifier.

[0323] The following θ is the parameter vector of the logistic function.

[0324] [Mathematical formula 7]

[0325] θ ∈ R m+1

[0326] The distribution of the parameter θ in this case is set to the following normal distribution.

[0327] [Mathematical formula 8]

[0328] p(θ|α) = N(θ|0, diag(α))

[0329] In addition, the distribution of the hyperparameter α for setting the distribution of the parameter w is set in the following manner, and thus the distributions of the respective parameters are estimated by performing hierarchical Bayesian estimation.

[0330] [Mathematical formula 9]

[0331]

[0332] In the expressions given above, a 0 and b 0 represent the parameters of the gamma distribution for determining the hyperparameters. The symbol α is the parameter vector representing the variance of the normal distribution of the vector θ, and the i-th element of this vector is α i .

[0333] This sparse logistic regression is disclosed in the following literature.

[0334] Literature: Okito Yamashita, Masa aki Sato, Taku Yoshioka, Frank Tong, and Yukiyasu Kamitani. “Sparse Estimation automatically selects voxels relevant for the decoding of fMRI activity patterns.” NeuroImage, Vol. 42, No. 4, pp. 1414 - 1429, 2008.

[0335] (Outer ring feature extraction)

[0336] Figure 11 is a conceptual diagram for explaining outer ring feature extraction.

[0337] In Figure 6 the outer ring feature extraction in step S106 of Figure 11 as shown, during inner ring feature extraction, the set of participants is divided into K subsets (K: a natural number of 2 or more), and the remaining one subset other than the (K - 1) subsets used for inner ring feature extraction is used as a “test pool” including the test data to be used for training.

[0338] That is to say, as Figure 11 shown, for example, the test pool is divided into L groups, and one group is sequentially selected from these L groups (which is indicated by diagonal lines within the Figure 11 dotted square of

[0339] . Then, for the set of participants other than the selected group, the first union of the functional connection elements FC of the correlation matrix is used as an explanatory variable, and a first classifier for predicting disease labels is generated by sparse logistic regression. At this time, the "elements FC of the correlation matrix of functional connections" are additionally selected by sparse logistic regression. This is referred to as the "elements FC of the correlation matrix of functional connections selected in the outer loop".

[0340] In this way, one group is sequentially selected from the L groups, and this process is repeated L times to set the union of the extracted elements FC of the correlation matrix of functional connections as the "union of the elements FC of the functional connections selected in the outer loop" (the second union).

[0340] In this way, the test set to be used for outer loop feature extraction is always independent of the data set used for dimensionality reduction in inner loop feature extraction.

[0341] As an example, it is particularly assumed that each of the L groups includes one participant. In this case, in the above repeated process, the prediction process for the excluded one participant by the generated first classifier, and the process of accumulating the error between the disease labels and the prediction results for the excluded one participant correspond to performing the so-called leave-one-out cross-validation (LOOCV). Therefore, in Figure 11 the process of

[0342] (Generation of the first classifier)

[0343] Returning to the reference Figure 6 , the data processing unit 32 then performs the process of generating a classifier by sparse logistic regression based on the result of the outer loop feature extraction (the second union) (step S108).

[0344] Figure 13 is a diagram for explaining the concept of the process of generating a classifier in step S108.

[0345] As Figure 13 shown, in step S108, using the second union extracted from the first union through outer loop feature extraction as an explanatory variable, a classifier is generated for all participants by sparse logistic regression.

[0346] The classifier data (data related to the function form and parameters) 3112 for identifying the generated classifier is stored in the non-volatile storage device 2080, and then when MRI measurement data (test data) different from the MRI measurement data (test data) used in the above training is input, this classifier data is used for the discrimination process when estimating the disease label for the test data.

[0347] That is, based on the doctor's preliminary diagnosis, the correlation (connection) of the activity levels of the brain regions (regions of interest) of the participants classified into a group of healthy individuals and a group of depression patients is measured, and the following classifier is used as a biomarker for depressive symptoms. This classifier is generated by machine learning for the measurement results to discriminate whether the test data for new different participants corresponds to either depressive symptoms or a healthy state.

[0348] At this time, the classifier is generated by logistic regression, so the "disease label" as the output of the biomarker can include the probability of having a disease (or the probability of being healthy). For example, a display such as "the probability of having a disease is ○○%". This probability can be used as a "disease marker".

[0349] In addition, the attributes output by the classifier are not always limited to the discrimination of diseases, and can be the output of other attributes. In this case, the discrete discrimination result of the class to which the attribute belongs can be output, or the continuous value (e.g., probability) of the attributes inherent in certain classes can be output.

[0350] That is, in the learning (creation) of the classifier, in order to create a biomarker for depressive symptoms, rs-fc MRI data is input to extract features through the above-mentioned inner-loop feature extraction and outer-loop feature extraction, and the classifier generated using the extracted features as explanatory variables is used to discriminate between the depressive symptom label and the healthy individual label.

[0351] (Processing of inner-loop feature extraction)

[0352] Figure 8 is a diagram for explaining the concept of the inner-loop feature extraction process.

[0353] Figure 9 is an explanatory diagram showing the result of the iterative process when performing inner-loop feature extraction for specific hyperparameters λ1 and λ2 as an example.

[0354] Reference Figure 8 represents one repetition of L1-SCCA in the feature extraction with a nested structure, and in this case, the canonical variables are only connected to the "disease label".

[0355] In Figure 8In it, the fact that the canonical variable is only connected to the "disease label" is represented by connecting the disease label to only one canonical variable w1 T x1 with a solid line.

[0356] In addition, the symbol c T on the dotted line connecting the canonical variable w1 T x1 to the canonical variable w2 j x2 respectively represents the correlation coefficient between these canonical variables.

[0357] When the canonical variable w1 T x1 is only connected to the "disease label", the elements of the non-diagonal lower triangular part of the correlation matrix of the functional connections connected to the corresponding canonical variable w2 T x2 are extracted as features.

[0358] As an example, Figure 9 shows the combination of the minimum hyperparameters that generate at least one canonical correlation for each person's attribute information in the initial outer loop.

[0359] The canonical variables are represented by circles.

[0360] The circles in the left column represent the canonical variable w1 T x1 obtained from the person's characteristic information and measurement conditions. On the other hand, the circles in the right column represent the canonical variable w2 T x2 obtained from the functional connection (FC).

[0361] As described above, the numbers on the dotted lines connecting these canonical variables represent the correlation coefficient between the canonical variable w1 T x1 and w2 T x2.

[0362] The label of the person's attribute information and the connection to the canonical variable w2 T x2 are represented by solid or dotted lines.

[0363] (Processing of inner loop feature extraction)

[0364] Figure 10 is a flowchart for explaining the inner loop feature extraction process in more detail.

[0365] In the inner loop feature extraction, as shown in (a) of Figure 7 , the set of participants is divided into K subsets, and the remaining (K - 1) subsets except for one specific subset among the K subsets are used.

[0366] Refer to Figure 10, when starting the inner-loop feature extraction process, the arithmetic processing unit (CPU) 2040 sets the hyperparameters λ1 and λ2 to the initial values (λ1, λ2) = (0.1, 0.1) (step S200).

[0367] Next, the value of the variable i is set to 1 (step S202), and while the value of i does not exceed the number of repetitions (K - 1) of the inner-loop feature extraction (being "no" in step S204), the CPU 2040 performs sparse canonical correlation analysis based on the person attribute information 3104 related to the participant and the correlation matrix data 3106 of the functional connectivity stored in the non-volatile storage device 2080, excluding the i-th data block, among the inner-loop training data (step S206).

[0368] When, for the current combination of values (λ1, λ2), there are canonical variables that are only connected to the disease label (being "yes" in step S208), the CPU 2040 extracts the elements (FC) of the correlation matrix of the functional connectivity corresponding to the canonical variables that are only connected to the disease label, and stores these elements in the non-volatile storage device 2080 (step S210). After the processing of step S210, or when, for the current combination of values (λ1, λ2), there are no canonical variables that are only connected to the disease label (being "no" in step S208), the CPU 2040 increments the value of i by 1, and the process returns to step S204.

[0369] Thus, the CPU 2040 repeats the processing from step S206 to step S212 (K - 1) times.

[0370] In step S204, when the value of i exceeds the number of repetitions (K - 1) of the inner-loop feature extraction (being "yes" in step S204), the CPU 2040 changes either λ1 or λ2 by a predetermined step amount according to a predetermined rule for (λ1, λ2) (step S214). In step S216, when the value of (λ1, λ2) falls within the variable range (being "yes" in step S216), the CPU 2040 returns the process to step S202.

[0371] When the processing is completed for all possible combinations of values (λ1, λ2) (being "no" in step S216), the CPU 2040 obtains the union set of the elements (FC) of the correlation matrix of the functional connectivity extracted in the processing performed so far, and stores this union set as the first functional connectivity union set data 3108 in the non-volatile storage device 2080, thereby ending this processing.

[0372] (Processing of outer-loop feature extraction)

[0373] Figure 12It is a flowchart for explaining the outer ring feature selection process in more detail.

[0374] In the outer ring feature extraction, as Figure 11 shown, the set of participants is divided into K subsets, and one subset that has not been used in the inner ring feature extraction among the K subsets is used as the test pool. The number of participants included in the test pool is set to NT.

[0375] In addition, the following assumes that the leave-one-out cross-validation (LOOCV) process is performed as an example for explanation.

[0376] Referring to Figure 12 , when starting the outer ring feature extraction process, the arithmetic processing device (CPU) 2040 sets the value of variable i to 1 (step S300), and when the value of i does not exceed the number of repetitions NT of the outer ring feature extraction ( "No" in step S302), the CPU 2040 performs sparse logistic regression (SLR) to generate a test classifier based on the person attribute information 3104 related to the participants and the correlation matrix data 3106 related to the functional connectivity stored in the non-volatile storage device 2080 except for the i-th data in the test pool (step S304).

[0377] Next, the CPU 2040 uses the generated test classifier to calculate the predicted value for the excluded i-th data through SLR (step S306).

[0378] In addition, the CPU 2040 stores the functional connectivity FC extracted by sparsification when generating the test classifier as the extracted feature in the non-volatile storage device 2080, and increments the value of variable i by 1 (step S310).

[0379] After repeating the processing of steps S304 to S308 NT times, when the value of variable i exceeds NT ( "Yes" in step S302), the CPU 2040 calculates the estimated square error as LOOCV (step S312).

[0380] Next, the CPU 2040 obtains the union set (second union set) of the extracted functional connectivity FC through NT times of repeated processing, and stores this union set as the second functional connectivity union set data 3110 in the non-volatile storage device 2080, thus ending this processing.

[0381] (Process of generating classifier)

[0382] Figure 13 It is a conceptual diagram for explaining the process of generating the final first classifier.

[0383] As Figure 13As shown, in the classifier generation process, inner-loop feature extraction processing and outer-loop feature extraction processing are performed to use the finally extracted second functional connection and set data 3110 as explanatory variables, and a first classifier is generated for all participants through sparse logistic regression. Information for identifying the generated first classifier (e.g., the parameter vector θ of the logistic function) is stored in the non-volatile storage device 2080.

[0384] As described above, in the above process, when "depression" is set as the disease label, 12 pairs of functional connections shown in Table 1 are used as explanatory variables for the finally extracted second functional connection and set data 3110, and a first classifier is generated.

[0385] All the functional connections shown in Table 1 can be used for the generation of a classifier represented by a logistic regression expression. However, at least one or both of the functional connection identification number 1 and the functional connection identification number 2 can be selected for use. In addition, multiple functional connections with high contribution degrees can be selected. Such selection of the functional connections to be used for the classifier from 12 pairs of functional connections can be performed by the data processing unit 32 or the CPU of the computer 300 described later. However, the selection can be made manually.

[0386] (Discrimination processing)

[0387] Figure 14 It is a flowchart for explaining the processing that the data processing unit 32 has to perform to discriminate the disease label based on the rs-fc MRI data of the subject by using the generated first classifier.

[0388] In step S401, the CPU 2040 acquires the rs-fc MRI measurement data 3113 of the subject in a resting state from the MRI imaging unit 25 via the interface unit 44.

[0389] In steps S402 and S403, the CPU 2040 performs the preprocessing described in the above section "2.", and extracts the elements of the correlation matrix for all or part of the functional connections shown in Table 1.

[0390] In step S404, the CPU 2040 extracts the Pearson correlation coefficient for the extracted elements of the correlation matrix. In steps S405 and S406, the CPU 2040 inputs the correlation coefficient into the classifier, and generates an index value for discriminating the disease label of depressive symptoms for at least one functional connection. In other cases, the CPU 2040 expects to input the correlation coefficient into the classifier, and generates a correlation weighted sum of multiple functional connections as an index value for discriminating the disease label of depressive symptoms.

[0391] Next, in step S407, the CPU 2040 compares the metric value with a reference value. For example, in the logistic regression expression of Equation 5 represented by Equation 6, the thresholds for the disease label for depressive symptoms and the label for healthy individuals are set to 0, so the reference value is 0. For example, in the case where the "correlation weighted sum" is used as the metric value of the biomarker, the reference value can be set to a value different from 0 to adjust the sensitivity or specificity of the first classifier. For example, when the reference value is set to be less than 0, the discrimination sensitivity for the disease level of "depression" increases, while the specificity decreases. In the case where the discrimination result using the classifier is intentionally used as reference information (auxiliary information) for a doctor to make a diagnosis based on other subsequent information, a setting that gives priority to the increase in sensitivity can be adopted.

[0392] In step S408, when the metric value is greater than 0 ("Yes: Y"), the CPU 2040 can determine that the subject has a label for depressive symptoms (step S409). On the contrary, when the metric value is less than 0 ("No: N"), the CPU 2040 can determine that the subject has a label for depressive symptoms (step S410).

[0393] 3-2. Discrimination device

[0394] In the description of the first embodiment, the brain activity measurement device (fMRI device) is configured to measure brain activity data measured at one measurement site, and use the brain activity data to generate a classifier through processing by the same computer or distributed processing, and use the classifier to estimate (predict) a disease label.

[0395] However, the following structure can be adopted: i) measurement (data collection) of brain activity data for training a classifier through machine learning; ii) processing of generating a classifier through machine learning, and processing of estimating (predicting) a disease label for a specific subject through the classifier (estimation processing); and iii) measurement of brain activity data of a specific participant at different sites in a distributed manner (measurement of the subject's brain activity).

[0396] Specifically, the data related to a group of healthy individuals and a group of depression patients is not limited to the case of being measured by the MRI device 10 itself, and data measured by other MRI devices can be integrated to generate a classifier. Furthermore, more generally, the data processing unit 32 does not always need to be a computer for controlling the MRI device, but can be a dedicated computer configured to receive measurement data from one or more MRI devices, and perform discrimination processing by generating a classifier and using the generated classifier.

[0397] Figure 15It is a functional block diagram for showing an exemplary case of distributed processing of data collection, estimation processing, and measurement of a subject's brain activity in another embodiment of the first embodiment of the present invention.

[0398] Reference Figure 15 , locations 100.1 to 100.N are facilities where brain activity measurement devices measure data related to participants including a group of depression patients and a group of healthy individuals, and the management server 200 is configured to manage the measurement data obtained at locations 100.1 to 100.N.

[0399] The computer 300 serving as the discrimination device 2 is configured to generate a classifier based on the data stored in the management server 200.

[0400] The MRI device 410 is set at another location using the result of the classifier on the computer 300 and is configured to measure brain activity data for a specific subject.

[0401] The computer 400 is installed at another location where the MRI device 410 is set and is configured to calculate relevant data of the functional connectivity of the brain of a specific subject based on the measurement data of the MRI device 410, send the relevant data of the functional connectivity to the computer 300, and use the discrimination result using the classifier returned.

[0402] The server 200 stores the fMRI measurement data 3102 of participants related to a group of depression patients and a group of healthy individuals sent from locations 100.1 to 100.N, and the personal attribute information 3104 of the participants associated with the fMRI measurement data 3102, and is configured to send these data to the computer 300 in response to an access from the computer 300.

[0403] The computer 300 is configured to receive the fMRI measurement data 3102 of the participants and the personal attribute information of the subject from the server 200 via the communication interface 2090.

[0404] The hardware structures of the server 200, the computer 300, and the computer 400 are substantially similar to Figure 2 the "data processing unit 32" shown, so the description of these hardware structures will not be repeated here.

[0405] Return reference Figure 15, the correlation matrix calculation unit 3002, the first feature selection unit 3004, the second feature selection unit 3006, the classifier generation unit 3008, the discrimination processing unit 3010, the data 3106 related to the correlation matrix of functional connections, the first functional connection and set data 3108, the second functional connection and set data 3110, and the classifier data 3112 are similar to those described in the first embodiment, so the description of these will not be repeated here.

[0406] The MRI device 410 is configured to measure brain activity data related to the subject for whom the diagnostic label is to be estimated, and the processing device 4040 of the computer 400 stores the measured MRI measurement data 4102 in the non-volatile storage device 4100.

[0407] In addition, similar to the correlation matrix calculation unit 3002, the processing device 4040 of the computer 400 is configured to calculate the data 4106 related to the correlation matrix of functional connections based on the MRI measurement data 4102 and the human attribute information related to the subject to be measured by the MRI device 410, and stores the data 4106 in the non-volatile storage device 4100.

[0408] The user of the computer 400 designates the disease to be diagnosed, and the computer 400 sends the data 4106 related to the correlation matrix of functional connections to the computer 300 according to the send instruction given by the user. In response to this, the discrimination processing unit 3010 calculates the discrimination result for the designated diagnostic label, and the computer 300 sends the result to the computer 400 via the communication interface 2090.

[0409] The computer 400 notifies the user of the discrimination result via a display device (not shown), for example.

[0410] With this structure, it is possible to provide the estimation result of the diagnostic label using the classifier to a larger number of users based on the data collected for a larger number of subjects.

[0411] In addition, a mode of separately managing the administrator management server 200 and the computer 300 can also be adopted. In this case, the computers that can access the server 200 are restricted, thereby improving the security of the information related to the subjects stored in the server 200.

[0412] In addition, regarding the management entity of the computer 300, the "service of providing the discrimination result" can be carried out without providing any information related to the classifier to "the side (computer 400) that receives the service of discriminating using the classifier" at all.

[0413] 3-4. Uses of the Classifier and the Discrimination Device

[0414] In addition to the discrimination of disease labels for depressive symptoms, the discrimination devices 1 and 2 can also be used to discriminate the level of depressive symptoms, generate information for discriminating the degree of treatment effect of the discriminated depressive symptoms, or generate auxiliary information when the subject is classified into subclasses in the case where depressive symptoms have been classified into multiple preset subclasses. How to use each of the discrimination devices 1 and 2 will be described later.

[0415] The first classifier can be used as a diagnostic biomarker for discriminating depressive symptoms. In addition, the first classifier can be used as a diagnostic biomarker for discriminating the level of depressive symptoms. The first classifier can be used as an efficacy biomarker for discriminating the treatment effect of depression. The first classifier can be used as a biomarker for classifying a subject into subclasses of depression.

[0416] 3-5. Computer programs for generating classifiers and computer programs for discriminating depressive symptoms

[0417] Another mode of the first embodiment includes the following programs: a program for causing a computer to execute a process including steps S100 to S108 (specifically including steps S200 to S218 and steps S300 to S314), thereby executing the function of classifier generation processing; and a program for causing a computer to execute a process including steps S401 to S410, thereby executing the function of discrimination processing. In addition, another mode of the first embodiment includes the following program, which is for causing a computer to execute the processes including steps S100 to S108 (specifically including steps S200 to S218 and steps S300 to S314) and steps S401 to S410, thereby executing the function of the discrimination device. These programs can be stored in a storage medium (for example, a hard disk drive, a semiconductor storage element such as a flash memory, or an optical disc). The format of storing the program in the storage medium is not limited as long as the CPU 2040 can read the program. The storage medium for storing the program is preferably a non-volatile storage medium.

[0418] 3-6. Method for discriminating depressive symptoms

[0419] The second embodiment of the present invention relates to a method for discriminating depressive symptoms. The discrimination method includes the following steps: generating an index value for evaluating depressive symptoms for an element of the correlation matrix of the functional connectivity measured for a subject in a resting state; and determining that the subject has depressive symptoms when the index value exceeds a reference value. The specific process conforms to steps S401 to S410, but all or part of this process can be performed manually except for the step of generating the index value.

[0420] 4. Device and method for discriminating the level of depressive symptoms

[0421] 4-1. Discrimination Device

[0422] The discrimination device 1 and the discrimination device 2 can be used to discriminate the level of depressive symptoms. That is, the discrimination device for the level of depressive symptoms according to another mode of the first embodiment includes a processor and a storage device. The processor performs the following processing: Based on the program stored in the storage device, by using the first classifier, for the elements of the correlation matrix of the functional connectivity measured for the subject in the resting state, an index value for evaluating depressive symptoms is generated. Then, the processor performs the following processing: comparing the index value with the reference range of the index values set in advance for each functional connectivity according to the level of depressive symptoms, and determining that the subject has the level of depressive symptoms corresponding to the reference range including the index value. In this case, the structures of the discrimination device 1 and the discrimination device 2 are similar to the structures described in the above part "3.", so the description of this structure is omitted here. The discrimination processing unit 3010 performs the processing of discriminating the input data based on the classifier identified by the classifier data 3112, and discriminates the level of the subject's depressive symptoms.

[0423] In this case, for the discrimination of the "level of depressive symptoms", for example, by classifying in advance the values that the value of the "correlation weighted sum" as the index value can take into multiple levels of reference ranges, the discrimination result corresponding to the "level of depressive symptoms" can be output.

[0424] Alternatively, in this case, the "disease labels" to be used for generating the classifier can be set in advance as multiple "level disease labels" corresponding to the levels of depression, and thus the machine learning of the classifier is performed.

[0425] 4-2. Discrimination Method

[0426] The third embodiment of the present invention relates to a discrimination method for the level of depressive symptoms. The discrimination method includes the following steps: generating an index value for evaluating depressive symptoms for the elements of the correlation matrix of the functional connectivity measured for the subject in the resting state; comparing the index value with the reference range of the index values set in advance for each functional connectivity according to the level of depressive symptoms; and determining that the subject has the level of depressive symptoms corresponding to the reference range including the index value.

[0427] Figure 16 It is a flowchart showing the processing that the data processing unit 32 has to perform to discriminate the level of depressive symptoms based on the rs-fc MRI subject data by using the generated classifier.

[0428] The CPU 2040 receives, for example, an input of the start of processing from the input unit 40 to execute Figure 14Steps S401 to S406 shown are performed to generate an index value. Then, the CPU 2040 Figure 16 compares the index value with a reference range of the index value set in advance according to the level of depressive symptoms in step S501 shown. 12 pairs of functional connections are related to, for example, the severity of depressive symptoms based on the BDI, and thus the reference range of the index value for determining the level of depressive symptoms can be determined in advance according to the correlation with the severity of depressive symptoms based on the BDI. In step S502, the CPU 2040 determines which level of depressive symptoms the index value generated in Figure 14 step S406 corresponds to (step S502). Then, in step S503, the CPU 2040 determines that the subject has the level of depressive symptoms determined in step S502. Regarding each step to be executed by the CPU 2040, a part or all of the steps other than the step of generating the index value can be performed manually.

[0429] In addition, another mode of the third embodiment includes a program for causing a computer to execute a process including the above steps S501 to S503, thereby performing the function of the discrimination device. Another mode of the third embodiment includes a program for causing a computer to execute a process including the above steps S401 to S406 and steps S501 to S503, thereby performing the function of the discrimination device. These programs can be stored in a storage medium (for example, a hard disk drive, a semiconductor storage element such as a flash memory, or an optical disc). The format of storing the program in the storage medium is not limited as long as the CPU 2040 can read the program. The storage medium for storing the program is preferably a non-volatile storage medium.

[0430] 5. Discrimination Device and Discrimination Method for Treatment Effect Using the First Classifier

[0431] 5-1. Discrimination Device

[0432] The discrimination device 1 and the discrimination device 2 can be used to generate information for discriminating the treatment effect. That is, the discrimination device for judging the treatment effect of the subject according to another mode of the first embodiment includes a processor and a storage device. The processor is configured to perform the following processing: based on the program stored in the storage device, by using the first classifier, generate a first value for evaluating depressive symptoms for the elements of the correlation matrix of the functional connectivity measured for the subject in the resting state at the first time point; by using the classifier, generate a second value for evaluating depressive symptoms for the elements of the correlation matrix of the same functional connectivity as the above-mentioned functional connectivity in the brain of the same subject as the above-mentioned subject in the resting state at the second time point after the start of the treatment and later than the first time point; compare the first value with the second value; and i) in the case where the second value is improved compared with the first value, determine that the treatment is effective for improving the depressive symptoms of the subject, and / or in the case where the second value is not improved compared with the first value, determine that the treatment is ineffective for improving the depressive symptoms of the subject. In this case, the structures of the discrimination device 1 and the discrimination device 2 are similar to the structures described in the above part "3.", and thus the description of these structures is omitted here. The discrimination processing unit 3010 performs the processing of discriminating the input data based on the classifier identified by the classifier data 3112, and generates information for discriminating the treatment effect of the subject.

[0433] Another mode of this embodiment includes using the discrimination device to perform drug reanalysis.

[0434] 5-2. Discrimination method

[0435] In the fourth embodiment of the present invention, the first classifier receives an input of elements of a correlation matrix of functional connections shown in Table 1, which are obtained from the resting-state fMRI data of a subject imaged at at least two time points, and generates information for discriminating treatment effects. Specifically, the fourth embodiment includes the following steps: generating a first value for evaluating depressive symptoms for elements of a correlation matrix of functional connections measured for a subject in a resting state at a first time point; generating a second value for evaluating depressive symptoms for elements of a correlation matrix of the same functional connections as those in the brain of the same subject as the above-mentioned subject in a resting state at a second time point; comparing the first value with the second value; and when the second value is improved compared with the first value, determining that the treatment is effective in improving the depressive symptoms of the subject, thereby generating information indicating the possibility of effectiveness. The first time point may be a time point before treatment, or may be a time point after a predetermined period has elapsed since the start of treatment. However, the first time point is preferably a time point before the start of treatment. In addition, also in the case where treatment has ended once and after treatment is restarted, the period from the end of treatment until the restart of treatment may be defined as "before treatment". The second time point is not limited as long as the second time point is a time point after the start of treatment and later than the first time point.

[0436] FIG. 17 is a flowchart for explaining a process for discriminating treatment effects based on fMRI data of a subject by using a first classifier.

[0437] In step S601 shown in FIG. 17, the CPU 2040 receives, for example, an input of the start of processing from the input unit 40 to obtain the fMRI measurement data 3113 of a subject in a resting state, which is obtained by imaging the subject with the MRI imaging unit 25 at a first time point via the interface unit 44.

[0438] In steps S602 and S603, the CPU 2040 performs the preprocessing described in the above-mentioned part "2.", and extracts elements of the correlation matrix at the first time point for all or a part of the functional connections selected from the 12 pairs of functional connections shown in Table 1.

[0439] In step S604, the CPU 2040 calculates the Pearson correlation coefficient for the elements of the correlation matrix at the first time point extracted in step S603.

[0440] In steps S605 and S606, the CPU 2040 inputs the correlation coefficient calculated in step S604 into the first classifier, and generates a first value for the selected functional connections. This first value is calculated, for example, as the index value described in the above-mentioned part "2.".

[0441] At the second time point, the CPU 2040 acquires, via the interface unit 44, the fMRI measurement data 3113 of the subject in a resting state obtained by imaging the subject using the MRI imaging unit 25 shown in step S607.

[0442] In steps S608 and S609, the CPU 2040 performs the preprocessing described in the above section "2.", and extracts elements of the correlation matrix at the second time point for all or a part of the functional connections selected from the 12 pairs of functional connections shown in Table 1.

[0443] In step S610, the CPU 2040 calculates the Pearson correlation coefficient for the elements of the correlation matrix at the second time point extracted in step S609.

[0444] In steps S611 and S612, the CPU 2040 inputs the correlation coefficient calculated in step S610 into the first classifier, and generates a first value for the selected functional connection. The second value is generated, for example, as the index value described in the above section "2.".

[0445] The CPU 2040 compares the first value generated in step S606 with the second value generated in step S612 (step S613).

[0446] In step S614, when the CPU 2040 determines that the second value has improved compared to the first value ("Yes: Y"), the CPU 2040 generates information indicating the state of treatment related to the improvement of the depressive symptoms of the subject (step S615). In this case, the CPU 2040 may then proceed to step S616 to prompt information indicating the possibility of continuing treatment. The doctor determines whether to continue treatment based on such information indicating the possibility of continuing treatment. For example, although not particularly limited, information on a preset region indicating the effectiveness of treatment may be displayed in a distinguishable manner on the output screen. "Information indicating the state of treatment" may be the index value to be output by the classifier, and "prompting information indicating the possibility of continuing treatment" may be to display the index value in the state of displaying the above region.

[0447] In addition, in step S614, when the CPU 2040 determines that the second value has not improved compared to the first value ("No: N"), the CPU 2040 generates information indicating the state of treatment related to the improvement of the depressive symptoms of the subject (step S617). In this case, the CPU 2040 may then proceed to step S618 to prompt information indicating the possibility of ending treatment.

[0448] The doctor determines whether to end the treatment based on the information indicating the possibility of ending the treatment. For example, although not particularly limited, the information of the preset area indicating the ineffectiveness of the treatment can be displayed on the output screen in a distinguishable manner. "The information indicating the state of the treatment" can be the index value to be output by the classifier, and "the information prompting the possibility of ending the treatment" can be to display the index value in the state of displaying the above area.

[0449] In addition, the CPU 2040 can enter step S619 to prompt the information indicating the possibility of changing the current treatment method to other treatment methods. In this case, the CPU 2040 can prompt the possibility of more specific treatment methods. In this case, the doctor can determine the treatment method to which the current treatment method is to be changed among the displayed potential treatment methods.

[0450] For example, in the expression represented by Mathematical Formula 6, the thresholds for the disease label of the depressive symptom and the label of the healthy individual are set to 0, so the reference value is 0.

[0451] For example, in the case of using "correlation weighted sum" as the index value of the biomarker, in step S614, as the depressive symptom becomes more severe, the index value becomes larger in the positive direction. Therefore, when the second value is less than the first value, the CPU 2040 can determine that the second value has improved compared with the first value. In addition, when the second value is greater than the first value, or when no significant difference is identified between the first value and the second value, the CPU 2040 can determine that the second value has not improved compared with the first value.

[0452] The first value and the second value can be index values, but the correlation coefficients calculated in step S604 and step S610 can be set as the first value and the second value respectively.

[0453] In addition, in the third embodiment, step S601 needs to be performed before step S607. However, steps S602 to S606 do not need to be performed before step S607. Steps S602 to S606 only need to be performed somewhere after step S601 and at least before step S613.

[0454] All or part of steps S601 to S619 can be performed manually.

[0455] The third embodiment can also be used to judge the effect of drug treatment. In the case of judging the effect of drug treatment by using the first classifier, it is preferably to set the functional connections with at least functional connection identification numbers 1 and 2 as the indexes.

[0456] Another mode of the fourth embodiment includes using a judgment method to perform drug reanalysis.

[0457] In addition, another mode of the fourth embodiment includes a program for causing a computer to execute the processes including the above-described steps S601 to S619, thereby executing the functions of the discrimination device. This program can be stored in a storage medium (for example, a hard disk drive, a semiconductor storage element such as a flash memory, or an optical disc). The format of storing the program in the storage medium is not limited as long as the CPU 2040 can read the program. The storage medium for storing the program is preferably a non-volatile storage medium.

[0458] 6. Classification Device and Classification Method for Depressed Patients

[0459] 6-1. Classification Device

[0460] Another mode of the first embodiment relates to a classification device for depressed patients. When depression is classified into a plurality of preset subclasses, the discrimination devices 1 and 2 can be respectively used as classification devices 1 and 2 (hereinafter collectively referred to as "classification devices") configured to assist doctors in classifying depressed patients. The classification device includes a processor and a storage device. That is, the processor of the discrimination device for classifying depressed patients performs the following processes: in the case of classifying a depressed patient into a plurality of preset subclasses, based on the program stored in the storage device, generating an index value for evaluating depressive symptoms for an element of the correlation matrix of the functional connectivity measured for the subject in a resting state; comparing the index value with a reference range of the index values preset for each subclass of the functional connectivity; and determining that the subject has a subclass corresponding to the reference range including the index value. In this case, the structures of the classification devices 1 and 2 are similar to the structures of the discrimination devices 1 and 2 described in the above section "3.", so the description of these structures is omitted here. Figure 19 and Figure 20 are functional block diagrams showing the classification devices 1 and 2. Except that the discrimination processing unit 3010 is replaced by a classification processing unit 3020, Figure 19 and Figure 20 are respectively the same as Figure 5 and Figure 15 The classification processing unit 3020 performs a process of classifying the input data based on the classifier identified by the classifier data 3112, performs discrimination for classifying depressed patients, and outputs the result.

[0461] Doctors use the discrimination result for classifying depressed patients output from the classification device 1 or 2 as auxiliary information to classify patients. In this sense, such a device can also be referred to as a "classification assistance device".

[0462] 6-2. Classification method

[0463] The fifth embodiment of the present invention relates to a method for classifying patients with depression. The discrimination method includes the following steps: in the case of classifying depression into a plurality of preset subclasses, generating an index value for evaluating depressive symptoms for an element of a correlation matrix of functional connections measured for a subject in a resting state; comparing the index value with a reference range of index values preset for each subclass of functional connections; and determining that the subject has a subclass corresponding to the reference range including the index value.

[0464] Figure 18 It is a flowchart for explaining the processing to be executed by the data processing unit 32, which is configured to generate information for discriminating subclasses of depressive symptoms based on the fMRI data of the subject by using the generated classifier.

[0465] The CPU 2040 receives, for example, an input of the start of processing from the input unit 40 to execute Figure 14 the steps S401 to S406 shown, thereby generating an index value. Then, in Figure 18 step S701 shown, the CPU 2040 compares the index value with a reference range of index values preset according to the subclasses of depression. The subclasses of depression are classified into subclasses of melancholic MDD, non-melancholic MDD, and treatment-resistant MDD according to clinical manifestations. Thus, for example, a classifier generated by setting melancholic MDD as a "disease label" can be used to determine whether the measurement data obtained from a subject whose brain activity has been measured corresponds to the disease label of melancholic MDD. Similarly, a classifier generated by setting non-melancholic MDD or treatment-resistant MDD as a "disease label" can be used to determine whether the measurement data obtained from a subject whose brain activity has been measured corresponds to the disease label of non-melancholic MDD or treatment-resistant MDD.

[0466] In step S702, the CPU 2040 determines which subclass the index value generated in Figure 15 step S406 corresponds to. Then, in step S703, the CPU 2040 determines that the subject belongs to the subclass of depression determined in step S702. A doctor uses the information related to the subclass output in this way as auxiliary information for classifying patients with depression, thereby classifying the patients. In this sense, this method can also be called a "classification assistance method". All or part of the steps to be executed by the CPU 2040, except for the step of generating the index value, can be performed manually.

[0467] In addition, another mode of the fifth embodiment includes a program for causing a computer to execute the processes including the above-described steps S701 to S703, thereby implementing the functions of the classification device. Another mode of the fifth embodiment includes a program for causing a computer to execute the processes including the above-described steps S401 to S406 and steps S701 to S703, thereby implementing the functions of the classification device. These programs can be stored in a storage medium (e.g., a hard disk drive, a semiconductor storage element such as a flash memory, or an optical disc). The format of storing the program in the storage medium is not limited as long as the CPU 2040 can read the program. The storage medium for storing the program is preferably a non-volatile storage medium.

[0468] 7. Generation of the second classifier and determination of the treatment effect using the second classifier

[0469] 7-1. Apparatus for determining treatment effect

[0470] A sixth embodiment of the present invention relates to an apparatus for determining treatment effect or a treatment assistance apparatus, which is configured to: perform a classifier generation process for generating a second classifier for determining a group of subjects among a plurality of subjects showing a treatment effect given by treatment and a group of subjects not showing a treatment effect; and generate information serving as an index for assisting a doctor in determining a treatment effect.

[0471] The structure of the apparatus for determining treatment effect is substantially the same as that of Figure 1 but Figure 1 the data processing unit 32 in Figure 1 is replaced by a data processing unit 62 in the sixth embodiment. In addition, in the apparatus for determining treatment effect, the structure of the data processing unit 62 is substantially the same as that of Figure 2 and Figure 1 the storage unit 36, control unit 42, input unit 40, interface unit 44, data collection unit 46, image processing unit 48, display unit 38, and display control unit 34 in Figure 2In [the original text], the computer main body 2010, memory drive 2020, disk drive 2030, processor (CPU) 2040, disk drive 2030, memory drive 2020, bus 2050, ROM 2060, RAM 2070, non-volatile storage device 2080, and communication interface 2090 of the data processing unit 32 are respectively replaced with a computer main body 6010, memory drive 6020, disk drive 6030, processor (CPU) 6040, disk drive 6030, memory drive 6020, bus 6050, ROM 6060, RAM 6070, non-volatile storage device 6080, and communication interface 6090. The hardware of the data processing unit 62 is not particularly limited as described above, and a general computer can be used as the hardware.

[0472] (Generation of classifier)

[0473] Figure 23 is a functional block diagram for performing a process of generating a second classifier and a discrimination process using the generated second classifier.

[0474] The non-volatile storage device 6080 stores information related to signals obtained by measuring signals representing brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and treated depression patients in advance and in time series using an MRI device. The stored information is rs-fc MRI measurement data 6102 of the subject measured at the first time point, rs-fc MRI measurement data 6104 of the subject measured at the second time point, and multiple treatment data 6105 associated with each subject for whom rs-fc MRI measurement data has been measured.

[0475] The treatment data 6105 contains information related to the disease label and treatment history of the subject (e.g., treatment method, name of administered drug, dose, or duration of administration). The CPU 6040 performs the following process: Based on the rs-fc MRI measurement data 6102 of the subject at the first time point, the rs-fc MRI measurement data 6104 of the subject at the second time point, and the treatment data 6105 of the subject, a classifier is generated for identifying a group of subjects ("remitted group") among multiple subjects showing treatment effects and a group of subjects ("non-remitted group") among multiple subjects not showing treatment effects. Thus, in this case, the disease label includes a "remission label" and a "non-remission label".

[0476] For each subject, the first correlation measurement unit 6002 calculates a correlation matrix of the functional connectivity of the brain activities in multiple predetermined regions based on the rs-fcMRI measurement data 6102 of the subject at the first time point. Data related to the calculated correlation matrix of the functional connectivity is stored as first correlation measurement data 6106 related to the correlation matrix of the functional connectivity for each subject in the non-volatile storage device 6080.

[0477] For each subject, the second correlation measurement unit 6004 calculates a correlation matrix of the functional connectivity of the brain activities in multiple predetermined regions based on the rs-fcMRI measurement data 6104 of the subject at the second time point. Data related to the calculated correlation matrix of the functional connectivity is stored as second correlation measurement data 6108 related to the correlation matrix of the functional connectivity for each subject in the non-volatile storage device 6080.

[0478] The second classifier generation unit 6008 obtains a regression expression based on the difference between the correlation at the first time point and the correlation at the second time point of multiple functional connectivities. The regression expression is stored as second classifier data 6020 for identifying the second classifier in the non-volatile storage device 6080. The treatment effect discrimination processing unit 6100 performs a process of discriminating the treatment effect on the input data based on the second classifier identified by the second classifier data 6020.

[0479] Figure 21 It is a flowchart for explaining the process that the data processing unit 62 needs to perform to generate the second classifier.

[0480] Now, refer to Figure 21 to explain in more detail Figure 23 the process shown.

[0481] In Figure 21 when the data processing unit 62, for example, receives an input to start processing from the input unit 70 and starts the process of generating a classifier (start), the data processing unit 62 reads the rs-fcMRI measurement data 6102 of each subject at the first time point, and measures the correlation at the first time point for each functional connectivity among the multiple functional connectivities shown in Table 1 (step S802). The measurement of this correlation is performed by calculating the Pearson correlation coefficient of each functional connectivity.

[0482] Next, the data processing unit 62 reads the rs-fcMRI measurement data 6104 of each subject at the second time point from the storage unit 66, and measures the correlation at the second time point for each functional connectivity among the multiple functional connectivities shown in Table 1 for each subject (step S803). The measurement of this correlation is performed by calculating the Pearson correlation coefficient of each functional connectivity.

[0483] Next, the data processing unit 62 calculates the difference between the correlation at the first time point and the correlation at the second time point for each of the multiple functional connections shown in Table 1 for each subject to be examined, such as the first functional connection and the second functional connection (step S804). This difference is calculated, for example, based on the following expression (4).

[0484] [Mathematical formula 10]

[0485]

[0486] In expression (4), FC1 represents the first connection strength in Table 1, and FC2 represents the second connection strength in Table 1. That is, as the result of calculating the correlation weighted sum, the connection with the largest absolute value of the weight is FC1, and the connection with the second largest absolute value of the weight is FC2. The superscripts "post" and "pre" represent the second time point and the first time point, respectively. Usually, the first time point is the time point before treatment (before administration), and the second time point is a predetermined time point after the start of treatment (after the start of administration). In addition, "sign(w1)" represents the sign of the weight of FC1, and "sign(w2)" represents the sign of the weight of FC2. For example, as described later, in Figure 33 the classifier created in f represents: accuracy: 0.75, AUC: 0.79, specificity: 0.88, and sensitivity: 0.43.

[0487] In addition, the data processing unit 62 obtains a regression expression that describes the relationship between the difference between the two correlations of each functional connection of each subject to be examined and the explanatory variables ("remission group" and "non-remission group" labels) included in the treatment data 6105. The regression expression is not particularly limited and can be obtained by linear regression analysis. This regression expression can be used to generate a classifier configured to discriminate a group of subjects showing a treatment effect and a group of subjects not showing a treatment effect in the correlation state space (step S804).

[0488] The term "correlation state space" here refers to a space in which the difference in strength between the first time point and the second time point of multiple functional connections (or, if necessary, the difference in strength multiplied by the sign of the weight) is expanded as the axis of this space. In the above example, the correlation state space is a two-dimensional space.

[0489] As described above, for example, by plotting the correlation state of the difference (Δsign(W)FC1) between the correlations at the first time point and the second time point of functional connection identification number 1 shown in Table 1 for each subject to be examined and the difference (Δsign(W)FC2) between the correlations at the first time point and the second time point of functional connection identification number 2 shown in Table 1 for each subject to be examined,Figure 33 f. Considering the relationship between the correlation between Δsign(W)FC1 and Δsign(W)FC2 and whether a treatment effect is exhibited in each subject, a linear regression expression with a negative slope can be obtained. Then, based on the regression line of this linear regression expression, subjects distributed above the regression line and subjects distributed below the regression line can be respectively identified as a group of subjects who do not exhibit a treatment effect and a group of subjects who exhibit a treatment effect.

[0490] The doctor uses the information related to the treatment effect output in this way as auxiliary information to judge the treatment effect for patients with depression.

[0491] (Δsign(W)FC1 and Δsign(W)FC2 as explanatory variables)

[0492] In the above example, for the convenience of display as a graph, the sign sign(W) of the weight of each connection strength is used. The intensity changes of FC1 and FC2 through treatment (drug administration) are essentially important.

[0493] As Figure 33 shown by f, when the remission period and the non-remission period are not distinguished from each other, there is no correlation between the intensity change Δsign(W)FC1 of the first functional connection and the intensity change Δsign(W)FC2 of the second functional connection between the time before the start of treatment and the time after a predetermined period has elapsed since the start of treatment, and no phenomenon of compensation for each change is observed. In other words, these changes in connection strength are independent of each other, which indicates that the selection of variables for describing the treatment effect is appropriate.

[0494] Therefore, as described above, instead of using the first functional connection and the second functional connection, other functional connections can also be used by using functional connections whose intensity changes before and after the start of treatment can be regarded as independent of each other. In addition, a larger number of functional connections can be selected from the functional connections shown in Table 1 as variables for describing the treatment effect.

[0495] (Discrimination process)

[0496] Figure 22 is a flowchart for explaining the process to be executed by the data processing unit 62, which is configured to discriminate the treatment effect based on the rs-fc MRI data of the subject by using the second classifier generated in the seventh embodiment of the present invention.

[0497] The CPU 6040 receives, for example, an input indicating the start of processing from the input unit 70, and acquires rs-fc MRI measurement data 6113 of the subject at rest at a first time point from the MRI imaging unit 25 via the interface unit 64 in step S901.

[0498] In steps S902 and S903, the CPU 6040 performs the preprocessing described in the above section "2.", extracts elements of the correlation matrix for all or part of the functional connections shown in Table 1, and measures the correlation of each functional connection at the first time point.

[0499] In step S904, the CPU 6040 acquires rs-fc MRI measurement data 6113 of the subject at rest at a second time point from the MRI imaging unit 25 via the interface unit 64.

[0500] In steps S905 and S906, the CPU 6040 performs the preprocessing described in the above section "2.", extracts elements of the correlation matrix for all or part of the functional connections shown in Table 1, and measures the correlation of each functional connection at the second time point.

[0501] Next, in step S907, for each functional connection among the multiple functional connections, the CPU 6040 calculates the difference between the correlation of each functional connection measured at the first time point in step S903 and the correlation of each functional connection measured at the second time point in step S906.

[0502] The CPU 6040 inputs the difference calculated in step S907 to the second classifier (step S908), and determines whether a treatment effect is exhibited in the subject by using the second classifier (step S908).

[0503] The method for measuring each correlation, the method for calculating the difference between the correlations, and the method for discriminating the treatment effect are the same as these three methods described in the method for generating the second classifier.

[0504] The first time point may be before the start of treatment, or may be a time point after a predetermined period has elapsed since the start of treatment. However, the first time point is preferably before the start of treatment. In addition, similarly, when the treatment ends once and after the treatment restarts, the period from the end of treatment until the restart of treatment can be defined as "before treatment". The second time point is not limited as long as the second time point is after the start of treatment and later than the first time point.

[0505] In addition, in the seventh embodiment, step S901 needs to be performed before step S606. However, steps S902 and S903 do not need to be performed before step S904. Steps S902 and S903 only need to be performed somewhere after step S901 and at least before step S907.

[0506] In addition, another mode of the sixth embodiment includes performing drug reanalysis using a second classifier, a treatment effect discrimination device, or a treatment assistance device.

[0507] 7-2. Method for discriminating treatment effect

[0508] The seventh embodiment relates to a method for discriminating treatment effect using a second classifier. The discrimination method includes the following steps: measuring a first correlation between a plurality of functional connections in the brain of a subject in a resting state at a first time point; measuring a second correlation between a plurality of functional connections in the brain of the same subject as the above-mentioned subject in a resting state at a second time point; and discriminating the treatment effect for the subject by using a classifier based on the difference between the first correlation and the second correlation of the plurality of functional connections of the subject. The specific process conforms to steps S901 to S909, but all or part of this process can be performed manually.

[0509] In addition, in this case, discrimination information related to the treatment effect output in this way is used as auxiliary information to judge the treatment effect for a depression patient.

[0510] Another mode of the seventh embodiment includes performing drug reanalysis using the method for discriminating treatment effect.

[0511] In addition, another mode of the seventh embodiment includes the following program for causing a computer to execute the processing including the above steps S801 to S804, thereby executing the function of the second classifier. Another mode of the seventh embodiment includes the following program for causing a computer to execute the processing including the above steps S801 to S804 and steps S901 to S909, thereby executing the function of the treatment effect discrimination device. Another mode of the seventh embodiment includes the following program for causing a computer to execute the processing including the above steps S901 to S909, thereby executing the function of the treatment effect discrimination device. These programs can be stored in a storage medium (for example, a hard disk drive, a semiconductor storage element such as a flash memory, or an optical disc). The format of storing the program in the storage medium is not limited as long as the CPU 6040 can read the program. The storage medium for storing the program is preferably a non-volatile storage medium.

[0512] 8. Judgment of the treatment effect of classified depression patients using a second classifier

[0513] 8-1. Discrimination device for treatment effect

[0514] The eighth embodiment of the present invention relates to a device configured to discriminate the treatment effect for the depression patients according to the fourth embodiment by the discrimination device for treatment effect according to the sixth embodiment using rs-fc MRI data related to the depression patients classified by the classification device.

[0515] That is, the eighth embodiment relates to a discrimination device for treatment effect configured to perform the first classifier generation process described in the above part "3.", the classification process described in the above part "6.", the second classifier generation process described in the above part "7.", the first correlation measurement process, the second correlation measurement process, and the process of discriminating the treatment effect for the subject using the second classifier.

[0516] Therefore, the descriptions of the above parts "3.", "6-1.", and "7-1." are incorporated into this embodiment.

[0517] Another mode of the eighth embodiment includes using the discrimination device for treatment effect to perform drug reanalysis.

[0518] 8-2. Discrimination method for treatment effect

[0519] The ninth embodiment of the present invention relates to a discrimination method for the treatment effect of classified depression patients using a second classifier.

[0520] Figure 24 It is a flowchart for discriminating the treatment effect of classified depression patients by using a second classifier.

[0521] In Figure 24 the steps S701 to S703 are the same as the steps in Figure 18 Therefore, the description of the above part "6-2." is incorporated into this embodiment.

[0522] The CPU 6040 executes steps S921 to S929 for the subject whose subclass is determined in step S703 shown in Figure 24 Each step from step S921 to S929 corresponds to the steps from step S901 to S909 described in the above part "7-2." and shown in Figure 22 Therefore, the description of the above part "7-2." is incorporated into this embodiment.

[0523] Although not particularly limited, in the above description, when the doctor determines based on the auxiliary judgment of the treatment effect discrimination device that the currently adopted treatment method (for example, the administration of a specific drug) is ineffective, the doctor can then determine to combine and use another treatment method (for example, the administration of another specific drug, neurofeedback, electroconvulsive therapy without convulsions, or repetitive transcranial magnetic stimulation therapy), or to change the current treatment method to another treatment method.

[0524] Another mode of the ninth embodiment includes using a discrimination method for treatment effects to perform drug reanalysis.

[0525] In addition, another mode of the eighth embodiment includes a program for causing a computer to execute the processes including the above-described steps S921 to S929, thereby executing the function of the second classifier. Another mode of the eighth embodiment includes a program for causing a computer to execute the processes including the above-described steps S701 to S703 and steps S921 to S929, thereby executing the function of the treatment effect discrimination device. Another mode of the seventh embodiment includes a program for causing a computer to execute the processes including the above-described steps S401 to S406, steps S701 to S703, and steps S901 to S909, thereby executing the function of the treatment effect discrimination device. These programs can be stored in a storage medium (for example, a hard disk drive, a semiconductor storage element such as a flash memory, or an optical disc). The format of storing the program in the storage medium is not limited as long as the CPU 6040 can read the program. The storage medium for storing the program is preferably a non-volatile storage medium.

[0526] 9. Neurofeedback

[0527] 9-1. Brain Activity Training Device

[0528] The tenth embodiment of the present invention relates to a brain activity training device for providing feedback on the brain activity of a trainee and performing training (neurofeedback training, also simply referred to as "training") to make the brain activity closer to the relevant (connected) state of a healthy individual.

[0529] More generally, regarding the relationship between the brain activities of a group of healthy individuals and a group of patients, such a brain activity training device can not only be used to make the functional connection state of the brain activity of the trainee closer to the functional connection state of the brain activity of healthy individuals, but also be used to make the current functional connection state of the brain activity of the trainee closer to the target functional connection state of the brain activity. As described later, the functional connection state of the brain activity of healthy individuals and the target connection state of the brain activity are referred to as "target patterns". The target patterns are stored in the storage device 10080 or the memory drive 10020.

[0530] In this case, the trainee is preferably: a subject (preferably a subject with MDD, or more preferably a subject with melancholic MDD) who has been assigned the label of "depressive symptoms" by the depressive symptom discrimination device 1, the depressive symptom discrimination device 2, or the depressive symptom discrimination method described in the above section "3."; a subject (preferably a subject with MDD, or more preferably a subject with melancholic MDD) who has been determined to have a certain level of depressive symptoms by the depressive symptom level discrimination device and discrimination method described in the above section "4."; or a subject (preferably a subject classified as MDD, or more preferably a subject classified as melancholic MDD) classified by the depression patient classification device and classification method described in the above section "6.". Therefore, the description of the depressive symptom discrimination device 1, the depressive symptom discrimination device 2, or the depressive symptom discrimination method described in the above section "3.", the description of the depressive symptom level discrimination device and discrimination method described in the above section "4.", and the description of the depression patient classification device and classification method described in the above section "6." are incorporated into this embodiment.

[0531] Figure 34 is a diagram for showing the concept of the structure of the brain activity training device.

[0532] For example, the structure of the MRI device 10 shown above Figure 1 can be used as the hardware structure of the brain activity training device. The hardware structure of the brain activity training device is basically the same as that of Figure 1 However, Figure 1 the data processing unit 32 in Figure 1 is replaced by the data processing unit 102 in the tenth embodiment. In addition, the structure of the data processing unit 102 in the brain activity training device is basically similar to that of Figure 2 and Figure 1In it, the storage unit 36, the control unit 42, the input unit 40, the interface unit 44, the data collection unit 46, the image processing unit 48, the display unit 38, and the display control unit 34 are respectively replaced with the storage unit 106, the control unit 112, the input unit 110, the interface unit 114, the data collection unit 116, the image processing unit 118, the display unit 108, and the display control unit 104. In addition, in Figure 2 it, the computer main body 2010, the memory drive 2020, the disk drive 2030, the processor (CPU) 2040, the disk drive 2030, the memory drive 2020, the bus 2050, the ROM 2060, the RAM 2070, the non-volatile storage device 2080, and the communication interface 2090 of the data processing unit 32 are respectively replaced with the computer main body 10010, the memory drive 10020, the disk drive 10030, the processor (CPU) 10040, the disk drive 6030, the memory drive 6020, the bus 10050, the ROM 10060, the RAM 10070, the non-volatile storage device 10080, and the communication interface 10090. The hardware of the data processing unit 62 is not particularly limited as described above, and a general computer can be used as the hardware.

[0533] The description of the method for driving the brain activity training device as described below by way of example is based on the assumption that a brain activity detection device for measuring a time series signal representing brain activity by brain functional imaging uses real-time fMRI.

[0534] Now, referring to Figure 34 and Figure 38 to describe the process of neurofeedback. First, the MRI device 10 detects the brain activity of the trainee as a time series signal representing the brain activity of a plurality of predetermined regions of the brain within a predetermined time period ( Figure 38 step S1001 of). During this time period, the trainee preferably tries to increase the reward value described later. Echo planar imaging (EPI) is performed as fMRI.

[0535] Next, the CPU 10040 of the data processing unit 102 performs the process of reconstructing the captured image in real time.

[0536] As in the first embodiment, functional connections between the regions of interest with the functional connection identification numbers (the "ID" in Table 1) 1 to 12 shown in Table 1 are selected by feature selection to discriminate the label of "depressive symptoms".

[0537] The subject further selects and extracts at least one specific functional connection from the above 12 pairs of functional connections as the "functional connection to be trained". Although not particularly limited, for example, it is assumed that the "functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex" (the connection with ID = 1 in Table 1), which has the greatest contribution to the discrimination of the depression label, is selected from the functional connections shown in Table 1 as the "functional connection to be trained". The term "contribution" here refers to "the absolute value of the weight in the relevant weighted sum".

[0538] In Figure 32 it, as described above, this connection is the connection with the greatest contribution among the connections that change in the direction opposite to that of healthy individuals due to the administration of the "antidepressant" as described above.

[0539] It is possible to select only the connection with ID = 1 as the "functional connection to be trained", or in addition to the connection with ID = 1, other connections can also be selected. At this time, in addition to the connection with ID = 1, the connection with ID = 2, which has the second greatest contribution, can also be selected as the functional connection to be selected. In other cases, in addition to the connection with ID = 1, other connections that change in the direction opposite to that of healthy individuals due to the administration of the "antidepressant" (for example, the connection with ID = 3) can also be selected. In addition to the connection with ID = 1, the other functional connections to be selected can be one or more.

[0540] In addition, the data processing unit 102 performs the following processing: for the region of interest corresponding to the "functional connection to be trained", calculate the temporal correlation of the functional connection within a predetermined time period ( Figure 38 step S1002).

[0541] That is, the data processing unit 102 first calculates the "activity level" of each region of interest based on the fMRI data (signals) at n consecutive measurement time points (n: natural number, n ≥ 1) measured in real time during the resting state by fMRI. This activity level can be set, for example, as the average of the measured values at n consecutive measurement time points. Hereinafter, the time period including n consecutive measurement time points is referred to as a "sample step".

[0542] Then, for the activity level of each sample step, the data processing unit 102 calculates the correlation of the activity level on the time axis (hereinafter referred to as "temporal correlation") for the measurement window within a predetermined time period through the following numerical expression. The measurement window is a time period including the activity levels of a plurality of consecutive sample steps on the time axis (for example, m (m: natural number, m ≥ 2)).

[0543] [Mathematical formula 11]

[0544]

[0545] In the above expression, x represents the "activity level" of a region of interest of the "functional connection to be trained", and y represents the "activity level" of another region of interest of the "functional connection to be trained". In addition, x(bar) (the symbol representing "average value" by assigning "-" to the top of the letter x is called x(bar). The same applies to other letters) represents the average value of the activity level x within the measurement window as described above. The same applies to "y(bar)". In the expression given above, Σ means, for example, to find the sum of the activity levels of m sample steps within the measurement window. In addition, although the measurement window is not particularly limited, for example, the measurement window can be set to about 10 to 20 seconds.

[0546] The temporal correlation of the functional connection can take a positive or negative value as its sign.

[0547] In addition, there may be cases such as, for example, in the case of a "healthy individual", the sign of the temporal correlation of the functional connection is negative, while in the case of a subject judged to have "depressive symptoms", the sign of the temporal correlation of the functional connection is positive. It should be understood that the opposite situation may exist, or the two signs may be the same. The sign and the magnitude of the absolute value of the temporal correlation of the functional connection are collectively referred to as the "pattern of temporal correlation".

[0548] In addition, for example, when a functional connection with a negative sign of temporal correlation is selected as the "functional connection to be trained" in the case of a "healthy individual", the "similarity" between the temporal correlation of the functional connection to be trained of the trainee and the temporal correlation of the healthy individual can be calculated by the following function F1.

[0549] i) Function F1, which takes a constant value V1 when the sign of the temporal correlation of the functional connection to be trained of the trainee is positive, and takes a value V2 (>V1≥0) when the sign is negative, so that as the absolute value of the temporal correlation increases, the value V2 increases.

[0550] Conversely, when a functional connection with a positive sign of temporal correlation is selected as the "functional connection to be trained" in the case of a "healthy individual", the "similarity" between the temporal correlation of the functional connection to be trained of the trainee and the temporal correlation of the healthy individual can be calculated by the following function F2.

[0551] (ii) Function F2, which takes a constant value V1 when the sign of the temporal correlation of the functional connection to be trained of the trainee is negative, and takes a value V2 (>V1≥0) when the sign is positive, so that as the absolute value of the temporal correlation increases, the value V2 increases.

[0552] The "similarity" is not limited to the above examples, and other functions can be adopted as long as the function is as follows: except for the sign, as the pattern becomes closer to the temporal correlation of the functional connection of a healthy individual, the "similarity" increases.

[0553] In addition, although there is no particular limitation, when selecting two or more "functional connections to be trained", the average value of the similarities can be set as the "similarity", or the average value of the similarities weighted by the contribution degrees of the respective connections can be set as the "similarity".

[0554] In addition, the data processing unit 102 calculates a reward value according to the similarity calculated as described above ( Figure 38 step S1003). According to the similarity with the connection strength of the target brain activity, the reward value (score SC) becomes higher as the similarity increases.

[0555] The data processing unit 102 displays the calculated score SC on the prompting device 6, thereby providing feedback to the trainee 2 ( Figure 38 step S1004). The information to be fed back can be the score value itself, or as Figure 35 shown, it can be a shape whose size changes according to the size of the value. Optionally, other structures can be adopted as long as the information is an image that enables the trainee to recognize the size of the score.

[0556] After that, the processes from EPI imaging to score feedback are repeated in real time within a predetermined time period.

[0557] Figure 36 is a diagram for showing an example of a training sequence in neurofeedback.

[0558] As Figure 36 shown, first, on the first day, the brain activity of the trainee in a resting state is measured in advance.

[0559] After that, training by neurofeedback is performed on the first day.

[0560] After that, neurofeedback training is also performed on the second day and the third day.

[0561] On the fourth day (the last day), after training by neurofeedback is performed, the subsequent brain activity of the trainee in a resting state is measured. The number of days of training may be less or more compared with this example.

[0562] In addition, after a predetermined time period has elapsed (for example, after two months), the brain activity of the trainee in a resting state is measured.

[0563] The therapeutic effect obtained through neurofeedback training can be judged by the discriminant device and discriminant method of the therapeutic effect using the first classifier described in the above section "5.", or by the discriminant method of the therapeutic effect of generating and using the second classifier described in the above section "7.".

[0564] Therefore, the description of the discriminant device and discriminant method of the therapeutic effect using the first classifier described in the above section "5.", or the discriminant method of the therapeutic effect of generating and using the second classifier described in the above section "7." is incorporated into this embodiment.

[0565] 9-2. Computer Program for Controlling Brain Activity Training Device

[0566] Another mode of the tenth embodiment includes a computer program for performing the processes of the above steps S1001 to S1004 and driving a brain activity training device. These programs can be stored in a storage medium (for example, a hard disk drive, a semiconductor storage element such as a flash memory, or an optical disk). The format of storing the program in the storage medium is not limited as long as the CPU 10040 can read the program. The storage medium for storing the program is preferably a non-volatile storage medium.

[0567] (Example)

[0568] I. Data Collection and Evaluation Method

[0569] 1. Data Collection Subjects

[0570] To select patients with major depressive disorder (MDD) based on the criteria of the Diagnostic and Statistical Manual of Mental Disorders (DSM)-IV, 105 patients recruited from Hiroshima University Hospital and clinics in Hiroshima City were screened using the Mini-International Neuropsychiatric Interview (M.I.N.I.). Patients with past or present manic symptoms, psychotic symptoms, alcohol dependence or abuse, drug dependence or abuse, or antisocial personality disorder were excluded. Finally, 93 patients with self-reported depressive symptoms were selected as the training data set for the MDD classifier. Before the start of drug administration or within 0 to 2 weeks after the start of drug administration, fMRI data of these patients were obtained under the conditions described later. In this area, 145 healthy individuals were recruited. These 145 healthy individuals underwent M.I.N.I. interviews, and individuals without a history of mental disorders were selected. The experiments in this example were conducted with the approval of the Ethics Committee of Hiroshima University. In addition, before the start of the experiment, all participants gave written informed consent.

[0571] In addition, a depression cohort was also collected from four completely independent facilities.

[0572] MDD includes three subtypes: melancholic MDD, non-melancholic MDD, and treatment-resistant MDD. Now, in the following examples, the group including all MDD patients, the group including melancholic MDD patients, the group including non-melancholic MDD patients, the group including treatment-resistant MDD patients, and the group of healthy individuals are set as the whole MDD group, melancholic MDD group, non-melancholic MDD group, treatment-resistant MDD group, and healthy control group, respectively.

[0573] The training dataset for generating the melancholic MDD classifier is limited to the subtype of depression (based on M.I.N.I.) with moderate depressive symptoms, where the age and gender are consistent with those of the healthy control group, based on the Beck Depression Inventory (BDI). The number of patients and the number of healthy individuals (whose age and gender are consistent with those of the patients and whose BDI-II score is less than 10) are set to be equal to each other so that no bias is generated between the groups (Table 2a). Regarding the scores of the Japanese Adult Reading Test (JART) for evaluating intelligence quotient (IQ), three missing data are found in the training dataset (one in the melancholic MDD group and two in the healthy control groups for both the whole MDD group and the melancholic MDD group), and two missing data are found in the treatment-resistant MDD group of the test dataset. Regarding the BDI scores, two missing data are found in the healthy control group only for the whole MDD group in the training dataset, and one missing data is found in the data after treatment with antidepressants.

[0574] In Table 2b, the details of the test data used in this example are shown.

[0575] [Table 2]

[0576]

[0577]

[0578] 2. Generalization of the independent external validation cohort

[0579] An independent validation cohort (the Chiba cohort) was created at the National Institute of Radiological Sciences in Japan. The lifetime history of mental disorders was evaluated for the participants based on M.I.N.I.

[0580] The MDD patients had no comorbid mental disorders, and the healthy control group had no physical, neurological, or mental disorders and no history of past or current substance abuse. All participants provided written informed consent prior to the experiment. The current experiment has been approved by the Radiopharmaceutical Safety Management Committee and the institutional review board of the National Institute of Radiological Sciences, Japan, according to the ethical standards defined by the Declaration of Helsinki in 1964 and its amendments after 1964.

[0581] 3. Evaluation of depression and depressive symptoms

[0582] The Beck Depression Inventory (BDI) was used to evaluate depression and depressive symptoms. In addition, the Hamilton Depression Rating Scale (HAMD) was also used to evaluate the treatment effect.

[0583] 4. Non-melancholic MDD and treatment-resistant MDD

[0584] The non-melancholic MDD group included all MDD patients with a BDI score of 17 or higher. Treatment-resistant MDD refers to those diagnosed with MDD whose depressive symptoms did not improve even when two or more antidepressants were administered.

[0585] 5. Evaluation of classifiers in other mental disorders

[0586] As fMRI data for patients with autism spectrum disorder (ASD) and schizophrenia spectrum disorder (SSD), the data described by Yahata et al. in the literature "Nature Communications|7:11254|DOI:10.1038 / ncomms 11254" were used. The ASD group was limited to those for whom antidepressants were ineffective to suppress the effects caused by the complication of depression. Similar to the MDD patients and healthy individuals, the fMRI data were acquired in the resting state with eyes open.

[0587] 6. Acquisition of fMRI data

[0588] When acquiring fMRI data, the subjects were required to keep watching the cross mark in the center of the monitor screen in a dimly lit scanning room without thinking about anything in particular and without sleeping. The details of the fMRI data acquisition conditions in each facility are shown in Table 3.

[0589] [Table 3]

[0590]

[0591]

[0592] 7. Preprocessing of fMRI imaging data and correlation between regions

[0593] All fMRI data were preprocessed using the same method described in the literature of Yahata et al.

[0594] SPM8 (Wellcome Trust Center for Neuroimaging, University College London, UK) of the following literature was used to preprocess the T1-weighted structural images and resting-state functional images: Noriaki Yahata, Jun Morimoto, Ryuichiro Hashimoto, Giuseppe Lisi, Kazuhisa Shibata, Yuki Kawakubo, Hitoshi Kuwabara, Miho Kuroda, Takashi Yamada, Fukuda Megumi, Hiroshi Imamizu, Jose′ E. Na′n~ez Sr, Hidehiko Takahashi, Yasumasa Okamoto, Kiyoto Kasai, Nobumasa Kato, Yuka Sasaki, Takeo Watanabe & Mitsuo Kawato, "A small number of abnormal brain connections predicts adult autism spectrum disorder", Nature Communications, DOI: 10.1038 / ncomms11254 Matlab R2014a (Mathworks Inc., USA). The functional images were preprocessed by slice-timing correction and alignment with the mean image. Then, the fMRI data were normalized using the normalization parameters obtained by segmenting the structural images synchronized by the mean functional image, and the data were resampled at 2×2×2 mm 3 voxel units. Finally, the functional images were smoothed with an isotropic full-width at half maximum Gaussian kernel of 6 mm. After these steps, any volumes (i.e., functional images) caused by additional head movement were removed by performing a cleaning procedure based on the relative changes between frames of the time-series data (for a summary of head movement, please refer to Table 4).

[0595] For each participant, fMRI data was extracted chronologically for each region of interest (ROI) among 137 ROIs that cover the entire cerebral cortex defined anatomically in the Brainvisa Sulci Atlas (BSA; http: / / brainvisa). In this embodiment, the structural and functional images of the cerebellum of the participant were not included in Facility 1, so the cerebellum was not included in the ROIs. After passing the data through a band-pass filter (0.008 Hz to 0.1 Hz), linear regression was performed on the following 9 parameters (6 head motion parameters after rearrangement, temporal fluctuations of white matter, temporal fluctuations of cerebrospinal fluid, fluctuations of the whole brain).

[0596] The pairwise Pearson correlations between the 137 ROIs were calculated to obtain the matrix of each functional connection among the 9316 functional connections (FCs) of each participant.

[0597] [Table 4]

[0598]

[0599]

[0600] II. Example 1: Selection of 12 pairs of functional connections for the classification of melancholic MDD

[0601] To select 12 pairs of functional connections for the classification of melancholic MDD, the rs-fMRI data of 66 melancholic MDD patients and 66 healthy individuals shown in Table 1a were used. Based on the method of creating a classifier for classifying autism spectrum disorder (ASD) reported in the above-mentioned literature by Yahata et al., the functional connections for classifying the melancholic MDD group were selected based on the process described in this embodiment.

[0602] The system uses L1-regularized sparse canonical correlation analysis (L1-SCCA) and sparse logistic regression (SLR). SLR is not used for classifying MDD, but has the ability to train a logistic regression model while pruning each functional connection in an objective manner. Before training by SLR, a certain amount of input was reduced by L1-SCCA, while reducing the influence of redundant variables (NV) that may cause fatal overfitting (overtraining). In this embodiment, facility, gender, and age were included in the random variables, so unnecessary factors were removed from these factors by L1-SCCA. In this method, consecutive steps of nested feature selection and leave-one-out cross-validation (LOOCV) using an inner loop and an outer loop as described later were used to avoid information leakage and overly optimistic results. As a result of SLR, 54 pairs of functional connections were output ( Figure 26)。In addition, LOOCV was performed to finally select 12 pairs of functional connections (Table 5 and Figure 26 )。

[0603] Consider whether the selected 12 pairs of functional connections FC are frequently selected with large weights throughout the outer loop. This is important for the stability and robustness of the finally selected 12 pairs of functional connections FC.

[0604] For this consideration, the following expression:

[0605] [Equation 12]

[0606]

[0607] defines the cumulative absolute weight of the k-th FC (k = 1, 2, …, 9316).

[0608] In the expression given above, N represents the number of times of LOOCV (i.e., the number of subjects), and w i k represents the weight related to the k-th functional connection FC of the i-th LOOCV.

[0609] The cumulative weight c i k The fact that it is a larger value means that: throughout the LOOCV, the k-th functional connection FC makes a greater contribution to the classification of MDD and HC.

[0610] [Table 5]

[0611]

[0612]

[0613] In Table 7, "L" and "R" in "Lat." represent the left and right brains in a differentiating manner. "BSA" represents Brodmann area, and "BA" represents the number of Brodmann areas. "rControl" represents the correlation coefficient of the healthy control group. "rMDD" represents the correlation coefficient of the melancholic MDD group. "Weight" represents the weight of the relevant weighted sum.

[0614] Figure 25 Shows the selected 12 pairs of functional connections and their respective weights.

[0615] Two functional connections with particularly large weights are the left dorsolateral prefrontal cortex (DLPFC, BA46) - left posterior cingulate cortex (PCC) / anterior cingulate gyrus and left inferior frontal gyrus (IFG operculum, BA44) - right DLPFC (BA9) / frontal eye field (FEF, BA8) / supplementary motor area (SMA, BA6). These functional connections overlap with the left DLPFC / IFG targeted by repetitive transcranial magnetic stimulation (rTMS) for the treatment of MDD in MDD.

[0616] In this way, a regression expression, i.e., a classifier, for classifying melancholic MDD patients and healthy individuals was created. In addition, a relevant weighted sum (associated weighted linear sum: WLS) was calculated as an index for determining whether each subject has melancholic MDD or is healthy.

[0617] In Figure 27 a of, the distributions of WLS in the melancholic MDD group and the healthy control group are shown. The black bars represent the melancholic MDD group, while the white bars represent the healthy control group. The accuracy was 70% (sensitivity: 64%, specificity: 77%, AUC: 0.77; and p =.049 in the permutation test: Figure 30 a of ). The results show that the classifier constructed from the selected 12 pairs of functional connections (hereinafter referred to as "the first classifier of the present invention") can distinguish melancholic MDD patients and healthy individuals.

[0618] In addition, the classifier obtained by the above method was also used to consider the cohort collected in Chiba City (including 11 melancholic MDD patients and 40 healthy individuals collected by Keio University Hospital). In Figure 27 b of, the results are shown. The accuracy of the cohort collected in Chiba City was 65% (sensitivity: 64%, specificity: 65%, AUC: 0.62; and p = 0.036 in the permutation test: Figure 30 b of ). The classifier of the present invention also successfully classified the cohort independent of the training data into melancholic MDD patients and healthy individuals.

[0619] Based on the above results, it is considered that the classifier of the present invention can classify melancholic MDD and is also generalized.

[0620] III. Example 2: Application of the classifier of the present invention to the non-melancholic MDD group and the entire MDD group

[0621] Figure 27c shows the following data, which is obtained by creating the classifier of the present invention using each of the entire MDD group, the melancholic MDD group, and the non-melancholic MDD group as training data (vertical direction), and considering the accuracy using each group as test data (horizontal direction). As a result of LOOCV, for example, the accuracy of the classifier generated by the melancholic MDD group for the non-melancholic MDD is 54% (sensitivity: 42%, specificity: 67%, AUC: 0.65). The accuracy of LOOCV for the entire MDD group is 66% (sensitivity: 58%, specificity: 74%, AUC: 0.74).

[0622] The results show the following fact: The classifier using the entire MDD group as training data has lower classification accuracy compared to the classifier using the melancholic MDD group as training data or the classifier using the melancholic MDD group as training data. In addition, the entire MDD group is considered to include more various subtypes of MDD.

[0623] Figure 28 d and e of Figure 29 f, g, and h of show the smoothed histograms of WLS and the AUC values obtained for each subject in the melancholic MDD group (d), non-melancholic MDD group (e), treatment-resistant MDD group (f), ASD group (g), and SSD group (h). The significant differences are calculated by the Benjamini-Hochberg corrected Kolmogorov-Smirnov test. The WLS data for each group is normalized to make the median and standard deviation of the healthy control group consistent. This normalization is not applicable to quantitative analysis. For the melancholic MDD group - healthy control group (d), AUC = 0.77 and p = 1.5×10 -5 value is obtained. For the non-melancholic MDD group - healthy control group (e), AUC = 0.65 and p =.051 value is obtained. For the treatment-resistant MDD group - healthy control group (f), AUC = 0.46 and p = 0.54 value is obtained. For the autism group - healthy control group (g), AUC = 0.51 and p = 0.74 value is obtained. For the schizophrenia group - healthy individual group (h), AUC = 0.43 and p = 0.038 value is obtained.

[0624] IV. Example 3: Evaluation of the severity of depression and evaluation of treatment effect using the WLS score

[0625] Verify the correlation between the BD score and the WLS score to study whether the WLS score is related to the severity of depression. As Figure 31As shown in a of [FIGURE REFERENCE], when calculating the correlation between the BDI score and the WLS score for the group including the entire MDD group and the healthy control group, values of r = 0.655 and p = 0.001 were obtained in the permutation test at n = 186, indicating that these two scores are correlated with each other. In addition, when only focusing on the entire MDD group ( Figure 31 b), values of r = 0.188 and p = 0.046 were obtained in the permutation test at n = 93, indicating that these two scores are relatively correlated with each other.

[0626] Next, it was considered whether the effect of escitalopram, a selective serotonin reuptake inhibitor (SSRI), could be evaluated by the WLS score. In Figure 31 c is shown the smoothed histogram of the WLS scores of each of the 24 melancholic MDD patients shown in Table 1a who had been alleviated after being administered escitalopram for 6 to 8 weeks. When the antidepressant was administered, the distribution of the WLS scores shifted towards the healthy control group. Values of AUC = 0.72 and p = 0.008 were obtained in the Benjamini-Hochberg corrected Kolmogorov-Smirnov test.

[0627] This result is related to the population and does not represent the relationship between WLS and individual improvement. Therefore, ΔWLS (specifically WLS post -WLS pre ) and the relationship between ΔBDI (specifically BDI post -BDI pre ), as well as the relationship between ΔWLS and ΔHAMD (specifically HAMD post -HAMD pre ) were considered. Significant differences were obtained through the permutation test. As Figure 31 shown in d, ΔBDI and ΔWLS are significantly correlated with each other (r = 0.373 and p = 0.040), but ΔHAMD and ΔWLS are not correlated with each other (r = 0.154 and p = 0.237). A deviation between BDI and HAMD has been reported.

[0628] Example 5: Changes in 12 pairs of functional connections before and after treatment

[0629] Next, the contribution to ΔWLS was considered for each functional connection to consider the effect of the SSRI (escitalopram) on each of the 12 pairs of functional connections.

[0630] Specifically, in the pre-treatment and subsequent analysis of the administration, the following numerical expression was used to calculate the contribution score of each functional connection.

[0631] [Mathematical formula 13]

[0632]

[0633] (In the expressions given above, N represents the number of patients who have received treatment, and w i represents the weight of the classifier.)

[0634] In addition, to obtain a reference point, the similarity contribution scores between the MDD and healthy control groups are also calculated through the following numerical expressions, and the contribution scores are thus calculated.

[0635] [Mathematical formula 14]

[0636]

[0637] (In the expressions given above, M represents the number of MDDs, and N represents the number of healthy controls.)

[0638] That is to say, s i represents the difference in the average connection strength between the healthy control group and the MDD group weighted by the weight of the classifier. In the classifier, MDD is positive (positive score), and healthy individuals are negative (negative score), so s i always takes a negative score. In addition, in Expression 2,

[0639]

[0640] represents the subtrahend (negative sign in subtraction). c i and s i taking large negative values means that the contribution of FC i in healthy individuals - MDD after treatment - before treatment is large. In other words, the fact that c i becomes negative means that the FC i after treatment is closer to the corresponding FC of healthy individuals. In contrast, the fact that c i becomes positive means that the FC i after treatment becomes closer to MDD. The number of members in the healthy control group and the MDD group is different from each other, so the Welch's t-test is used to evaluate the significant difference between c i and s i for each FC i . The Welch's t-test is calculated according to the following numerical expressions.

[0641] [Mathematical formula 15]

[0642]

[0643] (where: σ s 2 represents s iThe variance of the samples, and is calculated by using the law of total variance as FC i HC and FC i MDD (σ s 2 = σ HC 2 + σ MDD 2 ). Similarly, σ c 2 is also calculated as the sum of the variances of FC i post and FC i pre (σ c 2 = σ post 2 + σ pre 2 ). N s =(N CTRL + N MDD ) / 2 and N c =(N post + N pre ) / 2 represent the sample size. )

[0644] Adjust the p-value of each test for multiple comparisons by the Benjamini-Hochberg method. As Figure 32 shown by a, FC1 (i.e., DLPFC-precuneus / PCC) shows the largest significant difference between c i and s i . This indicates that after administration of the antidepressant, FC1 shifts towards the MDD group, which is opposite to the healthy control group.

[0645] Eight of the 12 functional connections used for the classification of melancholic MDD (the functional connection identification numbers shown in Table C can be represented by "FC#") shift towards the direction of the healthy control group and become "normal". However, four pairs of functional connections of FC#1, 3, 9, and 12 shift in the opposite direction to the healthy control group ( Figure 32 shown by a). Among them, FC#1 (left DLPFC-left precuneus / PCC FC) is the functional connection with the largest contribution among the 12 pairs of functional connections. In the paired comparison of the melancholic MDD group and the healthy control group, the significant difference in the change of FC#1 before and after treatment is p = 0.009 ( Figure 32 shown by a).

[0646] As a result of focusing on the two main functional connections (FC#1, FC#2) with the greatest contribution, significant differences were observed after treatment (p =.002 in the corresponding t-test), while no differences were observed before treatment (p = 0.96). In the test by two-way analysis of variance, there was a significant time interaction for the functional connections of FC#1 and FC#2 before and after treatment (FC# number × time, F(1,23) = 6.70, p = 0.016), and no significant differences were observed for each main effect. However, significant differences during the transition were observed in the paired t-tests of FC#1 and FC#2 after treatment (p = 0.002). In Figure 32 and Figure 33 In all graphs of, a shift in the positive direction indicates a shift towards melancholic MDD, while a shift in the negative direction indicates a shift towards healthy individuals. This is because the sign of the weight of the melancholic MDD classifier is multiplied by the difference of FC or FC. As described above, FC#1 has shifted in the positive direction after treatment with escitalopram, so it was verified whether the same results were obtained in a completely independent cohort. Considering 11 melancholic MDD patients collected in Chiba City (administered the serotonin-norepinephrine reuptake inhibitor (SNRI) duloxetine for 6 to 8 weeks) ( Figure 32 c). There was no significant difference before treatment (n.s., p = 0.77), while in the paired t-test after treatment, a difference in the trend level between FC#1 and FC#2 was observed (p = 0.10). Next, changes in FC#1 and FC#2 were observed when administering SSRI to healthy subjects (19 individuals). A significant difference between FC#1 and FC#2 was observed after a single administration of paroxetine (p = 0.033). There was no difference before treatment (p = 0.18), and no significant correlation was observed ( Figure 32 d).

[0647] In addition, by dividing the patients judged by clinicians to show treatment effects in the cohort in Hiroshima City into non-remission patients (n = 7) and remission patients (n = 7), the effect of antidepressant treatment was studied. As a result, a significant interaction was observed between the remission group and the non-remission group × the functional connections of FC#1 and FC#2 × the time points before and after treatment ( Figure 33For e, F(1, 22) = 8.86, p =.007), which indicates that there is a difference in the pattern of treatment response between FC#1 and FC#2. Specifically, after treatment with antidepressants in the remission-only group, the value of FC#2 decreased (p = 0.001), and a significant difference was observed in these values (p =.019). In addition, in the non-remission group, although there was no difference before treatment, a significant difference was observed between FC#1 and FC#2 after treatment with antidepressants (p = 0.033).

[0648] The changes in FC#1 and FC#2 were plotted for 24 patients with two-dimensional landmarks to consider how FC#1 and FC#2 changed in the remission and non-remission groups ( Figure 33 of f).

[0649] FC#1 and FC#2 were obtained through the following numerical expressions.

[0650] [Mathematical formula 16]

[0651] Δsign(W)FC1 = sign(w1)·(FC1 post - FC1 pre )

[0652] Δsign(W)FC2 = sign(w2)·(FC2 post - FC2 pre )

[0653] FC1 post represents FC1 after administration, FC1 pre represents FC1 before administration, FC2 post represents FC2 after administration, FC2 pre represents FC2 before administration, and "sign" represents the sign of the weight of each functional connection. The precision of this expression is: accuracy: 0.75, AUC: 0.79, specificity: 0.88, sensitivity: 0.43.

[0654] As a result, the changes in FC#1 and FC#2 contributed significantly to remission. On the other hand, no significant difference or correlation was observed between the change in FC#1 and the change in FC#2 (p = 0.57).

[0655] V. Example 4: Neurofeedback V-1. Selection of target functional connections for FC neurofeedback training

[0656] According to the following procedure, the first functional connection shown in Table 5 was selected as the target for neurofeedback.

[0657] First, as a biomarker of rs-fc MRI, a biomarker was constructed to predict the severity of depressive symptoms based on the classifier of the present invention and the BDI score (Yamashita A, Hayasaka S, Lisi G, Ichikawa N, Takamura M, Okada G, Morimoto J, Yahata N, Okamoto Y, Kawato M, Imamizu H (2015) Common functional connectivity between depression and depressed mood. The 37th annual meeting of the Japanese society of biological psychiatry; 24-26 September; Tokyo, Japan).

[0658] The target functional connectivity was set to include the above-mentioned biomarker.

[0659] During the neurofeedback training, the participants attempted to reduce the correlation of the target FC.

[0660] A total of 10 participants, including 3 participants with MDD and 7 participants with subclinical depression, participated in the training. The average value of the BDI-II of the subclinical depression participants measured at two different time points before the neurofeedback training was greater than 10.

[0661] During the neurofeedback training, the participants lay supine in the fMRI device, viewed the monitor in the device via the prism glasses, and imagined Figure 35 the connection index of the functional connectivity shown to move in the negative direction.

[0662] The neurofeedback score was calculated according to a well-known method.

[0663] Figure 37 Panel a shows the aggregation of the neurofeedback scores of the 3 MDD participants on their respective training days. The scores showed an increasing trend during the 4 days of training. When examining the significance of the results using a multiple regression model including 2 explanatory variables (each training day and training item) and one response variable (neurofeedback score), a significantly good effect was shown on the training day (95% confidence interval (CI) coefficient: 1.9 - 9.1). In a t-test conducted as a subsequent comparison, for all 3 MDD participants, significantly higher neurofeedback scores were obtained on the last day compared to the first day of training (t = 4.01, P <.001). These results indicate that all MDD participants learned to induce a negative correlation of the target FC through training. In addition, asFigure 37 As shown in b of [Figure 0], after training, for all three MDD participants, the Hamilton Depression Scale (HAMD), which represents the severity of depressive symptoms, decreased.

[0664] Similar to the MDD participants, seven subclinical depression participants had a tendency for increased neurofeedback scores during training ( Figure 37 as shown in c of [Figure 0]). When analyzing the significance of this result, in a one-way analysis of variance (one-way ANOVA), the main effect of the training day was significant (p = 0.0046). In the subsequent paired t-test for comparison, compared with the first day of training, the neurofeedback scores were significantly higher on the last day of training. There was also a tendency for the BDI scores to decrease after training (p =.07). Additionally, as Figure 37 shown in d of [Figure 0], five out of seven subclinical depression patients changed the target rs-fc MRI in the normal direction, and the change amount of rs-fc MRI before and after training was significantly correlated with the BDI score (r = 0.87, p = 0.011).

[0665] As described above, more than half of the participants reduced the functional connectivity shown in Table 5, which is related to the BDI score, through training. Therefore, this indicates that the functional connectivity shown in Table 5 can be used as the target functional connectivity for neurofeedback training to improve depressive symptoms. Additionally, this indicates that the functional connectivity shown in Table 5 can be used as a therapeutic method or biomarker for developing a neurofeedback training method to improve depressive symptoms.

[0666] V-2. Comparison with Other Antidepressant Treatments

[0667] The therapeutic effect on MDD obtained through the neurofeedback training method was compared with other treatment methods (administration of antidepressants, repetitive transcranial magnetic stimulation: rTMS, and electroconvulsive therapy without convulsions: ECT intervention). The neurofeedback training method improved the therapeutic effect compared with other antidepressant treatment methods.

[0668] Explanation of Reference Numerals

[0669] 2 Subjects, 6 Monitors, 10 MRI Devices, 11 Magnetic Field Application Mechanisms, 12 Static Magnetic Field Generation Coils, 14 Gradient Magnetic Field Generation Coils, 16 RF Irradiation Units, 18 Beds, 20 Receiver Coils, 21 Drive Units, 22 Static Magnetic Field Power Supplies, 24 Gradient Magnetic Field Power Supplies, 26 Signal Transmission Units, 28 Signal Reception Units, 30 Bed Drive Units, 32 Data Processing Units, 36 Storage Units, 38 Display Units, 40 Input Units, 42 Control Units, 44 Interface Units, 46 Data Collection Units, 48 Image Processing Units, 50 Network Interfaces.

Claims

1. A discrimination device, comprising: A storage device configured to store: A program; And Information for identifying a classifier, the classifier being generated through classifier generation processing based on resting state fMRI signals obtained by measuring signals of brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and depression patients in a resting state in advance and in chronological order by using a functional magnetic resonance imaging (fMRI) imaging device, Generate the classifier to discriminate a disease label of depressive symptoms based on a weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the multiple predetermined regions; A processor; And An interface provided between the processor and the storage device, Wherein, the processor is configured to: When the program is being executed, Generate a first value for evaluating depressive symptoms for elements of a correlation matrix of functional connections measured for a subject in a resting state at a first time point by using the classifier obtained from the storage device via the interface, Generate a second value for evaluating depressive symptoms for elements of a correlation matrix of the same functional connections inside the brain measured for the same subject in a resting state at a second time point, the second time point being a time point after the start of treatment and later than the first time point, by using the classifier, and Compare the first value with the second value to perform processing for discriminating the treatment effect on the subject based on the following criterion i) and / or criterion ii): i) When the second value is improved compared with the first value, it is determined that the treatment is effective for improving the depressive symptoms of the subject, and ii) When the second value is not improved compared with the first value, it is determined that the treatment is ineffective for improving the depressive symptoms of the subject, and Wherein, the multiple selected functional connections include at least one functional connection selected from the following: A first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex; and A second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area, The classifier generation processing includes the following operations: Calculate a correlation matrix of functional connections of the multiple predetermined regions for each participant based on the resting state fMRI signals of each participant, Sequentially select one subset from different K subsets extracted from multiple participants, and perform sparse canonical correlation analysis on multiple attribute information and elements of the correlation matrix for the K - 1 subsets other than the selected subset, thereby extracting specific attribute information from the multiple personal attribute information, and obtain a first union set that is a union set of elements of the extracted correlation matrix for the sequentially selected subset, where K is a natural number of 2 or more, When setting the remaining participants among the multiple participants except for K subsets as a test set and dividing the test set into N different groups, through sparse logistic regression, for the set of participants among the multiple participants except for the selected group among the N groups, a test classifier for estimating a disease label for discriminating depressive symptoms is calculated based on the first sum set, and elements of a correlation matrix that are explanatory variables of the test classifier due to sparsification are extracted. Further, one group is sequentially selected from the N groups, and feature extraction is repeated to obtain a second sum set, and Using the second sum set as an explanatory variable, the classifier for estimating a disease label for discriminating depressive symptoms is calculated through sparse logistic regression, and information for identifying the generated classifier is stored in the storage device. Among them, the multiple functional connections are the functional connection with the functional connection identification number 1 in Table 7 and the functional connection with the functional connection identification number 2. [Table 7] 2. A storage medium storing a program for causing a computer to execute the functions of the discrimination device according to claim 1.

3. A discrimination device, comprising a processor configured to: Perform a classifier generation process for pre-generating a classifier using information obtained by measuring the correlations at a first time point of a plurality of functional connections selected from functional connection identification numbers 1 to 12 shown in Table 1 and the correlations at a second time point of the plurality of functional connections for a plurality of subjects, wherein the classifier is configured to distinguish, in a correlation state space expanded by the difference between the correlations at the first time point and the second time point of the plurality of functional connections, a group of subjects showing a treatment effect among the plurality of subjects from a group of subjects not showing a treatment effect among the plurality of subjects, and the second time point is set to be after the start of treatment and later than the first time point; Calculate a first correlation of the plurality of functional connections inside the brain of a subject at rest at the first time point by using a functional magnetic resonance imaging (fMRI) imaging device; Using the fMRI imaging device, calculate the second correlation of the multiple functional connections inside the brain of the same subject at rest at the second time point. And Based on the difference between the first correlation and the second correlation of the multiple functional connections of the subject, by using the classifier to discriminate the treatment effect for the subject, a process for judging the treatment effect for a depression patient is performed: [Table 1] The classifier generation process includes the following operations: Based on the resting-state fMRI signals of each subject, calculate the correlation matrix of the functional connections of multiple predetermined regions of each brain for each subject. Sequentially select one subset from different K subsets extracted from multiple subjects, and perform sparse canonical correlation analysis on multiple attribute information and elements of the correlation matrix for the K - 1 subsets other than the selected subset, thereby extracting specific attribute information from the multiple personal attribute information, and obtaining a first sum set that is the sum set of the elements of the extracted correlation matrix for the sequentially selected subset. Here, K is a natural number of 2 or more. When setting the remaining subjects among the multiple subjects except for K subsets as a test set and dividing the test set into N different groups, through sparse logistic regression, for the set of subjects among the multiple subjects except for the selected group among the N groups, a test classifier for estimating a disease label for discriminating depressive symptoms is calculated based on the first sum set, and elements of a correlation matrix that are explanatory variables of the test classifier due to sparsification are extracted. Further, one group is sequentially selected from the N groups, and feature extraction is repeated to obtain a second sum set, and Using the second sum set as an explanatory variable, the classifier for estimating a disease label for discriminating depressive symptoms is calculated through sparse logistic regression, and information for identifying the generated classifier is stored in the storage device. Among them, the multiple functional connections are the functional connection with the functional connection identification number 1 and the functional connection with the functional connection identification number 2.

4. The discrimination device according to claim 3, wherein, The discrimination device is also used for drug reanalysis.

5. A discrimination method for assisting in judging the treatment effect for a depression patient, the discrimination method comprising the following steps: Using the information obtained by measuring the correlations at a first time point of a plurality of functional connections selected from functional connection identification numbers 1 to 12 shown in Table 2 for a plurality of subjects and the correlations at a second time point of the plurality of functional connections, a classifier is pre-generated, the classifier being configured to distinguish, in a correlation state space expanded by the difference between the correlations at the first time point and the second time point of the plurality of functional connections, a group of subjects showing a treatment effect among the plurality of subjects from a group of subjects not showing a treatment effect among the plurality of subjects, the second time point being set to be after the start of treatment and later than the first time point; Measuring, using a functional magnetic resonance imaging (fMRI) imaging device, a first correlation of the plurality of functional connections inside the brain of a subject at rest at the first time point; Measuring, using the fMRI imaging device, a second correlation of the plurality of functional connections inside the brain of the same subject at rest at the second time point; And Using the classifier to determine the treatment effect for the subject based on the difference between the first correlation and the second correlation of the plurality of functional connections of the subject; [Table 2] ， The classifier generation process includes the following operations: Calculating, for each subject, a correlation matrix of the functional connections of a predetermined region of each brain based on the resting-state fMRI signal of each subject; Sequentially selecting one subset from different K subsets extracted from a plurality of subjects, and performing sparse canonical correlation analysis on a plurality of attribute information and the elements of the correlation matrix for the K - 1 subsets other than the selected subset, thereby extracting specific attribute information from the plurality of personal attribute information, and obtaining a first union set that is the union set of the elements of the extracted correlation matrix for the sequentially selected subset, where K is a natural number greater than or equal to 2; When setting the remaining subjects other than the K subsets among the plurality of subjects as a test set and dividing the test set into N different groups, through sparse logistic regression, for the set of subjects other than the selected group among the N groups among the plurality of subjects, calculating a test classifier for estimating a disease label for discriminating depressive symptoms based on the first union set, and extracting the elements of the correlation matrix that are the explanatory variables of the test classifier due to sparsification, further sequentially selecting one group from the N groups, repeating feature extraction to obtain a second union set, and Calculating the classifier for estimating a disease label for discriminating depressive symptoms using the second union set as an explanatory variable through sparse logistic regression, and storing the information for identifying the generated classifier in a storage device; wherein the plurality of functional connections are a functional connection having a functional connection identification number 1 and a functional connection having a functional connection identification number 2.

6. A method for using a classifier to assist in determining whether a subject has depressive symptoms or the level of depressive symptoms, where the classifier is generated based on signals obtained by pre-measuring resting-state fMRI signals representing brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and patients with depression in a resting state using a functional magnetic resonance imaging (fMRI) camera device in chronological order, and through classifier generation processing. Generate the classifier to discriminate the disease label of depressive symptoms based on the weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the multiple predetermined regions. The multiple selected functional connections include at least one functional connection selected from the following: A first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex; and A second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. The method includes the following steps: Inputting an index value generated based on the elements of the correlation matrix of the plurality of functional connections of the subject into the classifier; The classifier generation process includes the following operations: Calculating, for each participant, a correlation matrix of the functional connections of the plurality of predetermined regions based on the resting-state fMRI signal of each participant; Successively select one subset from different K subsets extracted from multiple participants, and perform sparse canonical correlation analysis on multiple attribute information and elements of the correlation matrix for K-1 subsets other than the selected subset, thereby extracting specific attribute information from multiple personal attribute information, and obtain a first union set that is the union set of the elements of the extracted correlation matrix for the successively selected subset, where K is a natural number of 2 or more. When setting the remaining participants other than the K subsets among the multiple participants as a test set and dividing the test set into N different groups, through sparse logistic regression, for the set of participants other than one selected group among the N groups among the multiple participants, calculate a test classifier for estimating the disease label for discriminating depressive symptoms based on the first union set, and extract the elements of the correlation matrix that are the explanatory variables of the test classifier due to sparsification. Further successively select one group from the N groups, repeat feature extraction to obtain a second union set, and Use the second union set as an explanatory variable, and calculate the classifier for estimating the disease label for discriminating depressive symptoms through sparse logistic regression, and store the information for identifying the generated classifier in a storage device. Among them, the multiple functional connections are the functional connection with the functional connection identification number 1 in Table 8 and the functional connection with the functional connection identification number 2. [Table 8] 7. A discrimination device, comprising: A storage device configured to store a program and information for classifying depression into a plurality of preset subclasses. A processor configured to: When the program is in execution, Based on signals obtained by measuring, in advance and chronologically, resting-state fMRI signals representing brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and depression patients in a resting state using a functional magnetic resonance imaging (fMRI) imaging device, generate a first classifier, and Store the information of the generated first classifier in the storage device. Among them, the first classifier is generated to discriminate the disease label of depressive symptoms based on the weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the multiple predetermined regions. Among them, the selected multiple functional connections include at least one functional connection selected from the following: The first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex; and The second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area. Among them, the processor is further configured to: For multiple subjects, respectively measure the correlation at a first time point of multiple functional connections selected from the functional connection identification numbers 1 to 12 shown in Table 3 and the correlation at a second time point of the multiple functional connections, where the second time point is set after the start of treatment and is later than the first time point. Generate a second classifier, which is used to distinguish a group of subjects showing a treatment effect among the multiple subjects from a group of subjects not showing a treatment effect among the multiple subjects in the correlation state space expanded by the difference in the correlation between the first time point and the second time point of the multiple functional connections; and Store the information of the generated second classifier in the storage device, and wherein the processor is configured to: Perform a process of classifying subjects into multiple subclasses by using the first classifier; Perform a process of measuring the first correlation of the multiple functional connections in the resting state at the first time point for the subjects classified into a specific subclass through the classification process; Perform a process of measuring the second correlation of the multiple functional connections in the resting state at the second time point for the same subjects; and Based on the difference between the first correlation and the second correlation of the multiple functional connections of the subject, perform a process of determining the treatment effect for the subject by using the second classifier: [Table 3] ， The classifier generation process includes the following operations: Calculate the correlation matrix of the functional connections of the multiple predetermined regions for each participant based on the resting state fMRI signals of each participant, Sequentially select one subset from different K subsets extracted from multiple participants, and perform sparse canonical correlation analysis on multiple attribute information and the elements of the correlation matrix for the K-1 subsets other than the selected subset, thereby extracting specific attribute information from the multiple personal attribute information, and obtain a first union set as the union of the elements of the extracted correlation matrix for the sequentially selected subset, where K is a natural number greater than or equal to 2, When setting the remaining participants among the multiple participants other than the K subsets as a test set and dividing the test set into N different groups, through sparse logistic regression, for the set of participants other than the participants in one selected group among the N groups among the multiple participants, calculate a test classifier for estimating the disease label for discriminating depressive symptoms based on the first union set, and extract the elements of the correlation matrix that are the explanatory variables of the test classifier due to sparsification, further sequentially select one group from the N groups, and repeat feature extraction to obtain a second union set, and Use the second union set as the explanatory variable, calculate the classifier for estimating the disease label for discriminating depressive symptoms through sparse logistic regression, and store the information for identifying the generated classifier in the storage device, wherein the multiple functional connections are the functional connection with functional connection identification number 1 and the functional connection with functional connection identification number 2.

8. The discrimination device according to claim 7, wherein, The discrimination device is used for drug reanalysis.

9. A discrimination method for assisting in judging the treatment effect on a subject, the discrimination method comprising the following steps: In the case of classifying depressive symptoms into multiple preset subclasses, Classification step for classifying a subject into the multiple subclasses by using a first classifier generated by processing signals obtained by measuring resting-state fMRI signals representing brain activities of multiple predetermined regions of each brain of multiple participants including healthy individuals and patients with depression in a resting state in advance and chronologically by using a functional magnetic resonance imaging (fMRI) imaging device. The first classifier is generated to discriminate a disease label of depressive symptoms based on a weighted sum of multiple functional connections selected as being related to the disease label of depressive symptoms through feature selection via machine learning from the functional connections of the multiple predetermined regions. The multiple functional connections selected include at least one functional connection selected from the following: A first functional connection between the left dorsolateral prefrontal cortex and the left precuneus and the left posterior cingulate cortex, and A second functional connection between the left inferior frontal gyrus opercular part and the right dorsomedial prefrontal cortex and the right supplementary motor area; A first correlation measurement step for measuring a first correlation of the multiple functional connections in a resting state before the start of treatment for a subject classified into a specific subclass in the classification step; a second correlation measurement step for measuring a second correlation of the multiple functional connections in a resting state after a predetermined period of time has elapsed since the start of the treatment for the same subject; A discrimination step for discriminating a treatment effect for a subject classified into the specific subclass by using a second classifier. The second classifier is generated in advance by a second classifier generation process for measuring, for multiple subjects, a correlation at a first time point and a correlation at a second time point of multiple functional connections selected from functional connection identification numbers 1 to 12 shown in Table 4, thereby distinguishing, in a correlation state space expanded by a difference between the correlation at the first time point and the correlation at the second time point of the multiple functional connections, a group of subjects showing a treatment effect among the multiple subjects from a group of subjects not showing a treatment effect among the multiple subjects. The second time point is set to be after the start of the treatment and later than the first time point. And Based on a difference between the first correlation and the second correlation of the multiple functional connections of the subject, discriminating a treatment effect for the subject: [Table 4] ， The classifier generation process includes the following operations: Calculating a correlation matrix of the functional connections of the multiple predetermined regions for each participant based on the resting-state fMRI signals of each participant. Sequentially selecting one subset from different K subsets extracted from multiple participants, and performing sparse canonical correlation analysis on multiple attribute information and elements of the correlation matrix for the K - 1 subsets other than the selected subset, thereby extracting specific attribute information from the multiple personal attribute information, and obtaining a first union set that is a union set of elements of the extracted correlation matrix for the sequentially selected subset, where K is a natural number of 2 or more. When setting the remaining participants among the multiple participants except for K subsets as a test set and dividing the test set into N different groups, through sparse logistic regression, for the set of participants among the multiple participants except for the selected group among the N groups, a test classifier for estimating a disease label for discriminating depressive symptoms is calculated based on the first sum set, and elements of a correlation matrix that are explanatory variables of the test classifier due to sparsification are extracted. Further, one group is sequentially selected from the N groups, and feature extraction is repeated to obtain a second sum set, and using the second sum set as an explanatory variable, a classifier for estimating a disease label for discriminating depressive symptoms is calculated through sparse logistic regression, and information for identifying the generated classifier is stored in a storage device. wherein the multiple functional connections are a functional connection having a functional connection identification number 1 and a functional connection having a functional connection identification number 2.

10. A classifier generation device, comprising: A processor configured to measure, for multiple subjects, the correlation at a first time point of multiple functional connections selected from functional connection identification numbers 1 to 12 shown in Table 5 and the correlation at a second time point of the multiple functional connections, thereby generating a classifier configured to distinguish, in a correlation state space expanded by the difference between the correlation at the first time point and the correlation at the second time point of the multiple functional connections, a group of subjects among the multiple subjects who exhibit a treatment effect from a group of subjects among the multiple subjects who do not exhibit a treatment effect, and the second time point is set to be after the start of treatment and later than the first time point; and A storage device configured to store information for identifying the classifier generated by the processor: [Table 5] wherein the multiple functional connections are obtained using a functional magnetic resonance imaging (fMRI) imaging device; the multiple functional connections are a functional connection having a functional connection identification number 1 and a functional connection having a functional connection identification number 2.

11. A method for generating a second classifier, the method comprising the following steps: For multiple subjects, measure the correlation at a first time point of multiple functional connections selected from functional connection identification numbers 1 to 12 shown in Table 6 and the correlation at a second time point of the multiple functional connections, thereby pre-generating a classifier configured to distinguish, in a correlation state space expanded by the difference between the correlation at the first time point and the correlation at the second time point of the multiple functional connections, a group of subjects among the multiple subjects who exhibit a treatment effect from a group of subjects among the multiple subjects who do not exhibit a treatment effect, and the second time point is set to be after the start of treatment and later than the first time point: [Table 6] wherein the multiple functional connections are obtained using a functional magnetic resonance imaging (fMRI) imaging device; the multiple functional connections are a functional connection having a functional connection identification number 1 and a functional connection having a functional connection identification number 2.

Images