Method, information processing apparatus, program, and trained model

Abstract

A method for diagnosing, prognosing, or monitoring a response to treatment of a mental disorder in a subject to be executed by one or more information processing apparatuses, the method includes selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of a subject and a target region of the brain of the subject; calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; inputting the measure of directed flow of information for selected ROI pairs into a trained model; and outputting information pertaining to a mental disorder of the subject based on the trained model.

Description

METHOD, INFORMATION PROCESSING APPARATUS, PROGRAM, AND TRAINED MODELCROSS-REFERENCE TO RELATED APPLICATION

[0001] The present application claims priority to Japanese Patent Application No. 2024-137064 filed on August 16, 2024, and Japanese Patent Application No. 2025-048784 filed on March 24, 2025, the entire contents of which are incorporated herein by reference.

[0002] The present disclosure relates to a method, an information processing apparatus, a program, and a trained model.Background

[0003] Mental disorders are a tremendous burden to mankind, from economic, social, and personal perspectives. For example, schizophrenia affects about 24 million individuals worldwide, including 2.8 million adults in the USA (1.1% of the population), and 793,000 in Japan. Moreover, approximately half of these individuals have co-occurring mental and / or behavioral health disorders, like bipolar disorder, autism spectrum disorder, anxiety disorder, and depression. Also, autism spectrum disorder which affects 1 in 100 people worldwide. Today’s diagnosis of the Mental disorders relies on the Diagnostic and Statistical Manual of Mental Disorders (DSM-V) (for example, Non-Patent Literature (NPL) 1)

[0004] NPL 1: Diagnostic and Statistical Manual of Mental Disorders (DSM-5-TR) Internet <https: / / www.psychiatry.org / psychiatrists / practice / dsm>Summary

[0005] However, all mental disorders are difficult to diagnose, to prognose, and to monitor based on the behavior alone. This is because they all have a wide range of underlying maladaptations / malfunctions at a molecular level. It is desired to provide highly accurate information pertaining to mental disorders that can be used for diagnosis, prognosis, and monitoring. As such, it would be helpful to improve technology related to providing information pertaining to mental disorders.

[0006] In light of these circumstances, it is an aim of the present disclosure to improve technology related to providing information pertaining to mental disorders. (Solution to Problem)

[0007] (1) A method for diagnosing, prognosing, or monitoring a response to treatment of a mental disorder in a subject according to an embodiment of the present disclosure is a method to be executed by one or more information processing apparatuses, the method includes: selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of the subject and a target region of the brain of the subject; calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; inputting the measure of directed flow of information for selected ROI pairs into a trained model; and outputting information pertaining to a mental disorder of the subject based on the trained model.

[0008] (2) The method according to an embodiment of the present disclosure is the method according to (1), wherein the measure of directed flow of information is transfer entropy.

[0009] (3) The method according to an embodiment of the present disclosure is the method according to (2), wherein calculating transfer entropy for each ROI pair includes: calculating transfer entropy for each ROI pair based on a time delay for transferring information from a source region of the brain of the subject to a target region of the brain of the subject in each ROI pair, wherein the time delay has been calculated based on an average value of a time delay at which the transfer entropy for each ROI pair is a maximum.

[0010] (4) The method according to an embodiment of the present disclosure is the method according to any one of (1) to (3), wherein the mental disorder is schizophrenia, bipolar disorder, attention deficit / hyperactivity disorder, depression, drug-resistant depression, anxiety, early psychosis, autism spectrum disorder, epilepsy, Alzheimer's, or dementia.

[0011] (5) The method according to an embodiment of the present disclosure is the method according to any one of (1) to (4), wherein the method is that for monitoring a response to treatment, the selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of the subject and a target region of the brain of the subject, the calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair, the inputting the measure of directed flow of information for selected ROI pairs into a trained model, and the outputting information pertaining to a mental disorder of the subject based on the trained model are performed on the subject before and after starting the treatment; and wherein the method further comprises comparing the information pertaining to a mental disorder of the subject before starting the treatment and the information pertaining to a mental disorder of the subject after starting the treatment.

[0012] (6) A method for diagnosing, prognosing, or monitoring a response to treatment of a mental disorder in a subject according to an embodiment of the present disclosure is a method, including: selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of the subject and a target region of the brain of the subject, calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; and presenting, based on the measure of directed flow of information for selected ROI pairs, information pertaining to a mental disorder of the subject.

[0013] (7) The method according to an embodiment of the present disclosure is the method according to (6), wherein the measure of directed flow of information is transfer entropy.

[0014] (8) The method according to an embodiment of the present disclosure is the method according to (7), wherein the presenting, based on transfer entropy for each ROI pair, the information pertaining to the mental disorder of the subject includes: classifying the subject by a mathematical model based on the transfer entropy of selected ROI pairs and presenting the information pertaining to the mental disorder of the subject.

[0015] (9) The method according to an embodiment of the present disclosure is the method according to (8), wherein the mathematical model is a K-Nearest-Neighbor classifier, logistic regression classifier, support vector machine classifier or Naive Bayes classifier.

[0016] (10) The method according to an embodiment of the present disclosure is the method according to (8), wherein the mathematical model is a K-Nearest-Neighbor classifier and cosine similarity is used as a distance function.

[0017] (11) The method according to an embodiment of the present disclosure is the method according to any one of (6) to (10), wherein the mental disorder is schizophrenia, bipolar disorder, attention deficit / hyperactivity disorder, depression, drug-resistant depression, anxiety, early psychosis, autism spectrum disorder, epilepsy, Alzheimer's, or dementia.

[0018] (12) The method according to an embodiment of the present disclosure is the method according to any one of (6) to (11), wherein the method is that for monitoring a response to treatment, the selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of the subject and a target region of the brain of the subject, the calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; and the presenting, based on the measure of directed flow of information for selected ROI pairs, information pertaining to a mental disorder of the subject are performed on the subject before and after starting the treatment; and wherein the method further comprises comparing the information pertaining to a mental disorder of the subject before starting the treatment and the information pertaining to a mental disorder of the subject after starting the treatment.

[0019] (13) A method for diagnosing, prognosing, or monitoring a response to treatment of, a mental disorder in a subject to be executed by one or more information processing apparatuses, the method comprising: selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of the subject and a target region of the brain of the subject; calculating, for each ROI pair, a time delay for transferring information at which the transfer entropy from the source region to the target region is maximized; inputting the time delays for selected ROI pairs into a trained model; and outputting information pertaining to a mental disorder of the subject based on the trained model.

[0020] (14) The method according to (13), wherein the mental disorder is schizophrenia, bipolar disorder, attention deficit / hyperactivity disorder, depression, drug-resistant depression, anxiety, early psychosis, autism spectrum disorder, epilepsy, Alzheimer's, or dementia.

[0021] (15) The method according to (1), wherein the measure of directed flow of information is a time delay for transferring information.

[0022] (16) The method according to (2), wherein calculating transfer entropy for each ROI pair comprises: calculating transfer entropy for each ROI pair based on a time delay for transferring information from a source region of the brain of the subject to a target region of the brain of the subject in each ROI pair at which the transfer entropy from the source region to the target region is maximized.

[0023] (17) The method according to (16), wherein the mental disorder is schizophrenia, bipolar disorder, attention deficit / hyperactivity disorder, depression, drug-resistant depression, anxiety, early psychosis, autism spectrum disorder, epilepsy, Alzheimer's, or dementia.

[0024] (18) The method according to (1) or (16), wherein the mental disorder is Parkinson’s Disease.

[0025] (19) An information processing apparatus according to an embodiment of the present disclosure is an information processing apparatus including a controller, wherein the controller is configured to select a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of a subject and a target region of the brain of the subject; calculate a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; input the measure of directed flow of information for selected ROI pairs into a trained model; and output information pertaining to a mental disorder of the subject based on the trained model.

[0026] (20) A program according to an embodiment of the present disclosure is a program configured to cause a computer to execute operations including: selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of a subject and a target region of the brain of the subject; calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; inputting the measure of directed flow of information for selected ROI pairs into a trained model; and outputting information pertaining to a mental disorder of the subject based on the trained model.

[0027] (21) A trained model according to an embodiment of the present disclosure is a trained model to cause a computer to function to output an objective variable pertaining to a mental disorder of a subject based on explanatory variables pertaining to the data of the subject, wherein the explanatory variables comprise a measure of directed flow of information for each region of interest (ROI) pair, the measure of directed flow of information being calculated by selecting a plurality of ROI pairs, each ROI pair including a source region of a brain of a subject and a target region of the brain of the subject; and calculating based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair, and wherein the objective variables comprise information pertaining to the mental disorder of the subject. (Advantageous Effect)

[0028] According to an embodiment of the present disclosure, technology related to providing information pertaining to mental disorders is improved.

[0029] In the accompanying drawings: FIG. 1 is a block diagram illustrating a schematic configuration of an information processing apparatus according to an embodiment of the present disclosure; FIG. 2 is a flowchart illustrating operations of the information processing apparatus according to an embodiment of the present disclosure; FIG. 3 is a flowchart illustrating method according to an embodiment of the present disclosure; FIG. 4A is a result of the bootstrap estimates of the Control; FIG. 4B is a result of the bootstrap estimates of Schizophrenia; FIG. 5A is a diagram illustrating the locations of Source ROIs of the Control; FIG. 5B is a diagram illustrating the locations of Source ROIs of Schizophrenia; FIG. 6A is a diagram illustrating the locations of Target ROIs of the Control; FIG. 6B is a diagram illustrating the locations of Target ROIs of Schizophrenia; FIG. 7A is a diagram illustrating the inter-regional directed functional flow of information of the Control; FIG. 7B is a diagram illustrating the inter-regional directed functional flow of information of Schizophrenia; FIG. 8A is the result of the Schizophrenia versus Control Leave-One-Out Classification; FIG. 8B is the result of the Schizophrenia versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 9A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 9B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 10 is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Schizophrenia versus Control for the case of Stratified Cross-Validation in FIG. 8B; FIG. 11A is the result of the Schizophrenia versus Bipolar Leave-One-Out Classification; FIG. 11B is the result of the Schizophrenia versus Bipolar stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 12A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 12B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 13 is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Schizophrenia versus Bipolar for the case of Stratified Cross-Validation Classification in FIG 11B; FIG. 14A is the result of the Affective versus Non-Affective Psychosis Leave-One-Out Classification; FIG. 14B is the result of the Control versus Affective Psychosis Leave-One-Out Classification; FIG. 14C is the result of the Control versus Non-Affective Psychosis Leave-One-Out Classification; FIG. 15A is the result of the Affective versus Non-Affective Psychosis stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 15B is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Affective versus Non-Affective Psychosis for the case of Stratified Cross-Validation Classification in FIG 15A; FIG. 16A is the result of the Control versus Affective Psychosis stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 16B is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Control versus Affective Psychosis for the case of Stratified Cross-Validation Classification in FIG 16A; FIG. 17A is the result of the Control versus Non-Affective Psychosis stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 17B is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Control versus Non-Affective Psychosis for the case of Stratified Cross-Validation Classification in FIG 17A; FIG. 18A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 18B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 19A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 19B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 20A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 20B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 21A is the result of the Anxiety versus Depression Leave-One-Out Classification; FIG. 21B is the result of the Control versus Anxiety Leave-One-Out Classification; FIG. 21C is the result of the Control versus Depression Leave-One-Out Classification; FIG. 22A is the result of the Anxiety versus Depression stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 22B is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Anxiety versus Depression for the case of Stratified Cross-Validation in FIG 22A; FIG. 23A is the result of the Control versus Anxiety stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 23B is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Control versus Anxiety for the case of Stratified Cross-Validation in FIG 23A; FIG. 24A is the result of the Control versus Depression stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 24B is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Control versus Depression for the case of Stratified Cross-Validation in FIG 24A; FIG. 25A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 25B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 26A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 26B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 27A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 27B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 28A is the result of the Bipolar versus ADHD Leave-One-Out Classification; FIG. 28B is the result of the Schizophrenia versus ADHD Leave-One-Out Classification; FIG. 28C is the result of the Control versus Bipolar Leave-One-Out Classification; FIG. 28D is the result of the Control versus ADHD Leave-One-Out Classification; FIG. 29A is the result of the Bipolar versus ADHD stratified Cross-Validation Classification (1000 Repetitions, 50.0% Test size, per repetition); FIG. 29B is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Bipolar versus ADHD for the case of Stratified Cross-Validation Classification in FIG 29A; FIG. 30A is the result of the Schizophrenia versus ADHD stratified Cross-Validation Classification (1000 Repetitions, 50.0% Test size, per repetition); FIG. 30B is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Schizophrenia versus ADHD for the case of Stratified Cross-Validation Classification in FIG 30A; FIG. 31A is the result of the Control versus Bipolar stratified Cross-Validation Classification (1000 Repetitions, 50.0% Test size, per repetition); FIG. 31B is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Control versus Bipolar for the case of Stratified Cross-Validation Classification in FIG 31A; FIG. 32A is the result of the Control versus ADHD stratified Cross-Validation Classification (1000 Repetitions, 50.0% Test size, per repetition); FIG. 32B is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Control versus ADHD for the case of Stratified Cross-Validation Classification in FIG 32A; FIG. 33A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 33B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 34A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 34B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 35A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 35B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 36A is Representational Similarity Matrix (RSM) based on the selected TE; FIG. 36B is Representational Similarity Matrix (RSM) based on all of the TE; FIG. 37A is the distribution of τbestof Bipolar Disorder (BPD) vs Control; FIG. 37B is the distribution of τbestof Schizophrenia vs Control; FIG. 37C is the distribution of τbestof ADHD vs Control; FIG. 37D is the distribution of τbestof CHR-NC vs Control; FIG. 37E is the distribution of τbestof AP vs Control; FIG. 37F is the distribution of τbestof NAP; FIG. 37G is the distribution of τbestof MDD vs Control; FIG. 37H is the distribution of τbestof KET vs Control; FIG. 37I is the distribution of τbestof ECT; FIG. 37J is the distribution of τbestof Anxiety vs Control; FIG. 37K is the distribution of τbestof Epilepsy versus Control; FIG. 38A is the result of the Bipolar Disorder versus Control Leave-One-Out Classification; FIG. 38B is the result of the Bipolar Disorder versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 38C is Representational Similarity Matrix (RSM); FIG. 38D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Bipolar Disorder versus Control for the case of Stratified Cross-Validation in FIG. 38B; FIG. 39A is the result of the Schizophrenia versus Control Leave-One-Out Classification; FIG. 39B is the result of the Schizophrenia versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 39C is Representational Similarity Matrix (RSM); FIG. 39D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Schizophrenia versus Control for the case of Stratified Cross-Validation in FIG. 39B; FIG. 40A is the result of the ADHD versus Control Leave-One-Out Classification; FIG. 40B is the result of the ADHD versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 40C is Representational Similarity Matrix (RSM); FIG. 40D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for ADHD versus Control for the case of Stratified Cross-Validation in FIG. 40B; FIG. 41A is the result of the CHR-NC versus Control Leave-One-Out Classification; FIG. 41B is the result of the CHR-NC versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 41C is Representational Similarity Matrix (RSM); FIG. 41D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for CHR-NC versus Control for the case of Stratified Cross-Validation in FIG. 41B; FIG. 42A is the result of the AP versus Control Leave-One-Out Classification; FIG. 42B is the result of the AP versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 42C is Representational Similarity Matrix (RSM); FIG. 42D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for AP versus Control for the case of Stratified Cross-Validation in FIG. 42B; FIG. 43A is the result of the NAP versus Control Leave-One-Out Classification; FIG. 43B is the result of the NAP versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 43C is Representational Similarity Matrix (RSM); FIG. 43D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for NAP versus Control for the case of Stratified Cross-Validation in FIG. 43B; FIG. 44A is the result of the AP versus NAP Leave-One-Out Classification; FIG. 44B is the result of the AP versus NAP stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 44C is Representational Similarity Matrix (RSM); FIG. 44D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for AP versus NAP for the case of Stratified Cross-Validation in FIG. 44B; FIG. 45A is the result of the MDD versus Control Leave-One-Out Classification; FIG. 45B is the result of the MDD versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 45C is Representational Similarity Matrix (RSM); FIG. 45D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for MDD versus Control for the case of Stratified Cross-Validation in FIG. 45B; FIG. 46A is the result of the KET versus Control Leave-One-Out Classification; FIG. 46B is the result of the KET versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 46C is Representational Similarity Matrix (RSM); FIG. 46D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for KET versus Control for the case of Stratified Cross-Validation in FIG. 46B; FIG. 47A is the result of the ECT versus Control Leave-One-Out Classification; FIG. 47B is the result of the ECT versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 47C is Representational Similarity Matrix (RSM); FIG. 47D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for ECT versus Control for the case of Stratified Cross-Validation in FIG. 47B; FIG. 48A is the result of the KET versus ECT Leave-One-Out Classification; FIG. 48B is the result of the KET versus ECT stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 48C is Representational Similarity Matrix (RSM); FIG. 48D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for KET versus ECT for the case of Stratified Cross-Validation in FIG. 48B; FIG. 49A is the result of the MDD versus KET Leave-One-Out Classification; FIG. 49B is the result of the MDD versus KET stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 49C is Representational Similarity Matrix (RSM); FIG. 49D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for MDD versus KET for the case of Stratified Cross-Validation in FIG. 49B; FIG. 50A is the result of the MDD versus ECT Leave-One-Out Classification; FIG. 50B is the result of the MDD versus ECT stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 50C is Representational Similarity Matrix (RSM); FIG. 50D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for MDD versus ECT for the case of Stratified Cross-Validation in FIG. 50B; FIG. 51A is the result of the Anxiety versus Control Leave-One-Out Classification; FIG. 51B is the result of the Anxiety versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 51C is Representational Similarity Matrix (RSM); FIG. 51D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Anxiety versus Control for the case of Stratified Cross-Validation in FIG. 51B; FIG. 52A is the result of the Epilepsy versus Control Leave-One-Out Classification; FIG. 52B is the result of the Epilepsy versus Control stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 52C is Representational Similarity Matrix (RSM); FIG. 52D is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for Epilepsy versus Control for the case of Stratified Cross-Validation in FIG. 52B; FIG. 53A is the result of the Medicated ASD versus Control Leave-One-Out Classification; FIG. 53B is the result of the Not-Medicated ASD versus Control Leave-One-Out Classification; FIG. 53C is the result of the Not-Medicated ASD versus Medicated ASD Leave-One-Out Classification; FIG. 54A is the result of the Medicated ASD versus Control Leave-One-Out Classification; FIG. 54B is the result of the Not-Medicated ASD versus Control Leave-One-Out Classification; FIG. 54C is the result of the Not-Medicated ASD versus Medicated ASD Leave-One-Out Classification; FIG. 55A is the result of the Control versus Alzheimer’s Disease(AD) Leave-One-Out Classification; FIG. 55B is the result of the Control versus Alzheimer’s Disease(AD) stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 55C is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for the Control versus Alzheimer’s Disease(AD) for the case of Stratified Cross-Validation in FIG. 55B. FIG. 56A is the result of the Parkinson’s Disease (PD) versus PD Prodrome Leave-One-Out Classification; FIG. 56B is the result of the Parkinson’s Disease (PD) versus PD Prodrome stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 56C is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for the Parkinson’s Disease (PD) versus PD Prodrome for the case of Stratified Cross-Validation in FIG. 56B. FIG. 57 is the distribution of τbestof Control versus Alzheimer’s Disease(AD); FIG. 58A is the result of the Control versus Alzheimer’s Disease(AD) Leave-One-Out Classification; FIG. 58B is the result of the Control versus Alzheimer’s Disease(AD) stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 58C is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for the Control versus Alzheimer’s Disease(AD) for the case of Stratified Cross-Validation in FIG. 58B. FIG. 59 is the distribution of τbestof Parkinson’s Disease (PD) versus PD Prodrome; FIG. 60A is the result of the Parkinson’s Disease (PD) versus PD Prodrome Leave-One-Out Classification; FIG. 60B is the result of the Parkinson’s Disease (PD) versus PD Prodrome stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 60C is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for the Parkinson’s Disease (PD) versus PD Prodrome for the case of Stratified Cross-Validation in FIG. 58B. FIG. 61A is the result of the Control versus Alzheimer’s Disease(AD) Leave-One-Out Classification; FIG. 61B is the result of the Control versus Alzheimer’s Disease(AD) stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 61C is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for the Control versus Alzheimer’s Disease(AD) for the case of Stratified Cross-Validation in FIG. 61B. FIG. 62A is the result of the Parkinson’s Disease (PD) versus PD Prodrome Leave-One-Out Classification; FIG. 62B is the result of the Parkinson’s Disease (PD) versus PD Prodrome stratified Cross-Validation Classification (1000 repetitions, 50.0% Test size, per repetition); FIG. 62C is Receiver Operating Characteristic Curve (ROC) and Area Under the Curve (AUC) for the Parkinson’s Disease (PD) versus PD Prodrome for the case of Stratified Cross-Validation in FIG. 62B. DETAILED DESCRIPTION

[0030] Hereinafter, an embodiment of the present disclosure will be described.

[0031] (Outline of Embodiment) First, an outline of technology according to an embodiment of the present disclosure will be described and details thereof will be described later. The technology according to an embodiment of the present disclosure is to be executed by one or more information processing apparatus 10. First, a plurality of region of interest (hereinafter referred to as "ROI") pairs are selected. Each ROI pair includes a source region of a brain of a subject and a target region of the brain. Then, a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (hereinafter referred to as "fMRI") data pertaining to each ROI pair is calculated. Here, the measure of directed flow of information is a measure of the flow of information that starts at the ROI pertaining to the source region (hereinafter referred to as "source ROI") and ends at the ROI pertaining to the target region (hereinafter referred to as "target ROI"). The measure of directed flow of information is a measure that captures the statistical precedence between pairs of ROIs, thereby quantifying the directed functional effect of each one on the other. The measure of directed flow of information may contain information about a future observation of the target ROI, obtained from the simultaneous consideration of the past states of both the source ROI and the target ROI. Further, the information obtained from the past states of target ROI alone may be subtracted or otherwise accounted for to isolate or focus on the contribution of the source ROI. The measure of directed flow of information for selected ROI pairs is input into a trained model, and based on the trained model, information pertaining to the mental disorder of the subject is output. The mental disorder includes schizophrenia, bipolar disorder, attention deficit / hyperactivity disorder, depression, drug-resistant depression, anxiety, early psychosis, autism spectrum disorder, epilepsy, Alzheimer's, or dementia.

[0032] According to the present embodiment the measure of directed flow of information for each ROI pair is calculated based on the fMRI data pertaining to each ROI pair, and information pertaining to the mental disorder of the subject is output by the trained model. Therefore, the technology related to providing information on mental disorders of the subject is improved in that information on mental disorders based on the measure of directed flow of information for each ROI pair is obtained.

[0033] As used herein, the term “mental disorder” means any condition affecting the brain or central nervous system that impacts thought, mood, perception, cognition, behavior, or emotional regulation, including but not limited to psychiatric, neurological, neurocognitive, and brain-based disorders, whether congenital, acquired, traumatic, infectious, or degenerative in origin. The term “mental disorder” includes brain-based diseases and brain-based disorders.

[0034] (Configuration of Information Processing Apparatus) As illustrated in FIG. 1, the information processing apparatus 10 includes a controller 11, a memory 12, an input interface 13, a display 14, and a communication interface 15.

[0035] The controller 11 includes at least one processor, at least one dedicated circuit, or a combination thereof. The processor is a general purpose processor, such as a central processing unit (CPU) or a graphics processing unit (GPU), or a dedicated processor specialized for particular processing. The dedicated circuit is, for example, a field-programmable gate array (FPGA) or an application specific integrated circuit (ASIC). The controller 11 executes processes related to operations of the information processing apparatus 10 while controlling components of the information processing apparatus 10.

[0036] The memory 12 includes at least one semiconductor memory, at least one magnetic memory, at least one optical memory, or a combination of at least two of these. The semiconductor memory is, for example, random access memory (RAM) or read only memory (ROM). The RAM is, for example, static random access memory (SRAM) or dynamic random access memory (DRAM). The ROM is, for example, electrically erasable programmable read only memory (EEPROM). The memory 12 functions as, for example, a main memory, an auxiliary memory, or a cache memory. The memory 12 stores data to be used in the operations of the information processing apparatus 10 and data obtained by the operations of the information processing apparatus 10.

[0037] The input interface 13 includes at least one interface for input. The interface for input is, for example, a physical key, a capacitive key, a pointing device, or a touch screen integrally provided with a display. The interface for input may be, for example, a sound sensor that accepts audio input, a camera that accepts gesture input, or the like. The input interface 13 accepts an operation for inputting data to be used for the operations of the information processing apparatus 10. The input interface 13 may be connected to the information processing apparatus 10 as an external input device, instead of being included in the information processing apparatus 10. As the connection method, for example, any method such as Universal Serial Bus (USB), High-Definition Multimedia Interface (HDMI) (HDMI is a registered trademark in Japan, other countries, or both), or Bluetooth (Registered Trademark) can be used.

[0038] The display 14 includes at least one interface for display output. The interface for display output is, for example, a display for outputting information in the form of an image, or the like. The display is, for example, a liquid crystal display (LCD) or an organic electro luminescent (EL) display. The display 14 displays and outputs data obtained by the operations of the information processing apparatus 10. The display 14 may be connected to the information processing apparatus 10 as an external output device, instead of being included in the information processing apparatus 10. As the connection method, any technology such as USB, HDMI (HDMI is a registered trademark in Japan, other countries, or both), or Bluetooth (Bluetooth is a registered trademark in Japan, other countries, or both) can be used. The display 14 may be located in a position visible from the driver's seat of the vehicle 1.

[0039] The communication interface 15 includes at least one interface for external communication. The interface for communication may be either a wired or wireless communication interface. For wired communication, the interface for communication is, for example, a Local Area Network (LAN) interface or Universal Serial Bus (USB). For wireless communication, the interface for communication is, for example, an interface compliant with a mobile communication standard such as a Long Term Evolution (LTE), the 4th generation (4G) standard, or the 5th generation (5G) standard, or an interface compliant with a short-range wireless communication standard such as Bluetooth (Bluetooth is a registered trademark in Japan, other countries, or both). The communication interface 15 receives data to be used for the operations of the information processing apparatus 10, and transmits data obtained by the operations of the information processing apparatus 10.

[0040] The functions of the information processing apparatus 10 may be implemented by a processor corresponding to the controller 11 executing a program according to the present embodiment. That is, the functions of the information processing apparatus 10 are realized by software. The program causes a computer to execute the operations of the information processing apparatus 10, thereby causing the computer to function as the information processing apparatus 10. That is, the computer executes the operations of the information processing apparatus 10 in accordance with the program to thereby function as the information processing apparatus 10.

[0041] In the present embodiment, the program can be recorded on a computer readable recording medium. The computer readable recording medium includes a non-transitory computer readable medium and is, for example, a magnetic recording apparatus, an optical disc, a magneto-optical recording medium, or a semiconductor memory. The program is distributed, for example, by selling, transferring, or lending a portable recording medium such as a digital versatile disc (DVD) or a compact disc read only memory (CD-ROM) on which the program is recorded. The program may also be distributed by storing the program in a storage of an external server and transmitting the program from the external server to another computer. The program may be provided as a program product.

[0042] Some or all of the functions of the information processing apparatus 10 may be realized by a dedicated circuit corresponding to the controller 11. That is, some or all of the functions of the information processing apparatus 10 may be realized by hardware.

[0043] (Operations of Information Processing Apparatus) With reference to FIG. 2, the operations of the information processing apparatus 10 according to the present embodiment are now described. FIG. 2 is a flowchart illustrating an example of a method executed by the information processing apparatus 10 according to the present embodiment.

[0044] Step S10: the controller 11 of the information processing apparatus 10 selects a plurality of ROI pairs.

[0045] Any method can be employed to select a plurality of ROI pairs. For example, the memory 12 may store information pertaining to a plurality of ROI pairs that may be associated with information pertaining to mental disorders to be output in step S40 described below. In this case, the controller 11 may select the plurality of ROI pairs based on the information pertaining to the plurality of ROI pairs stored in the memory 12. Alternatively, control unit 11 may select multiple ROI pairs based on user operation to input interface 13.

[0046] Step S20: the controller 11 calculates the measure of directed flow of information for each ROI pair based on fMRI data pertaining to each ROI pair.

[0047] Any method can be employed to calculate the measure of directed flow of information. As described above, the measure of directed flow of information is a measure of the flow of information that starts at source ROI and ends at target ROI. The measure of directed flow of information typically represents an asymmetric quantification of the relationship between two or more variables. Examples of measures of directed flow of information include transfer entropy (hereinafter, referred to as "TE"), conditional entropy, graph- and network-based methodologies (e.g., random-walker graph Laplacian, Markov processes and networks, Bayesian network, etc.), convergent cross-mapping (CCM), Granger causality (hereinafter, referred to as "GC"), partial directed coherence, dynamic causal modeling and variation of predictive coding. In one embodiment, the measure of directed flow of information is calculated using Bayesian formalism, such as dynamic causal modeling. In other embodiments, the measure of directed flow of information may be calculated by assuming linearity for predictive relationships between a ROI pair. In this case, the measure of directed flow of information may be GC. In still other embodiments, nonlinear relationship is assumed for predictive relationships between a ROI pair. The measure of directed flow of information may be nonparametric. Such nonparametric measures may have many advantages, such as robustness due to the absence of assumptions about specific probability distributions, applicability to small sample sizes, the ability to capture complex relationships, and robustness to outliers. In other embodiments, measures that allow detection of nonlinear dependencies between time series may be used for the measure of directed flow of information. Those measures that allow detection of nonlinear dependencies may have many advantages, including the ability to capture complex relationships and threshold effects, resilience to misspecification and improved predictive power.

[0048] For example, the measure of directed flow of information may be defined by TE. TE is a measure that allows detection of linear and nonlinear (in particular, nonlinear) dependencies between time series. It can be expressed in terms of both, parametric as well as nonparametric formulation. It is this versatility that set its use apart from such measures as GC. For instance, whereas in the presence of guaranteed Guassianity of the processes under consideration, TE becomes equivalent to GC, its computation is not required to be limited / confined to such an assumption. In other words, TE can be expressed so as to capture the presence of predictive power of one process over another without fulfilling the linearity (i.e., Gaussianity) of these processes. The definition of the measure of the flow of information, here represented in terms of TE (i.e., Transfer Entropy), may be set arbitrarily. For example, TE may be defined by the following equation (1).

[0049] where τ is the time delay for transferring information between regions in each ROI pair, X is random variable pertaining to source ROI (here, fMRI data of source ROI), Y is random variable pertaining to target ROI (here, fMRI data of target ROI), Xtis the present state of random variable X, Xt-τis the state of random variable X, t-τ steps before, MI(X;Y|Z) = H(X|Z)- H(X|Y,Z) is the mutual information (MI) between X and Y conditioned on Z, and H(X|Z) and H(X|Y,Z) are the entropy (H) of X conditioned on Z and on Y&Z, respectively. Therefore, equation (1) can be expressed as in the following equation (2)

[0050]

[0051] Entropy of a random variable is the average level of "information", or "uncertainty" inherent to the variable's possible outcomes. For example, in discrete domain, Entropy is computed based on the following equation (3).

[0052] where "b" is the base of the logarithm (e.g., b=e means that Entropy is computed in unit of natural logarithm). This can be quantified in continuous domain using kernel- as well as nearest-neighbor based approaches.

[0053] In information theory, entropy of A conditioned on B (H(A|B)) is a conditional entropy which quantifies the amount of information needed to describe the outcome of a random variable A given that the value of another random variable B is known.

[0054] As shown here in equation (1), TE is calculated based on τ. In other words, the controller 11 calculate the TE for each ROI pair based on τ. τ determines the time-window within which the source process is effective in explaining the dynamics of the target process, i.e., the range of the asymmetric influence that governs the (directed) interplay between the source and destination processes. Any method can be employed to calculate τ. For example, τ can be calculated based on an average value of a best time delay (hereinafter referred to as "τbest") at which the transfer entropy for each ROI pair is a maximum (hereinafter referred to as "TEbest").

[0055] Step S30: the controller 11 inputs the measure of directed flow of information (here, TE) for selected ROI pairs into a trained model.

[0056] A learning model is created by machine learning using a machine learning algorithm. The learning model may, for example, be a machine learning model constructed based on a decision tree. Examples of machine learning models constructed based on decision trees include, but are not limited to, Random Forest, Light GBM and XGBoost. Alternatively, the learning model may be a model generated based on a machine learning algorithm such as a Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), or deep learning. The trained model is a model to cause the information processing apparatus 10 to function to output an objective variable pertaining to a subject's mental disorder based on explanatory variables pertaining to the subject's data. The explanatory variables include TE of each ROI pair. The objective variables include information pertaining to the mental disorder of the subject.

[0057] Step S40: the controller 11 outputs information pertaining to the mental disorder of the subject based on the trained model. In other words, the controller 11 outputs information pertaining to the mental disorder of the subject based on the measure of directed flow of information (here, TE) for selected ROI pairs and the trained model. Information pertaining to the mental disorder of the subject includes information that supports the diagnosis, prognosis, monitoring, and response to treatment of mental disorder by physicians and others.

[0058] As described above, according to the information processing apparatus 10 according to the present embodiment, the controller 11 of the information processing apparatus 10 selects a plurality of ROI pairs. Then the controller 11 calculates the measure of directed flow of information for each ROI pair based on fMRI data pertaining to each ROI pair. The controller 11 inputs the measure of directed flow of information for selected ROI pairs into the trained model, and based on the trained model, information pertaining to the mental disorder of the subject is output.

[0059] According to the present embodiment the measure of directed flow of information for each ROI pair is calculated based on the fMRI data pertaining to each ROI pair, and information pertaining to the mental disorder of the subject is output by the trained model. Therefore, the technology related to providing information on mental disorders of the subject is improved in that information on mental disorders based on the measure of directed flow of information for each ROI pair is obtained.

[0060] The technology according to the embodiment of the present disclosure is not limited to those executed by information processing equipment 10. For example, the technology of the embodiment can be applied as a method for diagnosing, prognosing, or monitoring a response to treatment of mental disorders by physicians and others. With reference to FIG. 3, an example of a method for the embodiment is described. FIG. 3 is a flowchart showing an example of the method of this embodiment.

[0061] Step S110: First, a plurality of ROI pairs are selected.

[0062] Any method can be employed to select a plurality of ROI pairs. For example, a plurality of ROI pairs may be selected that can be related to information pertaining to the mental disorder of the subject to be presented in step S130 described below.

[0063] Step S120: Based on the fMRI data for each ROI pair, the measure of directed flow of information (here, TE) for each ROI pair is calculated. Any method can be used to calculate TE. For example, TE may be calculated by the above formula (1).

[0064] Step S130: Based on the measure of directed flow of information (here, TE) for selected ROI pairs, information pertaining to the mental disorder of the subject is presented. Such information includes information that supports diagnosis, prognosis, monitoring, and response to treatment of mental disorders by physicians and others.

[0065] Presenting the information pertaining to the mental disorder of the subject may include classifying the subject by a mathematical model and presenting the information pertaining to the mental disorder of the subject. The mathematical model may be a K-Nearest-Neighbor classifier, logistic regression classifier, support vector machine classifier or Naive Bayes classifier. When the K-Nearest-Neighbor classifier is applied, cosine similarity, correlation similarity, or Euclidean similarity may be used as a distance function.

[0066] (Example) The following is an example of the embodiment of the present disclosure. Each of the data used in this example, the preprocessing, the calculation of τ, and the analysis of information pertaining to the mental disorder output based on TE are described below, respectively.

[0067] (fMRI Data) In this example, the inventors used the following datasets. (i) National Institute of Health (NIMH) Human Connectome Projects 1. HCP-Early Psychosis (HCP-EP): This dataset includes 155 subjects that covers three phenotypes: Non-Affective Psychosis (78 subjects), Affective Psychosis (25 subjects), and Control (52 subjects). 2. Boston Adolescent Neuroimaging of Depression & Anxiety (BANDA): It includes 162 subjects, covering three phenotypes: Anxiety (59 subjects), Depression (53 subjects), and Control (50 subjects). (ii) UCLA Consortium for Neuropsychiatric Phenomics LA5c Study (UCLA): It includes 209 subjects, comprising 4 phenotypes: Bipolar Disorder (37 subjects), Schizophrenia (33 subjects), Attention Deficit Hyperactivity Disorder (ADHD, 36 subjects), and Control (103 subjects).

[0068] “Affective disorder” is also known as “mood disorder” [1]. The major examples include mania, depression, and seasonal affective disorder. Although it is also known under the rubric of “Affective Psychosis” (i.e., its maniac manifestation), it differs from disorders such as schizophrenia that is classified under “Personality Disorders” in the Diagnostic and Statistical Manual of Mental Disorders (DSM-5). The National Institutes of Health (NIH) defines it as “The manifestation of psychotic symptoms, such as hallucinations or delusions, in the presence of a mood disorder” [2].

[0069] In the DSM-5, non-affective psychoses are reported under the “schizophrenia spectrum disorders” and include schizophrenia, other psychotic disorders, and schizotypal personality disorder that find a common denominator in the presence of delusions, hallucinations, disorganized thinking, disorganized behavior, and negative symptoms (anhedonia, loss of interest / appetite, etc.) [3, 4].

[0070] The numbers that the inventor report in this example correspond to the finalized list of participants. In other words, they correspond to the number of participants that passed all the preprocessing and postprocessing steps and not the actual number of participants that participated in the original neuroimaging studies. The inventors follow the conventions that are practiced within the resting-state functional neuroimaging (rs-fMRI) community [5].

[0071] (Paired Phenotypes) The inventors report the results for the following paired phenotypes. (i) UCLA 1. Control versus Schizophrenia 2. Bipolar versus Schizophrenia (ii) HCP-EP 1. Control versus Affective Psychosis 2. Control versus Non-Affective Psychosis 3. Affective Psychosis versus Non-Affective Psychosis (iii) BANDA 1. Control versus Anxiety 2. Control versus Depression 3. Anxiety versus Depression (iv) UCLA 1. Control versus Bipolar 2. Control versus ADHD 3. Bipolar versus ADHD 4. Schizophrenia versus ADHD

[0072] (Parcellation) The inventors parcellated the individuals' brain data into 232 cortical and subcortical regions of interests (ROIs). To obtain 200 cortical ROIs, the inventors used the Schaefer et al. [6] scale-200 version of the local-global functional parcellation. The inventors then augmented these cortical ROIs with Tian et al. [7] 32 subcortical ROIs.

[0073] (Calculation of the TE) To capture the statistical precedence between pairs of ROIs, thereby quantifying the directed functional effect of each one on the other, TE between the pairs of ROIs were calculated based on the above equation (1).

[0074] (Calculation of τ) To compute TE, an estimate of the time delay τ between the paired Source and Target ROIs are required. The inventors computed τ for a certain paired phenotype (for example, Control and Schizophrenic samples) as follows. First, the inventors found τ, per Control and Schizophrenia samples, per individual, per pair of ROIs, such that it maximized from Source to Target transfer of information. This was performed through a brute-force strategy [8] in which τ was incremented one data-point at the time. As a result, this increment step size becomes dependent on the fMRI device Repetition Time (TR) i.e., its sampling rate because TR determines the distance between subsequent recorded data points. For example, when the inventors computed τ based on the dataset of UCLA with TR = 2.0 sec, τ’s increment in a 1-data-point corresponded to ≒ 2-second increment steps. The inventors carried out this increment within the range τ ∈ [1, ...., 15] data points i.e., approximately 2 through 30 seconds. The inventors calculated the value of τ (i.e. τbest), per individual, per sample, per paired ROIs, that resulted in maximum TE (i.e. TEbest) for a given paired ROI. As a result, 53,824 τbestvalues (i.e. 2322τbestvalues), per individual, per Control / Schizophrenia samples were obtained.

[0075] The inventors then bootstrapped (10,000 random sampling with replacement at 95% confidence interval (CI95%)) these τbestvalues within each Control and Schizophrenia samples (i.e., Control and Schizophrenia samples, separately), thereby estimating the μτ, per sample.

[0076] FIG. 4A and FIG. 4B show the distribution of τbestof Control and Schizophrenia, respectively. According to the data of FIG. 4A and FIG. 4B, mean (M), Standard deviation (SD), median (Mdn) the range of CI95%are the following: Control: M = 3.8906 SD = 0.0005 Mdn = 3.8906 CI95%= [3.8895, 3.8916] Schizophrenia: M = 3.8931 SD = 0.0008 Mdn = 3.8931 CI95%= [3.8916, 3.8947] Both groups’ bootstrapped estimation resulted in μτ≒4 i.e., 8 seconds. The inventors set all individuals’ τ, per group, per ROI pairs, to μτ(=4) and recomputed their respective all TE. τbestand μτfor other phenotypes are calculated in the similar manner. As mentioned above, it is noted that the increment 1-data-point step size depends on the TR. The datasets of BANDA and HCP-EP adopted TR = 0.8 sec, so that τ for BANDA and HCP-EP were incremented in a 1-data-point increment i.e., ≒ 0.8-second increment steps within the range τ ∈ [1, ...., 30] data points i.e., approximately 0.8 through 24 seconds.

[0077] (Selecting a plurality of ROI) As described above, in one embodiment of the present disclosure, a plurality of ROI pairs are selected that can be related to information pertaining to the mental disorder of the subject being presented. In other words, ROI pairs whose importance exceeds a predetermined level are selected. Here, the method of selecting ROI pairs in the case where the target mental disorder is schizophrenia is described.

[0078] After computing TE for all possible pairs (i.e., 2322= 53,824 TE values) using μτ(=4), the inventors determined the significant difference between two phenotypes (e.g., Control vs. Schizophrenia), per ROI pair, through following steps.

[0079] The inventors performed bootstrap (i.e., 10,000 repetitions with replacement of samples) test of significant mean-difference (i.e., μPhenotype1- μPhenotype2) at 95% confidence interval (i.e., p < 0.05) between the two phenotypes. Here, Phenotype 1 is Control and Phenotype 2 is Schizophrenia. If this difference is significant, then the inventors check its computed p-value against the Bonferroni-corrected p-value (i.e., 0.05 / 232 ≒ 0.00022). If the computed p-value is less than Bonferroni-corrected p-value, then the inventors considered the mean-difference between the given ROI pair significant and selected it as a pair-of-interest. Otherwise, the inventors consider their mean-difference non-significant and discard them (i.e., even if their calculated bootstrapped p-value is less than 0.05, the inventors did not select them as pair-of-interest).

[0080] For the example of Control and Schizophrenic Patients, the inventors found 60 statistically significant TE values. For the example of Schizophrenia and Bipolar Patients, the inventors found 52 statistically significant TE values. It is apparent that this a substantial reduction: only a small (i.e., << 53,824) portion of computed pairwise TE values in the above two cases are required for the purpose of diagnosis of these patients from Control.

[0081] (Selected Pairs of ROI) For example, out of total of 232 cortical and subcortical regions, the inventors found some source ROIs (FIG. 5A, FIG. 5B) that contributed to the significant differences between Control and Schizophrenia brain’s inter-regional information out-flow at rest. These ROIs were distributed among a number of brain cortical and subcortical regions and networks. Some or the ROIs that contributed to the significant differences between Schizophrenia and Control are described below.

[0082] Subcortical ROIs included the following: (1) the basal ganglia (6 ROIs; left hemisphere (LH): anterior and posterior caudate (aCAU & pCAU), anterior putamen (aPUT), and ante-rior globus pallidus (aGP), right hemisphere (RH): nucleus accumbens core and shell (NAc-core & NAc-shell)), (2) the thalamus (4 ROIs; LH: ventral posterior (THA-VP), dorsal anterior (THA-DA), ventral anterior (THA-VA), RH: ventral posterior (THA-VP)), and (3) the amygdala (1 ROI, LH: lateral (lAMY)).

[0083] Cortical ROIs expanded over a number of cortical networks and regions. Cortical networks included the following: (1) the dorsal attention network (5 ROIs; LH: frontal eye fields (DorsAttnFEF1 & DorsAttnFEF2) and posterior (DorsAttnPost3 & DorsAttnPost7& DorsAttnPost8)), (2) the salient ventral attention network (4 ROIs; LH: medial (SalVentAttnMed2) and parietal operculum (SalVentAttnParOper3), RH: medial (Sal-VentAttnMed1) and frontal operculum insula (SalVentAttnFrOperIns2)), (3) the default mode network (DMN) (7 ROIs; LH: prefrontal cortex (PFC) (DefaultPFC1 & DefaultPFC8), temporal (DefaultTemp2), parietal (DefaultPar2), precuneus posterior cingulate cortex (PCC) (DefaultpCunPCC3), RH: temporal (DefaultTemp5) and PCC (DefaultpCunPCC1)), and (4) the front-parietal network (FPN) (4 ROIs; LH: lateral PFC (ContPFCl1 & ContPFCl5) and cingulate cortex (ContCing1), RH: medial posterior PFC (ContPFCmp1)).

[0084] Moreover, there were three cortical regions: (1) the limbic system (4 ROIs; LH: orbital frontal cortex (OFC) (LimbicOFC1) and temporal pole (LimbicTempPole2 & LimbicTempPole4), RH: temporal (LimbicTempPole3)) (2) the visual cortex (6 ROIs; LH: Vis1 & Vis3 & Vis13 & Vis14, RH: Vis5 & Vis12), and (3) the somatomotor cortex (5 ROIs; RH: SomMot1 & SomMot5 & SomMot11 & SomMot17 & SomMot19).

[0085] Although the source ROIs were present in both, the left and the right hemispheres, their distribution was left-dominated (FIG. 5A and FIG. 5B, 30 ROIs in LH versus 16 of them in RH), indicating a left-hemispheric tendency in the source ROIs’ differences between the Control and the Schizophrenia samples.

[0086] FIG. 6A and 6B show the target ROIs that contributed to the difference between the Control and Schizophrenia samples. The number of target ROIs was considerably smaller than source ROIs (≒ half of the source ROIs in FIG. 5A and FIG. 5B). Similar to the case of source ROIs, the target ROIs were also distributed among a number of cortical and subcortical regions.

[0087] The subcortical ROIs were in (1) the basal ganglia (4 ROIs; LH: posterior caudate (pCAU) and nucleus accumbens shell (NAc-shell), RH: anterior globus pallidus (aGP) and posterior putamen (pPUT)) and (2) the amygdala (1 ROI; LH: medial (mAMY)).

[0088] In the case of cortical regions, they corresponded to (1) the dorsal attention network (3 ROIs; LH: frontal eye field (DorsAttnFEF1), RH: posterior (DorsAt-tnPost1 & DorsAttnPost10), (2) the salient ventral attention network (3 ROIs; LH: frontal operculum insula (SalVentAttnFrOperIns3) and parietal operculum (SalVen-tAttnParOper2), RH: frontal operculum insula (SalVentAttnFrOperIns4)), (3) the DMN (2 ROIs; LH: Parahippocampal Cortex (DefaultPHC1), RH: temporal (Default-Temp1)), (4) the FPN (5 ROIs; LH: Parietal (ContPar1), RH: lateral PFC (ContPFCl2 & ContPFCl4), medial posterior PFC (ContPFCmp1), and cingulate cortex (ContCing1)), (5) the limbic system (2 ROIs; LH: temporal Pole LimbicTempPole2, RH: Orbital frontal cortex (LimbicOFC3)), and (6) the somatomotor cortices (2 ROIs; SomMot2 in both the left and the right hemispheres). Contrary to the case of source ROIs, these regions did not exhibit any hemispheric tendency (FIG. 6A and FIG. 6B): 10 of them were in LH and 12 of them were in RH.

[0089] Four ROIs were present in both, the Source as well as the Target. In the left hemisphere, these ROIs were from (1) the basal ganglia (pCAU), (2) the dorsal attention network (DorsAttnFEF1), and (3) the limbic system (TempPole2Pole2). In the right hemisphere, it was a single ROI from the FPN (ContPFCmp1).

[0090] FIG. 7A and FIG 7B shows the inter-regional directed functional flow of information that significantly differed between Control and Schizophrenia samples. The unique Source ROIs and Target ROIs constituted some significantly different directed functional connectivity (out of total of 2322= 53,824 pairwise directed functional connectivity).

[0091] (Analysis) The inventors report the following two types of classification results. (i) Leave-One-Out Classification: where the inventors isolate every individual (whether from Control or Patient phenotypes), train a model on the remainder of sample, and then test the model’s performance to predict the isolated subject’s phenotype. The inventors report the accuracy of our model in terms of “its accuracy averaged over all isolated cases.” For instance, in the case of BANDA’s Anxiety versus Control (59 + 50 = 109 subjects), the inventors perform 109 trainings, each of which is followed by its corresponding prediction test. This results in 109 predictions. The final accuracy of the model is then its accuracy averaged over these 109 cases. (ii) Stratified Cross-Validation: where the inventors divide the sample into Train and Test sets. 1. To prevent biased results that can be incurred due to the sample imbalance (i.e., unequal number of participants in two phenotype groups, resulting in the model’s performance to lean toward the dominant subsample), the inventors employ “stratification” strategy. Concretely, regardless of the difference in the two subsamples, the inventors ensure that equal proportion of each subsample is present in both Train and Test sets. The inventors divide the entire sample equally between the Train and Test sets i.e., 50.0% (with equal proportion of the two (imbalanced) phenotypes) is assigned to the Train and 50.0% to the Test sets. 2. The inventors carry out the above step for 1000 repetitions. For each repetition, the inventors randomly assign half i.e., 50.0% of each of two phenotypes to the Train set and the remainder 50.0% to the Test set. Next, the inventors train our model using the Train set. Subsequently, the inventors evaluate our trained model’s accuracy on the other 50.0% in the Test set. a. It is crucial to note that this a highly unconventional and unorthodox splitting of the data for testing the model’s performance. Specifically, literature adapts a 5.0%-30.0% division that is (for the most part) unstratified. b. This also explains the difference that the inventors observe between the results of Leave-One-Out Classification versus Stratified Cross-Validation. Concretely, whereas the former uses N-1 (where N is the total number of participants in a given paired phenotypes) samples for training the model and uses this model to predict the single isolated subject, the latter has only 50.0% (with equal proportion) of the entire sample available for its proper training and recognition of the difference between two phenotypes. It then uses this model to predict the remaining 50.0% of sample in the Test set (i.e., an equal number of unknown versus known cases). 3. The inventors report the accuracy of the model as “its accuracy averaged over these 1000 repetitions.” 4. The inventors visualize the model’s performance using Receiver Operating Characteristic (ROC) curve and its corresponding Area Under the ROC curve (AUC) plot.

[0092] It is noted that “Leave-One-Out” classification is more important than the “Stratified Cross-Validation” results. This is because the former is the real-life scenario: patients are diagnosed on individual basis not after a waiting list for diagnosis reaches a certain threshold. However, the latter plays a crucial role from an analytical perspective: it ensures that the results are stable within the population under investigation and that these results do not deviate (statistically) substantially as the sample size changes.

[0093] (Results of Control versus Schizophrenia) FIG. 8A shows the result of the Control versus Schizophrenia Leave-One-Out Classification by KNN (K = 7) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 60 TE that showed significant differences between Control and Schizophrenia samples.

[0094] FIG. 8A verifies that the 60 TE between Schizophrenia and Control exhibited a high specificity in distinguishing the two samples apart. Specifically, these 60 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 93% accuracy) and the Schizophrenia (i.e., 97% accuracy) samples.

[0095] FIG. 8B shows the result of the Schizophrenia versus Control stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 7) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 60 TE that showed significant differences between Control and Schizophrenia samples.

[0096] FIG. 8B verifies that the 60 TE between Schizophrenia and Control exhibited a high specificity in distinguishing the two samples apart. Specifically, these 60 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 93% accuracy) and the Schizophrenia (i.e., 91% accuracy) samples.

[0097] FIG. 9A and FIG. 9B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Schizophrenia and the Control samples based on the 60 TE (using “cosine similarity” measure) revealed that (FIG. 9A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 9B) resulted in rather a uniform spatial representation of the two samples.

[0098] FIG. 10 shows the result of the receiver operating characteristic curve (ROC) analysis for Schizophrenia versus Control. As shown in FIG. 10, using 60 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.97 ± 0.01 ≡ 97.00% ± 0.01%).

[0099] (Results of Bipolar versus Schizophrenia) FIG. 11A shows the result of the Bipolar versus Schizophrenia Leave-One-Out Classification by KNN (K = 5) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 52 TE that showed significant differences between Bipolar and Schizophrenia samples.

[0100] FIG. 11A verifies that the 52 TE between Bipolar versus Schizophrenia exhibited a high specificity in distinguishing the two samples apart. Specifically, these 52 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Bipolar (i.e., 100% accuracy) and the Schizophrenia (i.e., 100% accuracy) samples.

[0101] FIG. 11B shows the result of the Bipolar versus Schizophrenia stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 5) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 52 TE that showed significant differences between Bipolar and Schizophrenia samples.

[0102] FIG. 11B verifies that the 52 TE between Bipolar versus Schizophrenia exhibited a high specificity in distinguishing the two samples apart. Specifically, these 52 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Bipolar (i.e., 96% accuracy) and the Schizophrenia (i.e., 100% accuracy) samples.

[0103] FIG. 12A and FIG. 12B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Bipolar and Schizophrenia samples based on the 52 TE (using “cosine similarity” measure) revealed that (FIG. 12A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 12B) resulted in rather a uniform spatial representation of the two samples.

[0104] FIG. 13 shows the result of the receiver operating characteristic curve (ROC) analysis for Bipolar versus Schizophrenia. As shown in FIG. 13, using 52 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.99 ± 0.00 ≡ 99.00%).

[0105] (Results of Affective versus Non-Affective Psychosis, Control versus Affective Psychosis, and Control versus Non-Affective Psychosis) FIG. 14A shows the result of the Affective versus Non-Affective Psychosis Leave-One-Out Classification by KNN (K = 5) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 100 TE that showed significant differences between Affective and Non-Affective Psychosis.

[0106] FIG. 14A verifies that the 100 TE between Affective versus Non-Affective Psychosis exhibited a high specificity in distinguishing the two samples apart. Specifically, these 100 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Affective (i.e., 100% accuracy) and the Non-Affective Psychosis (i.e., 94% accuracy) samples.

[0107] FIG. 14B shows the result of the Control versus Affective Psychosis Leave-One-Out Classification by KNN (K = 3) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 116 TE that showed significant differences between Control and Affective Psychosis.

[0108] FIG. 14B verifies that the 116 TE between Control versus Affective Psychosis exhibited a high specificity in distinguishing the two samples apart. Specifically, these 116 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 96% accuracy) and the Affective Psychosis (i.e., 100% accuracy) samples.

[0109] FIG. 14C shows the result of the Control versus Non-Affective Psychosis Leave-One-Out Classification by KNN (K = 5) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 67 TE that showed significant differences between Control and Non-Affective Psychosis.

[0110] FIG. 14C verifies that the 67 TE between Control versus Non-Affective Psychosis exhibited a high specificity in distinguishing the two samples apart. Specifically, these 67 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 92% accuracy) and the Non-Affective Psychosis (i.e., 86% accuracy) samples.

[0111] FIG. 15A shows the result of the Affective versus Non-Affective Psychosis stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 5) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 100 TE that showed significant differences between Affective and Non-Affective Psychosis samples.

[0112] FIG. 15A verifies that the 100 TE between Affective versus Non-Affective Psychosis exhibited a high specificity in distinguishing the two samples apart. Specifically, these 100 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Affective (i.e., 98% accuracy) and the Non-Affective Psychosis (i.e., 94% accuracy) samples.

[0113] FIG. 15B shows the result of the receiver operating characteristic curve (ROC) analysis. As shown in FIG. 15B, using 100 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.99 ± 0.01 ≡ 99.00% ± 0.01%).

[0114] FIG. 16A shows the result of the Control versus Affective Psychosis stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 3) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 116 TE that showed significant differences between Control and Affective Psychosis samples.

[0115] FIG. 16A verifies that the 116 TE between Control versus Affective Psychosis exhibited a high specificity in distinguishing the two samples apart. Specifically, these 116 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 96% accuracy) and Affective Psychosis (i.e., 95% accuracy) samples.

[0116] FIG. 16B shows the result of the receiver operating characteristic curve (ROC) analysis for Control versus Affective Psychosis. As shown in FIG. 16B, using 116 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.98 ± 0.01 ≡ 98.00% ± 0.01%).

[0117] FIG. 17A shows the result of the Control versus Non-Affective Psychosis stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 5) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 67 TE that showed significant differences between Control and Non-Affective Psychosis samples.

[0118] FIG. 17A verifies that the 67 TE between Control versus Affective Psychosis exhibited a high specificity in distinguishing the two samples apart. Specifically, these 67 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 91% accuracy) and the Non-Affective Psychosis (i.e., 87% accuracy) samples.

[0119] FIG. 17B shows the result of the receiver operating characteristic curve (ROC) analysis for Control versus Non-Affective Psychosis. As shown in FIG. 17B, using 67 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.95 ± 0.02 ≡ 95.00% ± 0.02%).

[0120] FIG. 18A and FIG. 18B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Affective versus Non-Affective Psychosis samples based on the 100 TE (using “cosine similarity” measure) revealed that (FIG. 18A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 18B) resulted in rather a uniform spatial representation of the two samples.

[0121] FIG. 19A and FIG. 19B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Control versus Affective Psychosis samples based on the 116 TE (using “cosine similarity” measure) revealed that (FIG. 19A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 19B) resulted in rather a uniform spatial representation of the two samples.

[0122] FIG. 20A and FIG. 20B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Control versus Non-Affective Psychosis samples based on the 67 TE (using “cosine similarity” measure) revealed that (FIG. 20A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 20B) resulted in rather a uniform spatial representation of the two samples.

[0123] (Results of Anxiety versus Depression, Control versus Anxiety, and Control versus Depression) FIG. 21A shows the result of the Anxiety versus Depression Leave-One-Out Classification by KNN (K = 30) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 27 TE that showed significant differences between Anxiety and Depression.

[0124] FIG. 21A verifies that the 27 TE between Anxiety versus Depression exhibited a high specificity in distinguishing the two samples apart. Specifically, these 27 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Anxiety (i.e., 90% accuracy) and the Depression (i.e., 83% accuracy) samples.

[0125] FIG. 21B shows the result of the Control versus Anxiety Leave-One-Out Classification by KNN (K = 20) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 27 TE that showed significant differences between Control and Anxiety.

[0126] FIG. 21B verifies that the 27 TE between Control versus Anxiety exhibited a high specificity in distinguishing the two samples apart. Specifically, these 27 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 92% accuracy) and the Anxiety (i.e., 93% accuracy) samples.

[0127] FIG. 21C shows the result of the Control versus Depression Leave-One-Out Classification by KNN (K = 9) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 33 TE that showed significant differences between Control and Anxiety.

[0128] FIG. 21C verifies that the 33 TE between Control versus Depression exhibited a high specificity in distinguishing the two samples apart. Specifically, these 33 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 86% accuracy) and the Depression (i.e., 83% accuracy) samples.

[0129] FIG. 22A shows the result of the Anxiety versus Depression stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 30) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 27 TE that showed significant differences between Anxiety and Depression samples.

[0130] FIG. 22A verifies that the 27 TE between Anxiety versus Depression exhibited a high specificity in distinguishing the two samples apart. Specifically, these 27 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Anxiety (i.e., 88% accuracy) and the Depression (i.e., 82% accuracy) samples.

[0131] FIG. 22B shows the result of the receiver operating characteristic curve (ROC) analysis for Anxiety versus Depression. As shown in FIG. 22B, using 27 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.91 ± 0.05 ≡ 91.00% ± 0.05%).

[0132] FIG. 23A shows the result of the Control versus Anxiety stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 20) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 27 TE that showed significant differences between Control and Anxiety samples.

[0133] FIG. 23A verifies that the 27 TE between Control versus Anxiety exhibited a high specificity in distinguishing the two samples apart. Specifically, these 27 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 88% accuracy) and the Anxiety (i.e., 94% accuracy) samples.

[0134] FIG. 23B shows the result of the receiver operating characteristic curve (ROC) analysis for Control versus Anxiety. As shown in FIG. 23B, using 27 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.97 ± 0.01 ≡ 97.00% ± 0.01%).

[0135] FIG. 24A shows the result of the Control versus Depression stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 9) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 33 TE that showed significant differences between Control and Depression samples.

[0136] FIG. 24A verifies that the 33 TE between Control versus Depression exhibited a high specificity in distinguishing the two samples apart. Specifically, these 33 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 88% accuracy) and the Depression (i.e., 82% accuracy) samples.

[0137] FIG. 24B shows the result of the receiver operating characteristic curve (ROC) analysis for Control versus Depression. As shown in FIG. 24B, using 33 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.90 ± 0.04 ≡ 90.00% ± 0.04%).

[0138] FIG. 25A and FIG. 25B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Anxiety versus Depression samples based on the 27 TE (using “cosine similarity” measure) revealed that (FIG. 25A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 25B) resulted in rather a uniform spatial representation of the two samples.

[0139] FIG. 26A and FIG. 26B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Control versus Anxiety samples based on the 27 TE (using “cosine similarity” measure) revealed that (FIG. 26A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 26B) resulted in rather a uniform spatial representation of the two samples.

[0140] FIG. 27A and FIG. 27B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Control versus Depression samples based on the 33 TE (using “cosine similarity” measure) revealed that (FIG. 27A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 27B) resulted in rather a uniform spatial representation of the two samples.

[0141] (Results of Bipolar versus ADHD, Schizophrenia versus ADHD, Control versus Bipolar and Control versus ADHD) FIG. 28A shows the result of the Bipolar versus ADHD Leave-One-Out Classification by KNN (K = 9) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 33 TE that showed significant differences between Bipolar and ADHD.

[0142] FIG. 28A verifies that the 33 TE between Bipolar versus ADHD exhibited a high specificity in distinguishing the two samples apart. Specifically, these 33 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Bipolar (i.e., 95% accuracy) and the ADHD (i.e., 94% accuracy) samples.

[0143] FIG. 28B shows the result of the Schizophrenia versus ADHD Leave-One-Out Classification by KNN (K = 3) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 36 TE that showed significant differences between Schizophrenia and ADHD.

[0144] FIG. 28B verifies that the 36 TE between Schizophrenia versus ADHD exhibited a high specificity in distinguishing the two samples apart. Specifically, these 36 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Schizophrenia (i.e., 95% accuracy) and the ADHD (i.e., 94% accuracy) samples.

[0145] FIG. 28C shows the result of the Control versus Bipolar Leave-One-Out Classification by KNN (K = 5) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 30 TE that showed significant differences between Control and Bipolar.

[0146] FIG. 28C verifies that the 30 TE between Control versus Bipolar exhibited a high specificity in distinguishing the two samples apart. Specifically, these 30 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 92% accuracy) and the Bipolar (i.e., 86% accuracy) samples.

[0147] FIG. 28D shows the result of the Control versus ADHD Leave-One-Out Classification by KNN (K = 7) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 36 TE that showed significant differences between Control and ADHD.

[0148] FIG. 28D verifies that the 36 TE between Control versus ADHD exhibited a high specificity in distinguishing the two samples apart. Specifically, these 36 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 89% accuracy) and the ADHD (i.e., 83% accuracy) samples.

[0149] FIG. 29A shows the result of the Bipolar versus ADHD stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 9) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 33 TE that showed significant differences between Bipolar and ADHD samples.

[0150] FIG. 29A verifies that the 33 TE between Bipolar versus ADHD exhibited a high specificity in distinguishing the two samples apart. Specifically, these 33 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Bipolar (i.e., 93% accuracy) and the ADHD (i.e., 95% accuracy) samples.

[0151] FIG. 29B shows the result of the receiver operating characteristic curve (ROC) analysis for Bipolar versus ADHD. As shown in FIG. 29B, using 33 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.98 ± 0.01 ≡ 98.00% ± 0.01%).

[0152] FIG. 30A shows the result of the Schizophrenia versus ADHD stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 3) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 36 TE that showed significant differences between Schizophrenia and ADHD samples.

[0153] FIG. 30A verifies that the 36 TE between Schizophrenia versus ADHD exhibited a high specificity in distinguishing the two samples apart. Specifically, these 36 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Schizophrenia (i.e., 96% accuracy) and the ADHD (i.e., 97% accuracy) samples.

[0154] FIG. 30B shows the result of the receiver operating characteristic curve (ROC) analysis for Schizophrenia versus ADHD. As shown in FIG. 30B, using 37 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.98 ± 0.02 ≡ 98.00% ± 0.02%).

[0155] FIG. 31A shows the result of the Control versus Bipolar stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 5) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 30 TE that showed significant differences between Control and Bipolar samples.

[0156] FIG. 31A verifies that the 30 TE between Control versus Bipolar exhibited a high specificity in distinguishing the two samples apart. Specifically, these 30 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 91% accuracy) and the Bipolar (i.e., 82% accuracy) samples.

[0157] FIG. 31B shows the result of the receiver operating characteristic curve (ROC) analysis for Control versus Bipolar. As shown in FIG. 31B, using 30 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.93 ± 0.04 ≡ 93.00% ± 0.04%).

[0158] FIG. 32A shows the result of the Control versus ADHD stratified (test size = 50%) classification (1,000 repetitions) by KNN (K = 7) classifier using the cosine similarity as its distance measure (distance function), in which the inventors used the 36 TE that showed significant differences between Control and Bipolar samples.

[0159] FIG. 32A verifies that the 36 TE between Control versus ADHD exhibited a high specificity in distinguishing the two samples apart. Specifically, these 36 TE (out of total of 2322= 53,824 TE) showed a substantially above average accuracy for both the Control (i.e., 92% accuracy) and the ADHD (i.e., 81% accuracy) samples.

[0160] FIG. 32B shows the result of the receiver operating characteristic curve (ROC) analysis for Control versus ADHD. As shown in FIG. 32B, using 36 TE resulted in a mean ROC that was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.94 ± 0.03 ≡ 94.00% ± 0.03%).

[0161] FIG. 33A and FIG. 33B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Bipolar versus ADHD samples based on the 33 TE (using “cosine similarity” measure) revealed that (FIG. 33A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 33B) resulted in rather a uniform spatial representation of the two samples.

[0162] FIG. 34A and FIG. 34B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Schizophrenia versus ADHD samples based on the 36 TE (using “cosine similarity” measure) revealed that (FIG. 34A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 34B) resulted in rather a uniform spatial representation of the two samples.

[0163] FIG. 35A and FIG. 35B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Control versus Bipolar samples based on the 30 TE (using “cosine similarity” measure) revealed that (FIG. 35A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 35B) resulted in rather a uniform spatial representation of the two samples.

[0164] FIG. 36A and FIG. 36B show the results in terms of the representational similarity matrix (RSM). An examination of RSM of the Control versus ADHD samples based on the 36 TE (using “cosine similarity” measure) revealed that (FIG. 36A) these two samples formed two distinct and well-separated sub-spaces. However, the use of all of the 2322= 53,824 TE (FIG. 36B) resulted in rather a uniform spatial representation of the two samples.

[0165] (Results of Control versus Alzheimer’s Disease(AD)) FIG. 55A to 55C show the result of the Control versus Alzheimer’s Disease (AD) Leave-One-Out Classification by Support Vector Machine (SVM) classifier using a radial basis function (RBF) kernel, the stratified (test size = 50%) classification (1,000 repetitions) by SVM classifier using RBF, and the receiver operating characteristic curve (ROC) analysis for Control versus AD, respectively. FIG. 55A and FIG. 55B verify that the TE between Control versus AD exhibited a high specificity in distinguishing the two samples apart. Also as illustrated in FIG. 55C, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.89 ± 0.02 ≡ 89.00% ± 0.02%). Importantly, the major target region implicated in the case of TEControl> TEADwas identified as the right lateral amygdala. This is notable because the amygdala, particularly in the right hemisphere, is strongly associated with Alzheimer’s pathology [9-11], underscoring the biological plausibility and relevance of the observed TE differences.

[0166] (Results of Parkinson’s Disease (PD) versus PD Prodrome) FIG. 56A to 56C show the result of the PD versus PD Prodrome Leave-One-Out Classification by KNN (K=20) classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN (K=20) classifier using the cosine similarity as its distance measure (distance function), and the receiver operating characteristic curve (ROC) analysis for PD versus PD Prodrome, respectively. Prodrome is a medical term for early signs or symptoms of an illness or health problem that appear before the major signs or symptoms start. FIG. 56A and FIG. 56B verify that the TE between PD versus PD Prodrome exhibited a high specificity in distinguishing the two samples apart. Also as illustrated in FIG. 56C, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.89 ± 0.03 ≡ 89.00% ± 0.03%). It is noted that two of the Source ROIs in this analysis were identified as the Caudate and Putamen, which are core components of the basal ganglia and part of the basal ganglia-prefrontal cortex (PFC) dopaminergic pathway. This is important because in Parkinson’s disease, this pathway undergoes functional disruptions, and the lateral PFC - critical for executive function - is known to be affected by altered dopaminergic signaling. Moreover, the globus pallidus, another basal ganglia structure, plays a pivotal role in PD pathology and is a major therapeutic target for both pharmacological treatments and deep brain stimulation (DBS), supporting the biological significance of the observed TE alterations.

[0167] (τ as a Distinct Marker for mental disorder) In the method according to the above embodiment, the method comprises calculating a measure of directed flow of information (i.e. TE) for each ROI pair based on fMRI data pertaining to each ROI pair, inputting the measure of directed flow of information for selected ROI pairs into a trained model, and outputting information pertaining to a mental disorder of the subject based on the trained model; however, it is not limited thereto. Considering this critical role of τ, it is natural to ask whether differential timing in the brain’s information transfer can be a marker of mental disorder, i.e., independent of their corresponding TE. In other words, the measure of directed flow of information can be a time delay for transferring information. As a first modification, for example, the method may comprise calculating, for each ROI pair, a time delay for transferring information at which the transfer entropy from the source region to the target region is maximized; inputting the time delays for selected ROI pairs into a trained model; and outputting information pertaining to a mental disorder of the subject based on the trained model. In this manner, the time delay τ, associated with the calculation of TE, may also be used as a distinct marker instead of TE.

[0168] For example, Control and Major Depressive Disorder (MDD) samples, first, the inventors found τ, per Control and MDD samples, per individual, per pair of ROIs, such that it maximized from Source to Target transfer of information. This was performed through a brute-force strategy [8] in which τ was incremented one data-point at the time. As a result, this increment step size becomes dependent on the fMRI device Repetition Time (TR) i.e., its sampling rate because TR determines the distance between subsequent recorded data points. For example, when the inventors computed τ based on the dataset of UCLA with TR = 2.0 sec, τ’s increment in a 1-data-point corresponded to ≒ 2-second increment steps. The inventors carried out this increment within the range τ ∈ [1, ...., 15] data points i.e., approximately 2 through 30 seconds. The inventors calculated the value of τ (i.e. τbest), per individual, per sample, per paired ROIs, that resulted in maximum TE (i.e. TEbest) for a given paired ROI. As a result, 53,824 τbestvalues (i.e. 2322τbestvalues), per individual, per Control / MDD samples were obtained.

[0169] The inventors then bootstrapped (10,000 random sampling with replacement at 95% confidence interval (CI95%)) these τbestvalues within each Control and MDD samples (i.e., Control and MDD samples, separately). Thereafter, by using each τbestvalue, a plurality of ROI pairs are selected that can be related to information pertaining to the mental disorder of the subject being presented. In other words, ROI pairs whose importance exceeds a predetermined level are selected. Here, the method of selecting ROI pairs in the case where the target mental disorder is MDD is described.

[0170] After computing τbestfor all possible pairs (i.e., 2322= 53,824 τbestvalues) the inventors determined the significant difference between two phenotypes (e.g., Control vs. MDD), per ROI pair, through following steps.

[0171] The inventors performed bootstrap (i.e., 10,000 repetitions with replacement of samples) test of significant mean-difference (i.e., μPhenotype1- μPhenotype2) at 95% confidence interval (i.e., p < 0.05) between the two phenotypes. Here, Phenotype 1 is Control and Phenotype 2 is MDD. If this difference is significant, then the inventors check its computed p-value against the Bonferroni-corrected p-value (i.e., 0.05 / 232 ≒ 0.00022). If the computed p-value is less than Bonferroni-corrected p-value, then the inventors considered the mean-difference between the given ROI pair significant and selected it as a pair-of-interest. Otherwise, the inventors consider their mean-difference non-significant and discard them (i.e., even if their calculated bootstrapped p-value is less than 0.05, the inventors did not select them as pair-of-interest).

[0172] The inventors use the significantly different τ-Based Source to Destination between two phenotypes (e.g., Control versus MDD) as input to mathematical models, and perform diagnosis prediction based on these τ-Based Source to Destination.

[0173] It is of significant importance to note that in all reported scenarios, a simple K-Nearest-Neighbor (KNN) model with cosine similarity function (K≦20) was sufficient to achieve the high diagnosis prediction accuracies that the inventors present below. Table 1 summarizes the K value associated with each of the paired-phenotype diagnosis prediction.

[0174] Here, K values associated with diagnosis prediction of each of paired-phenotypes, using K-Nearest-Neighbor (KNN) model and cosine similarity function. In KNN, K represents the number of neighbors based on which an individual's diagnosis is predicted (Mean = 6.3333, Median = 3.00, Standard Deviation = 5.6293, 95.0% Confidence Interval (CI95%)) = [3.0, 20.00]).

[0175] (The results of τ as a Distinct Marker for mental disorder) FIG. 37A to FIG. 37K illustrate the distribution of τbestof Bipolar Disorder (BPD) vs Control, Schizophrenia vs Control, ADHD vs Control, Clinically High Risk - Not Converted (CHR-NC) vs Control, Affective Psychosis (AP) vs Control, Non-Affective Psychosis (NAP), MDD vs Control, Ketamine Therapy for MDD (KET) vs Control, Electroconvulsive Therapy for MDD (ECT), Anxiety vs Control, and Epilepsy versus Control, respectively.

[0176] FIG. 38A to FIG. 38D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Bipolar Disorder versus Control, respectively. FIG. 38A and FIG. 38B verify that the τbestbetween Bipolar Disorder versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 38C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 38D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.92 ± 0.02 ≡ 92.00% ± 0.02%).

[0177] FIG. 39A to FIG. 39D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Schizophrenia versus Control, respectively. FIG. 39A and FIG. 39B verify that the τbestbetween Schizophrenia versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 39C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 39D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.95 ± 0.02 ≡ 95.00% ± 0.02%).

[0178] FIG. 40A to FIG. 40D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for ADHD versus Control, respectively. FIG. 40A and FIG. 40B verify that the τbestbetween ADHD versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 40C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 40D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.96 ± 0.02 ≡ 96.00% ± 0.02%).

[0179] FIG. 41A to FIG. 41D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Clinically High Rish Psychosis - Not Converted (CHR-NC) versus Control, respectively. FIG. 41A and FIG. 41B verify that the τbestbetween CHR-NC versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 41C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 41D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.89 ± 0.03 ≡ 89.00% ± 0.03%).

[0180] FIG. 42A to FIG. 42D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Affective Psychosis (AP) versus Control, respectively. FIG. 42A and FIG. 42B verify that the τbestbetween AP versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 42C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 42D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.99 ± 0.01 ≡ 99.00% ± 0.01%).

[0181] FIG. 43A to FIG. 43D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Non-Affective Psychosis (NAP) versus Control, respectively. FIG. 43A and FIG. 43B verify that the τbestbetween NAP versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 43C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 43D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.97 ± 0.01 ≡ 97.00% ± 0.01%).

[0182] FIG. 44A to FIG. 44D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Affective Psychosis (AP) versus Non-Affective Psychosis (NAP), respectively. FIG. 44A and FIG. 44B verify that the τbestbetween AP versus NAP exhibited a high specificity in distinguishing the two samples apart. FIG. 44C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 44D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.99 ± 0.01 ≡ 99.00% ± 0.01%).

[0183] FIG. 45A to FIG. 45D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for MDD versus Control, respectively. FIG. 45A and FIG. 45B verify that the τbestbetween MDD versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 45C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 45D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.97 ± 0.02 ≡ 97.00% ± 0.02%).

[0184] FIG. 46A to FIG. 46D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Ketamine Therapy for MDD (KET) versus Control, respectively. FIG. 46A and FIG. 46B verify that the τbestbetween KET versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 46C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 46D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.99 ± 0.01 ≡ 99.00% ± 0.01%).

[0185] FIG. 47A to FIG. 47D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Electroconvulsive Therapy for MDD (ECT) versus Control, respectively. FIG. 47A and FIG. 47B verify that the τbestbetween ECT versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 47C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 47D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.99 ± 0.00 ≡ 99.00% ± 0.00%).

[0186] FIG. 48A to FIG. 48D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Ketamine Therapy for MDD (KET) versus Electroconvulsive Therapy for MDD (ECT), respectively. FIG. 48A and FIG. 48B verify that the τbestbetween KET versus ECT exhibited a high specificity in distinguishing the two samples apart. FIG. 48C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 48D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.98 ± 0.01 ≡ 98.00% ± 0.01%).

[0187] FIG. 49A to FIG. 49D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for MDD versus Ketamine Therapy for MDD (KET), respectively. FIG. 49A and FIG. 49B verify that the τbestbetween MDD versus KET exhibited a high specificity in distinguishing the two samples apart. FIG. 49C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 49D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.99 ± 0.00 ≡ 99.00% ± 0.00%).

[0188] FIG. 50A to FIG. 50D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for MDD versus Electroconvulsive Therapy for MDD (ECT), respectively. FIG. 50A and FIG. 50B verify that the τbestbetween MDD versus ECT exhibited a high specificity in distinguishing the two samples apart. FIG. 50C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 50D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.98 ± 0.01 ≡ 98.00% ± 0.01%).

[0189] FIG. 51A to FIG. 51D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Anxiety versus Control, respectively. FIG. 51A and FIG. 51B verify that the τbestbetween Anxiety versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 51C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 51D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.98 ± 0.01 ≡ 98.00% ± 0.01%).

[0190] FIG. 52A to FIG. 52D illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN classifier using the cosine similarity as its distance measure (distance function), the representational similarity matrix (RSM), and the receiver operating characteristic curve (ROC) analysis for Epilepsy versus Control, respectively. FIG. 52A and FIG. 52B verify that the τbestbetween Epilepsy versus Control exhibited a high specificity in distinguishing the two samples apart. FIG. 52C revealed that these two samples formed two distinct and well-separated sub-spaces. Also as illustrated in FIG. 52D, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.94 ± 0.02 ≡ 94.00% ± 0.02%).

[0191] (Results of Control versus Alzheimer’s Disease(AD)) FIG. 57 illustrates the distribution of τbestof Control versus Alzheimer’s Disease (AD). FIG. 58A to 58C show the result of the Control versus AD Leave-One-Out Classification by a linear Support Vector Machine (SVM) classifier, the stratified (test size = 50%) classification (1,000 repetitions) by a linear SVM, and the receiver operating characteristic curve (ROC) analysis for Control versus AD, respectively. FIG. 58A and FIG. 58B verify that the τbestbetween Control versus AD exhibited a high specificity in distinguishing the two samples apart. Also as illustrated in FIG. 58C, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.87 ± 0.02 ≡ 87.00% ± 0.02%).

[0192] (Results of Parkinson’s Disease (PD) versus PD Prodrome) FIG. 59 illustrates the distribution of τbestof Parkinson’s Disease (PD) versus PD Prodrome. FIG. 60A to 60C show the result of the PD versus PD Prodrome Leave-One-Out Classification by Naive Bayes classifier, the stratified (test size = 50%) classification (1,000 repetitions) by Naive Bayes, and the receiver operating characteristic curve (ROC) analysis for Control versus AD, respectively. FIG. 60A and FIG. 60B verify that the τbestbetween PD versus PD Prodrome exhibited a high specificity in distinguishing the two samples apart. Also as illustrated in FIG. 60C, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.84 ± 0.04 ≡ 84.00% ± 0.04%).

[0193] (max-TE-based marker for mental disorder) As a second modification, for example, the method may comprise calculating transfer entropy for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; inputting transfer entropy for selected ROI pairs into a trained model; and outputting information pertaining to a mental disorder of the subject based on the trained model, wherein transfer entropy for each ROI pair is calculated based on a time delay for transferring information from the source region to the target region at which the transfer entropy from the source region to the target region is maximized. In this manner, the maximized TE may also be used as a distinct marker. Similar to the above embodiment and the first modification, three approaches are based on TE. All three markers have different information content and must be interpreted differently. Especially, the first modification and the second modification offer complementary information on timing of information flow and information flow itself.

[0194] (Comparison between TE-based marker and τ-based marker for ASD) Although both TE-based marker and τ-based marker exhibited a high specificity in distinguishing the above mental disorders, in distinguishing Autism Spectrum Disorder (ASD, both medicated and not-medicated), the inventors found TE-based marker is superior to τ-based marker. FIG. 53A to FIG. 53C illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function) for Medicated ASD versus Control, Not-Medicated ASD versus Control, and Not-Medicated ASD versus Medicated ASD using TE-based marker, respectively. Here, TE for each ROI pair is calculated based on the time delay, wherein the time delay has been calculated based on an average value of a time delay at which the transfer entropy for each ROI pair is a maximum. On the other hand, FIG. 54A to FIG. 54C illustrate the result of the Leave-One-Out Classification by KNN classifier using the cosine similarity as its distance measure (distance function) for Medicated ASD versus Control, Not-Medicated ASD versus Control, and Not-Medicated ASD versus Medicated ASD using τ-based marker, respectively. Comparing FIG. 53A to FIG. 53C and FIG. 54A to FIG. 54C, TE-based marker exhibited a higher specificity in distinguishing the two samples apart than τ-based marker. Accordingly, in diagnosing, prognosing, or monitoring a response to treatment of ASD, it is better to utilize TE-based marker instead of τ-based marker.

[0195] (Results of Control versus Alzheimer’s Disease(AD)) FIG. 61A to 61C show the result of the Control versus AD Leave-One-Out Classification by linear Support Vector Machine (SVM) classifier, the stratified (test size = 50%) classification (1,000 repetitions) by linear SVM, and the receiver operating characteristic curve (ROC) analysis for Control versus AD, respectively. FIG. 61A and FIG. 61B verify that the maximized TE between Control versus AD exhibited a high specificity in distinguishing the two samples apart. Also as illustrated in FIG. 61C, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.90 ± 0.02 ≡ 90.00% ± 0.02%).

[0196] (Results of Parkinson’s Disease (PD) versus PD Prodrome) FIG. 62A to 62C show the result of the PD versus PD Prodrome Leave-One-Out Classification by KNN (K=20) classifier using the cosine similarity as its distance measure (distance function), the stratified (test size = 50%) classification (1,000 repetitions) by KNN (K=20) classifier using the cosine similarity as its distance measure (distance function), and the receiver operating characteristic curve (ROC) analysis for PD versus PD Prodrome, respectively. FIG. 62A and FIG. 62B verify that the maximized TE between PD versus PD Prodrome exhibited a high specificity in distinguishing the two samples apart. Also as illustrated in FIG. 62C, the mean ROC was significantly above chance level (Area Under the (ROC) Curve (AUC) = 0.87 ± 0.03 ≡ 87.00% ± 0.03%).

[0197] Although the present disclosure is based on embodiments and drawings, it is to be noted that various changes and modifications will be apparent to those skilled in the art based on the present disclosure. Therefore, such changes and modifications are to be understood as included within the scope of the present disclosure. For example, the functions and the links included in the various units and steps may be reordered in any logically consistent way. Furthermore, a plurality of units and steps may be combined into one, or a single unit or step may be divided.

[0198] For example, the configuration and operations of the information processing apparatus 10 in the above embodiment may be distributed to a plurality of computers capable of communicating with each other.

[0199] For example, in the above embodiments, the measure of directed flow of information for each ROI pair, such as TE, is based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair, other data can be used instead of fMRI data. For example, Electroencephalography (EEG), Magnetoencephalography (MEG) data may also be utilized.

[0200] The terms "diagnosing" and "diagnosis" as used herein refer to methods by which the skilled artisan can estimate and even determine whether or not a subject is suffering from a given disease or condition (e.g., mental disorders) or may develop a given disease or condition (e.g., mental disorders) in the future.

[0201] In some embodiments, a subject is diagnosed with mental disorders based on the information presented or output by the method of the present disclosure.

[0202] In some embodiments, diagnosing includes a step comparing the information presented or output by the method of the present disclosure between a subject and a healthy control.

[0203] In some embodiments, the information presented or output by the method of the present disclosure of a subject may be compared with that of the subject at different time points.

[0204] In some embodiments, the method of the present disclosure may be used for a support method for diagnosing.

[0205] The terms "prognosing" or "prognosis" as used herein refers to the prediction of the probable progression and outcome of a given disease or condition (e.g., mental disorders).

[0206] In some embodiments, prognosis also refers to the ability to demonstrate a positive or negative response to therapy or other treatment regimens, for a given disease or condition (e.g., mental disorders) in the subject. In some embodiments, prognosis refers to the ability to predict the presence or diminishment of symptoms associated with a given disease or condition (e.g., mental disorders).

[0207] In some embodiments, prognosis of subject with mental disorders is done based on the information presented or output by the method of the present disclosure.

[0208] In some embodiments, the correlation between prognosis and information may also be stored in a database. In some embodiments, prognosis of subject with mental disorders may be done with comparison of such a database and the information presented or output by the method of the present disclosure.

[0209] In some embodiments, the method of the present disclosure may be used for a support method for prognosing.

[0210] The term "monitoring a response to treatment" as used herein refers to assessing a subject suffering from a given disease or condition (e.g., mental disorders) at successive time intervals during or following treatment to determine whether disease symptoms have worsened, stabilized, or improved (i.e., become less severe) as a result of the treatment.

[0211] In some embodiments, monitoring a response to treatment is done based on the information presented or output by the method of the present disclosure.

[0212] As used herein, "treatment" or "treating" refers to any type of intervention or process performed on, or the administration of drugs to the subject with the objective of reversing, alleviating, ameliorating, inhibiting, slowing down or preventing the onset, progression, development, severity or recurrence of a symptom, complication or condition.

[0213] As a non-limiting example, a treatment can include a pharmacological agent, an electroconvulsive therapy, a cognitive rehabilitation therapy, a surgical treatment, behavioral therapy, physical therapy, gene therapy, radiation therapy, or a combination of these modalities.

[0214] In some embodiments, monitoring a response to treatment includes a step comparing the information presented or output by the method of the present disclosure between a subject before starting the treatment and that of the subject after starting the treatment.

[0215] In some embodiments, monitoring a response to treatment includes a step comparing the information presented or output by the method of the present disclosure between a subject before starting the treatment and that of the subject prior to the onset of a given disease or condition (e.g., mental disorders).

[0216] In some embodiments, the information presented or output by the method of the present disclosure may be obtained at successive time intervals during or following treatment.

[0217] In some embodiments, the method of the present disclosure may be used for a support method for monitoring a response to treatment.

[0218] In some embodiments, a method for monitoring a response to treatment may comprise: selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of a subject before starting the treatment and a target region of the brain of the subject before starting the treatment, calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair, presenting, based on the measure of directed flow of information for selected ROI pairs, information pertaining to a mental disorder of the subject before starting the treatment, starting the treatment against the subject, selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of a subject after starting the treatment and a target region of the brain of the subject after starting the treatment, calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair, presenting, based on the measure of directed flow of information for selected ROI pairs, information pertaining to a mental disorder of the subject after starting the treatment, comparing the information pertaining to a mental disorder of the subject before starting the treatment and the information pertaining to a mental disorder of the subject after starting the treatment. REFERENCES The following references are hereby incorporated herein by reference in their entirety for all purposes: [1] A. Stevens and J. Price, Evolutionary Psychiatry: a New Beginning, Routledge Mental Health Classic Edition, 2nd Edition (2000). [2] https: / / www.ncbi.nlm.nih.gov / medgen / 7916#:~:text=Definition,from%20NCI%5D [3] American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. 5th ed. American Psychiatric Association; Washington, DC, USA: 2013. [4] De Giorgi, R., De Crescenzo, F., D’Alo, G.L., Rizzo Pesci, N., Di Franco, V., Sandini, C. and Armando, M., 2019. Prevalence of non-affective psychoses in individuals with autism spectrum disorders: a systematic review. Journal of clinical medicine, 8(9), p.1304. [5] Iraji, A., Fu, Z., Faghiri, A., Duda, M., Chen, J., Rachakonda, S., DeRamus, T., Kochunov, P., Adhikari, B.M., Belger, A. and Ford, J.M., 2022. Canonical and Replicable Multi-Scale Intrinsic Connectivity Networks in 100k+ Resting-State fMRI Datasets, Human Brain Mapping, 44(17), 5729-5748 (2023). [6] Schaefer, A., Kong, R., Gordon, E.M., Laumann, T.O., Zuo, X.N., Holmes, A.J., Eickhoff, S.B., Yeo, B.T.: Local-global parcellation of the human cerebral cortex from intrinsic functional connectivity mri. Cerebral Cortex 28(9), 3095-3114 (2018) [7] Tian, Y., Margulies, D.S., Breakspear, M., Zalesky, A.: Topographic organization of the human subcortex unveiled with functional connectivity gradients. Nature neuroscience 23(11), 1421-1432 (2020) [8] Wibral, M., Pampu, N., Priesemann, V., Siebenhu hner, F., Seiwert, H., Lindner, M., Lizier, J.T., Vicente, R.: Measuring information-transfer delays. PloS one 8(2), 55809 (2013) [9] Braak, H. and Braak, E., 1991. Neuropathological stageing of Alzheimer-related changes. Acta neuropathologica, 82(4), pp.239-259.

[0010] Knafo, S., 2012. Amygdala in Alzheimer’s disease. The Amygdala-A Discrete Multitasking Manager. IntechOpen, pp.375-383.

[0011] Stouffer, K.M., Grande, X., Duzel, E., Johansson, M., Creese, B., Witter, M.P., Miller, M.I., Wisse, L.E. and Berron, D., 2024. Amidst an amygdala renaissance in Alzheimer’s disease. Brain, 147(3), pp.816-829

[0219] 10 Information processing apparatus 11 Controller 12 Memory 13 Input interface 14 Output interface 15 Communication interface

Claims

1. A method for diagnosing, prognosing, or monitoring a response to treatment of, a mental disorder in a subject to be executed by one or more information processing apparatuses, the method comprising: selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of the subject and a target region of the brain of the subject; calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; inputting the measure of directed flow of information for selected ROI pairs into a trained model; and outputting information pertaining to a mental disorder of the subject based on the trained model.

2. The method according to claim 1, wherein the measure of directed flow of information is transfer entropy.

3. The method according to claim 2, wherein calculating transfer entropy for each ROI pair comprises: calculating transfer entropy for each ROI pair based on a time delay for transferring information from a source region of the brain of the subject to a target region of the brain of the subject in each ROI pair, wherein the time delay has been calculated based on an average value of a time delay at which the transfer entropy for each ROI pair is a maximum.

4. The method according to claim 1, wherein the mental disorder is schizophrenia, bipolar disorder, attention deficit / hyperactivity disorder, depression, drug-resistant depression, anxiety, early psychosis, autism spectrum disorder, epilepsy, Alzheimer's, or dementia.

5. The method according to claim 4, wherein the method is that for monitoring a response to treatment, the selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of the subject and a target region of the brain of the subject, the calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair, the inputting the measure of directed flow of information for selected ROI pairs into a trained model, and the outputting information pertaining to a mental disorder of the subject based on the trained model are performed on the subject before and after starting the treatment; and wherein the method further comprises comparing the information pertaining to a mental disorder of the subject before starting the treatment and the information pertaining to a mental disorder of the subject after starting the treatment.

6. A method for diagnosing, prognosing, or monitoring a response to treatment of a mental disorder in a subject, comprising: selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of the subject and a target region of the brain of the subject, calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; and presenting, based on the measure of directed flow of information for selected ROI pairs, information pertaining to a mental disorder of the subject.

7. The method according to claim 6, wherein the measure of directed flow of information is transfer entropy.

8. The method according to claim 7, wherein the presenting, based on transfer entropy for each ROI pair, the information pertaining to the mental disorder of the subject comprises: classifying the subject by a mathematical model based on the transfer entropy of selected ROI pairs and presenting the information pertaining to the mental disorder of the subject.

9. The method according to claim 8, wherein the mathematical model is a K-Nearest-Neighbor classifier, logistic regression classifier, support vector machine classifier or Naive Bayes classifier.

10. The method according to claim 8, wherein the mathematical model is a K-Nearest-Neighbor classifier and cosine similarity is used as a distance function.

11. The method according to claim 6, wherein the mental disorder is schizophrenia, bipolar disorder, attention deficit / hyperactivity disorder, depression, drug-resistant depression, anxiety, early psychosis, autism spectrum disorder, epilepsy, Alzheimer's, or dementia.

12. The method according to claim 6, wherein the method is that for monitoring a response to treatment, the selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of the subject and a target region of the brain of the subject, the calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; and the presenting, based on the measure of directed flow of information for selected ROI pairs, information pertaining to a mental disorder of the subject are performed on the subject before and after starting the treatment; and wherein the method further comprises comparing the information pertaining to a mental disorder of the subject before starting the treatment and the information pertaining to a mental disorder of the subject after starting the treatment.

13. A method for diagnosing, prognosing, or monitoring a response to treatment of, a mental disorder in a subject to be executed by one or more information processing apparatuses, the method comprising: selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of the subject and a target region of the brain of the subject; calculating, for each ROI pair, a time delay for transferring information at which the transfer entropy from the source region to the target region is maximized; inputting the time delays for selected ROI pairs into a trained model; and outputting information pertaining to a mental disorder of the subject based on the trained model.

14. The method according to claim 13, wherein the mental disorder is schizophrenia, bipolar disorder, attention deficit / hyperactivity disorder, depression, drug-resistant depression, anxiety, early psychosis, autism spectrum disorder, epilepsy, Alzheimer's, or dementia.

15. The method according to claim 1, wherein the measure of directed flow of information is a time delay for transferring information.

16. The method according to claim 2, wherein calculating transfer entropy for each ROI pair comprises: calculating transfer entropy for each ROI pair based on a time delay for transferring information from a source region of the brain of the subject to a target region of the brain of the subject in each ROI pair at which the transfer entropy from the source region to the target region is maximized.

17. The method according to claim 16, wherein the mental disorder is schizophrenia, bipolar disorder, attention deficit / hyperactivity disorder, depression, drug-resistant depression, anxiety, early psychosis, autism spectrum disorder, epilepsy, Alzheimer's, or dementia.

18. The method according to claim 1 or claim 16, wherein the mental disorder is Parkinson’s Disease.

19. An information processing apparatus comprising a controller, wherein the controller is configured to select a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of a subject and a target region of the brain of the subject; calculate a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; input the measure of directed flow of information for selected ROI pairs into a trained model; and output information pertaining to a mental disorder of the subject based on the trained model.

20. A program configured to cause a computer to execute operations comprising: selecting a plurality of region of interest (ROI) pairs, each ROI pair including a source region of a brain of a subject and a target region of the brain of the subject; calculating a measure of directed flow of information for each ROI pair based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair; inputting the measure of directed flow of information for selected ROI pairs into a trained model; and outputting information pertaining to a mental disorder of the subject based on the trained model.

21. A trained model to cause a computer to function to output an objective variable pertaining to a mental disorder of a subject based on explanatory variables pertaining to the data of the subject, wherein the explanatory variables comprise a measure of directed flow of information for each region of interest (ROI) pair, the measure of directed flow of information being calculated by selecting a plurality of ROI pairs, each ROI pair including a source region of a brain of a subject and a target region of the brain of the subject; and calculating based on functional magnetic resonance imaging (fMRI) data pertaining to each ROI pair, and wherein the objective variables comprise information pertaining to the mental disorder of the subject.

Images