Information processing device and brain lesion estimation method

Abstract

To provide means for cerebral lesion estimation optimized on the basis of the purpose.SOLUTION: An information processing device for selecting a test and an algorithm used for estimating a lesion risk in the brain of a subject from tests and algorithms prepared beforehand includes; an input reception unit for receiving an input of the purpose of the estimation or information for estimating the purpose of the estimation; and an estimation algorithm selection unit for selecting the algorithm used for the estimation on the basis of the input of the input reception unit, and selecting the test for acquiring a variable reflecting the state of the brain of the subject on the basis of the input of the input reception unit.SELECTED DRAWING: Figure 2A

Description

[Technical field]

[0001] The present invention relates to a brain lesion estimation technique with optimized resolution and sensitivity. [Background technology]

[0002] Information on brain lesions is necessary for diagnosis, treatment, and prognosis. Technologies for analyzing information on brain lesions are disclosed in, for example, Patent Document 1 and Patent Document 2.

[0003] Patent Document 3 also discloses an information processing system including a memory unit that stores test-brain state related information that associates test results, which are the result of the subject's behavior in response to a specified test, with the subject's brain state for a plurality of subjects, an input unit that accepts a first test result, which is the result of the first subject's behavior in response to the specified test, a control unit that estimates a brain state from the first test result based on the test-brain state related information, and an output unit that outputs the estimated brain state.

[0004] Patent Document 1 uses imaging technology to monitor brain lesions. Patent Document 2 uses an invasive probe to detect the characteristics of lesions. Patent Document 3 visualizes brain conditions and optimizes rehabilitation programs to provide early feedback on the effects of rehabilitation to patients and medical staff. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] U.S. Patent Application Publication No. 2015 / 0325139A1 [Patent Document 2] Special Publication No. 2019-513520 [Patent Document 3] Patent Publication No. 2021-49064 Summary of the Invention [Problem to be solved by the invention]

[0006] According to the inventors' study, one of the constraints in estimating brain lesions is the imbalance of the data set. That is, in some cases, lesions are present uniformly in many places in the subject's brain, while in other cases, the overall lesion level is low and lesions are present sparsely.

[0007] In this case, for example, if a lesion estimation model is generated based on data in which many lesions are uniformly distributed, it may not be effective for subjects in which the lesion is less severe and only sparsely distributed.

[0008] In actual tests, it is difficult to obtain a dataset with an appropriate distribution of brain lesions suitable for brain lesion estimation. For example, in a dataset with many brain divisions and small lesions (lesions), lesions may be present in only 5% or less of the subjects in approximately 70% of the areas. If an estimation model for estimating brain lesions is trained based on such a dataset, it is difficult to increase the detection sensitivity (or sensitivity), and there is a risk of overlooking lesion detection.

[0009] There are several methods for dividing the brain into regions where lesions are present, and detection sensitivity can be improved by reducing the number of brain divisions to broaden the region and loosening the criteria for positive judgment. However, there was no method for selecting and presenting tests that optimize the trade-off between resolution and detection sensitivity according to the purpose.

[0010] An object of the present invention is to provide a means for estimating brain lesions that is optimized based on a purpose. [Means for solving the problem]

[0011] A preferred aspect of the present invention is an information processing device that selects a test and an algorithm to be used to estimate a risk of a lesion in a subject's brain from pre-prepared tests and algorithms, the information processing device comprising: an input receiving unit that accepts input of a purpose of the estimation or information for estimating the purpose of the estimation; and an estimation algorithm selection unit that selects an algorithm to be used for the estimation based on the input from the input receiving unit, and an estimation algorithm selection unit (also referred to as an estimation formula selection unit 202) that determines the required resolution and estimation performance from the purpose of the estimation and selects a test to obtain variables reflecting the state of the subject's brain, based on the input from the input receiving unit.

[0012] Another preferred aspect of the present invention is a brain lesion estimation method for selecting a test and an algorithm to be used to estimate a lesion risk in a subject's brain from pre-prepared tests and algorithms using an information processing device, the method comprising: a first step of accepting an input of a purpose of the estimation or information for estimating the purpose of the estimation; and a second step of selecting an algorithm to be used for the estimation based on the input, and selecting a test for obtaining a variable reflecting the state of the subject's brain based on the input.

[0013] Further features related to the present disclosure will become apparent from the description of the present specification and the accompanying drawings.

[0014] The descriptions in this specification are exemplary and illustrative only and are not intended to limit the scope or application of the present disclosure in any way. Effect of the Invention

[0015] According to the present invention, a means for estimating a brain lesion that is optimized based on a purpose can be provided. [Brief description of the drawings]

[0016] [Figure 1]Conceptual diagram showing examples of pathology risk estimation at different resolutions. [Figure 2A] FIG. 2 is a block diagram showing the configuration of an environment development system 200. [Figure 2B] FIG. 3 is a block diagram showing the configuration of an estimation calculation system 300. [Diagram 3] FIG. 2 is a block diagram showing an example of the configuration of a system that realizes an environment development system 200 and an estimation calculation system 300. [Figure 4A] FIG. 2 is a process flow diagram of the environment development system 200. [Figure 4B] 13 is a process flow diagram of the environment development system 200 (continued). [Figure 5A] An image showing an example of a user interface for new subjects. [Figure 5B] FIG. 13 is an image showing an example of a user interface for registered subjects. [Figure 6] FIG. 2 is a table diagram showing an example of a subject-environment-test table 203. [Figure 7] FIG. 13 is a table diagram showing an example of a lesion estimation formula table 205. [Figure 8] FIG. 13 is a table diagram showing an example of a variable test estimation equation table 207. [Figure 9] FIG. 2 is a table showing an example of the subject database 204. [Figure 10] FIG. 11 is a process flow diagram of the subject data selection process S408. [Figure 11] A tabular diagram illustrating the concept of lesion sparsity computation. [Figure 12] FIG. 13 is a graph showing the relationship between the probability of lesion occurrence in subject data and the sparsity of the lesion. [Figure 13] FIG. 11 is a process flow diagram showing the process flow S409 of mathematical modeling. [Figure 14] 13 is an image showing an example of a user interface developed by the test development unit 209. [Figure 15] FIG. 4 is a process flow diagram showing the estimation process by the estimation resolution unit 302. [Figure 16A] An image showing an example of a user interface for estimation results for new subjects. [Figure 16B] FIG. 13 is an image showing an example of a user interface for estimation results for enrolled subjects. [Figure 17] A table showing the relationship between applications, models, and datasets for model development. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0017] The embodiments will be described in detail with reference to the drawings. However, the present invention should not be interpreted as being limited to the description of the embodiments shown below. It will be easily understood by those skilled in the art that the specific configuration can be changed without departing from the concept or purpose of the present invention.

[0018] In the configurations of the embodiments described below, the same parts or parts having similar functions are denoted by the same reference numerals in different drawings, and duplicated explanations may be omitted.

[0019] When there are multiple elements having the same or similar functions, they may be described by using the same reference numerals with different subscripts. However, when there is no need to distinguish between multiple elements, the subscripts may be omitted.

[0020] The designations "first," "second," "third," and the like in this specification are given to identify components, and do not necessarily limit the number, order, or content. Furthermore, numbers for identifying components are used in different contexts, and a number used in one context does not necessarily indicate the same configuration in another context. Furthermore, a component identified by a certain number does not prevent the component from also having the function of a component identified by another number.

[0021] In order to facilitate understanding of the invention, the position, size, shape, range, etc. of each component shown in the drawings, etc. may not represent the actual position, size, shape, range, etc. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings, etc.

[0022] All publications, patents, and patent applications cited herein are hereby incorporated by reference in their entirety.

[0023] Brain lesions can be defined by labeling using hierarchical levels (e.g., hemisphere, lobe, and gyrus levels) according to the size of the lesion area. Sparse lesions can be labeled as larger lesion areas with reduced resolution. As a result, the detection sensitivity of lesion estimation can be improved.

[0024] The system shown in the embodiment can take into account the trade-off between resolution and estimation performance. Based on the purpose (e.g. screening, prognostic monitoring, etc.), the subject's medical history, or the location (clinic, hospital, rehabilitation facility, home, etc.), the required resolution and estimation performance are determined, and the test and estimation algorithm corresponding to the optimal resolution and detection sensitivity are selected. The selected algorithm can minimize the functional tests required to obtain an estimate of the lesion.

[0025] Figure 1 shows an example of pathology risk estimation at different resolutions. The color bar 101 shows a measure of the probability of the estimated pathology risk. The probability is shown, for example, from a minimum of 0% to a maximum of 100%. Image 102 shows the highest resolution (here, at the gyrus level). The lobe and hemisphere levels are visualized in images 103 and 104, respectively.

[0026] A gyrus is a raised wrinkled part of the cerebral cortex. A lobe is an anatomical division of the cerebrum, and there are six lobes, the frontal lobe, the temporal lobe, the parietal lobe, the occipital lobe, the insular lobe, and the limbic lobe, which are bounded by the major sulci. A hemisphere is the largest part of the brain, and the left and right hemispheres form a pair to make up the cerebrum.

[0027] Grey areas represent areas where lesion risk is estimated. Darker grey represents higher lesion risk. Lesion risk within a lobe or hemisphere determines the lesion risk at the relevant lobe or hemisphere. The risk probability scale at low resolution is adjusted by the highest probability at high resolution.

[0028] For example, the risk probability of the frontal lobe 105 is determined by the highest risk probability among the Middle Frontal Gyrus (MFG) 106, the Superior Frontal Gyrus 107, and a part of the Inferior Frontal Gyrus 108. That is, the lesion risk probability of the MFG 106 becomes the risk probability of the frontal lobe 105. Also, the risk probability of the hemisphere 109 is estimated by the risk probability of the Rolandic operculum 110, which is the highest probability at high resolution. In this way, the estimation of the lesion risk can be performed at multiple levels of resolution as necessary.

[0029] 2A is a block diagram showing a system configuration for environment development in lesion examination. The environment development system 200 provides an examination environment including an interface with a user. The environment development system 200 includes six units and a database of analysis information and patient data.

[0030] The input receiving unit 201 is a system interface for user settings, and receives information about the purpose of the examination or an input for estimating the purpose of the examination from a user (e.g., a medical professional performing the examination). Information about the purpose of the examination is, for example, whether the examination is for detecting or screening a lesion, or for measuring the effect of rehabilitation, etc. Information for estimating the purpose of the examination is, for example, information about the subject and information about the examination location.

[0031] The information about the subject is, for example, characteristic data and history data of the subject. The characteristic data of the subject is, for example, information about the age, sex, height, weight, physical condition, subjective symptoms, etc. The history data of the subject is, for example, the medical history of the subject, the presence or absence of a medical condition, a treatment history, and the results of past examinations.

[0032] The information on the testing location is information indicating the system location and the testing location, for example, whether the testing location is a clinical site or a non-clinical site. Generally, clinical is an environment where health insurance applies, including hospital inpatient wards, outpatient clinics, testing for medical purposes, and rehabilitation facilities. Non-clinical is within the home, medical checkups, and the like.

[0033] The estimation formula selection unit 202 selects an estimation algorithm (estimation formula) based on the input set by the input receiving unit 201. The selection is made using a subject-environment-test table 203, a subject database 204, and a lesion estimation formula table 205 stored in the database. If there is no suitable estimation algorithm based on the set input, a new estimation algorithm is developed in the formula development unit 206 using a variable test estimation formula table 207 stored in the database.

[0034] After the estimation algorithm is selected, the necessary algorithm variables related to the functional task are extracted in the formula input extraction unit 208 .

[0035] The information of the required functional tasks is managed by a test development unit 209, which creates a user interface that requests input for the subject's functional test. The created user interface is displayed on a display unit 210.

[0036] The environment development system 200 can specifically select an estimation formula according to user input, support the development of an estimation formula if there is no appropriate formula in response to the user input, and develop an estimation environment (i.e., a user interface for performing lesion risk testing).

[0037] The environment development system 200 can be realized, in a typical example, by a computer. Like a general computer, it includes an input device, an output device, a processing device, and a storage device, but these are not shown in FIG. 2A.

[0038] An input receiving unit 201, an estimated formula selection unit 202, a formula development unit 206, a formula input extraction unit 208, and a test development unit 209, which are indicated by dotted lines, can be realized by a processing device executing software stored in a storage device such as a semiconductor memory, and cooperating with an input device, an output device, and a storage device. A subject-environment-test table 203, a subject database 204, a lesion estimated formula table 205, and a variable test estimated formula table 207 are stored as data in a storage device such as a hard disk drive. A display unit 210 is an example of an output device.

[0039] 2B is a block diagram showing the configuration of an estimation calculation system 300. The estimation calculation system 300 performs estimation after information reflecting the state of the subject's brain is input to a user interface that is created and displayed via a test input receiving unit 301.

[0040] The estimation calculation system 300 can be realized by a computer in a typical example. It may be the same computer as the environment development system 200, or it may be a different computer. Like a general computer, it includes an input device, an output device, a processing device, and a storage device, but these are not shown in FIG. 2B.

[0041] A test input receiving unit 301 and an estimation resolution unit 302 indicated by dotted lines can be realized by a processing device executing software stored in a storage device such as a semiconductor memory. A subject database 204 and a lesion estimation formula table 205 are stored as data in a storage device such as a hard disk drive. A display unit 210 is an example of an output device.

[0042] The test input receiving unit 301 receives information or a score reflecting the brain state obtained by the subject performing a functional task (test). Examples of the functional task include questions to the subject, or finger tapping as disclosed in, for example, Patent Document 3.

[0043] The estimation process is performed using the lesion estimation formula table 205 and the estimation resolution unit 302. The estimation resolution unit 302 estimates the lesion risk with optimized resolution and detection performance (sensitivity, specificity, etc.) using an algorithm selected from the lesion estimation formula table 205 based on the quantified information obtained based on the information reflecting the brain state. The current estimate can be compared with previous estimates in the subject database 204, which can refer to the treatment history such as medication, rehabilitation, etc. The estimation result is displayed on the display unit 210.

[0044] With this configuration, the estimation calculation system 300 can calculate an estimate of disease risk based on the results of tests entered into the user interface, and determine the estimation resolution and detection performance (sensitivity, specificity, etc.) depending on purposes such as screening, prognosis recovery monitoring, etc.

[0045] 3 is a block diagram showing an example of a hardware configuration in which the functions of the environment development system 200 and the estimation calculation system 300 are connected via a network. A database server 3001, an environment development server 3002 in which the software of the environment development system 200 is implemented in memory 3021, and an estimation calculation server 3003 in which the software of the estimation calculation system 300 is implemented in memory 3031 are connected via a network 3004. The resources of each software and hardware are available via an input device 3032 of the estimation calculation server 3003.

[0046] The estimation calculation server 3003 is placed at a clinical site or a non-clinical site and can be used as a brain function testing device. The estimation calculation server 3003 can also execute environment development by the environment development server 3002 using the input device 3032.

[0047] 4A and 4B show the process flow of the environment developing system 200. The environment developing system 200 executes the following procedure.

[0048] In Fig. 4A, the estimation formula selection unit 202 receives an input set by a user on the input device 3032 via the input receiving unit 201 (S401). The user is, for example, a medical professional who performs a test. The input is, for example, a classification of a subject to be tested (e.g., a subject with a medical history, a normal or asymptomatic subject) and / or a system environment (e.g., a clinical site, a non-clinical site) and / or a test purpose / use (e.g., rehabilitation effect measurement, screening).

[0049] The estimation formula selection unit 202 acquires one or more appropriate test lists for estimating a disease risk by checking the set input against the subject-environment-test table 203 (S402). The subject-environment-test table 203 will be described in detail later with reference to FIG. 6.

[0050] If the subject has already taken a test in the past and performed estimation, a test list can be obtained by checking against the subject database 204. For example, if the subject has previously performed estimation with the same input in S401, an estimation formula similar to the test list used in the past can be selected.

[0051] One or more scores can be obtained from one test. The test is, for example, a question to the subject or a request for an action, and the score is a numerical answer or response to the question or request for an action. In Patent Document 3, the total movement distance in the finger tapping test, the MMSE (mini-mental state examination), the FIM (Functional Independence Measure), and the CAT (Clinical Assessment for Attention) scores are given as score information, and these can be used. These scores can be associated with the subject's brain state (see, for example, Patent Document 3).

[0052] The scores become variables in a estimating equation (e.g., a logistic regression model). The estimating equations are assumed to be independent for each brain region. Each region may have one or more equations with different variables.

[0053] The estimation formula selection unit 202 refers to the lesion estimation formula table 205 and selects an estimation formula for each region according to the subject classification and system environment (or examination purpose) specified in S403.

[0054] There may be cases where all variables of the estimation formula are not obtained from the scores obtained in the appropriate test. When there are multiple applicable estimation formulas in the same area, an estimation formula that uses a larger proportion of scores obtained in the appropriate test as variables, that is, an estimation formula that has a larger ratio of the number of available (obtained) score types to the number of all score types used in the applicable estimation formula, is selected from the Lesion Estimation Formula Table 205 (S403). The Lesion Estimation Formula Table 205 will be described in detail later with reference to FIG. 7.

[0055] Then, the estimation formula selection unit 202 checks whether the selected estimation formula covers all brain regions (S404). Here, covering a certain brain region means that a predetermined number (e.g., 50%) or more of scores can be used for the score types used in the estimation formula for the corresponding region.

[0056] If all brain regions are covered by the selected equation (Yes in S404), the estimated equation selection unit 202 rechecks whether unavailable scores are used in the selected estimated equation, i.e., whether all scores used for variables required in the estimated equation are obtained from the appropriate test list (S405).

[0057] If unavailable scores are not used in the selected estimation equation (No in S405), estimation is possible using the selected test (Figure 4B).

[0058] If some (e.g., less than 50%) of the scores used for variables required in the estimation equation cannot be obtained from the appropriate test list (Yes in S405), the estimation equation selection unit 202 determines whether the unavailable scores can be replaced with scores that can be obtained from the appropriate test list (S406).

[0059] Whether or not the scores that cannot be obtained from the appropriate test list can be replaced with scores that can be obtained from the appropriate test list is evaluated (S406) using the variable test estimation equation table 207. The variable test estimation equation table 207 will be described in detail later with reference to FIG.

[0060] If the variable can be replaced (Yes in S406), the estimation equation associated with the replaced variable is modified (S407). Estimation is performed using the modified estimation equation (FIG. 4B). The above steps S401 to S407 are executed by the estimation equation selection unit 202.

[0061] If the scores cannot be replaced with scores from the appropriate test list (No in S406), or if the estimated equations for each selected region do not cover all brain regions (No in S404), new estimated equations for the uncovered brain regions using variables from the appropriate test list are developed by the equation development unit 206 (S408-9). To develop the new equations, the subject database 204 stored in the database is used.

[0062] Not all subject data stored in the subject database 204 is used unconditionally. Subject data having variables in an appropriate test list or variables that can be substituted for variables in an appropriate test list according to the criteria for the suitability of the model is selected (S408). The subject data selection method will be described later.

[0063] By using the selected subject data (lesion data, variables, etc.), the lesion risk in the uncovered areas is modeled and an estimation equation is obtained (S409).

[0064] The development of new estimated equations will be described in detail later in Figures 10 and 13. After the entire brain region is covered by estimated equations with appropriate variables, we proceed to the estimation process in Figure 4B.

[0065] In Fig. 4B, a estimating equation with appropriate variables is used to estimate the lesion risk covering the entire brain region. First, scores corresponding to the variables required in the selected estimating equation are extracted by the formula input extraction unit 208 (S410).

[0066] A user interface for obtaining the extracted scores is created by the test development unit 209 based on the input items corresponding to the extracted scores (S411). The test development unit 209 informs the user which variables (e.g., Test A Items 1, 2, ...) are required, for example by a user interface displayed on the display unit 210. The user (e.g., a medical professional) then asks the subject to perform a test (e.g., a digitized test, a physical examination, etc.) corresponding to the variables (S412). The user interface will be described later in FIG. 14.

[0067] The test development unit 209 may directly display the test contents on the display unit 210 without the user's intervention, and prompt the subject to input the contents. The test to be performed may be a general test known for evaluating brain or body functions, such as that disclosed in Patent Document 3.

[0068] To obtain these variables, subjects may perform tests, but only purposefully selected tests (questions, items, etc.) are performed, thus saving the time required for variable collection and pathology risk estimation.

[0069] The above process requires less up-front user input to select the appropriate tests. The selection and development of estimation formulas based on user inputs is performed automatically. The estimation environment is automatically created with variable entries depending on the set test requirements and test variables required for the estimation formulas.

[0070] In the above process, a test is selected in S401 and then an algorithm is selected in S402, but the order may be reversed.

[0071] 5A and 5B are examples of user interfaces displayed on the display unit 210 for the user to enter input in step S401 executed by the input receiving unit 201. Depending on the subject, either a new subject interface 500a in FIG. 5A or a registered subject interface 500b in FIG. 5B may be selected as the user interface.

[0072] When setting up a subject 501, the input items change depending on whether the subject is new or already registered. When a new subject 501a is selected, data 502 of the target subject is newly registered, and the subject's information such as new ID 503, age 504, and gender 505 is saved in the subject database 204 (FIG. 5A).

[0073] For registered subjects 501b, the user interface calls up the data stored in the subject database 204 and displays it with a list of IDs 507, and provides a subject selection 506 function (FIG. 5B).

[0074] Also, subject classifications 508 are defined, for example, subjects with pre-existing conditions 508a or normal or asymptomatic subjects 508b. These are just examples and other classifications may be defined.

[0075] Further, a testing environment 509 is defined. The testing environment may be, for example, a clinical site 509a or a non-clinical site 509b. These are just examples, and other classifications may be defined.

[0076] To implement the above inputs in the estimation process, press the "Set OK" button 510.

[0077] As described above, we provide a practical interface for registering subjects and specifying the subject type and system environment. Based on the subject type and system environment, a suitable estimation model is then selected.

[0078] FIG. 6 is an example of the subject-environment-test table 203. The subject-environment-test table 203 lists tests 601 that are commonly used to evaluate brain functions. A test is a process for obtaining input that reflects the state of the subject by requesting the subject to perform a specific action. The test may be a publicly known test. For example, Test A is set to finger tapping, Test B to MMSE, and Test C to FIM. The contents of each test are separately converted into data and associated with each other.

[0079] There are items of clinical 603a, 604a, non-clinical 603b, 604b, subjects with a medical history 603, classification of normal or asymptomatic subjects 604, and test interval 602. Applicability of the matching test according to the combination of each item of clinical 603a, 604a, non-clinical 603b, 604b, subjects with a medical history 603, and normal or asymptomatic subjects 604 is indicated by a check mark.

[0080] The subject-environment-test table 203 is created by determining applicability in advance based on the knowledge and know-how of doctors, medical staff, researchers, etc. This is created by taking into consideration the practicality of a general brain-body function test that is executed in a specific system environment by a specific target subject.

[0081] For example, in Figure 6, for "normal / asymptomatic" and "non-clinical," Test E and Test F are applicable, while Test A to Test D are inappropriate. For example, when detecting brain lesions, in the early stages of the lesion, the subject is asymptomatic, and the lesion is generally observed with a low and sparse occurrence. In order to prevent lesions from being overlooked due to the absence of symptoms, estimation performance that improves the detection sensitivity of lesions is prioritized. Therefore, by adopting a test with high detection sensitivity even if it has low resolution, it can contribute to the purpose of screening lesions in the early stages in medical checkups, etc.

[0082] For example, in Figure 6, for "history of stroke" and "clinical", Test A to Test C are applicable, and Test D to Test F are inappropriate. In the later stages of lesions with obvious symptoms, it is necessary to obtain a more accurate lesion location to identify the cause of functional impairment. For this reason, resolution is prioritized over detection sensitivity.

[0083] In addition, for those with a history of stroke and in the non-clinical setting, a test with a higher resolution than the detection sensitivity is selected in rehabilitation facilities etc. to evaluate the effectiveness of rehabilitation.

[0084] Thus, subject-environment-test table 203 allows one to identify tests appropriate for a particular subject target (eg, patient, normal / asymptomatic) and system environment (eg, clinical, non-clinical site).

[0085] 7 is an example of the lesion estimation formula table 205. An estimation formula is prepared for each brain region, subject target (e.g., patient, normal / asymptomatic person), and environment (e.g., clinical, non-clinical).

[0086]

number

[0087] Equation 1 shows an example of an estimation equation for estimating the lesion probability. Equation 1 is the lesion probability Pregion1 The estimation formula for obtaining is shown below. The formula is defined for each brain region, and k or more estimation formulas with uniquely defined coefficients are required for k regions. A test may include two or more evaluation items (Item1, Item2, etc.). Equation 1 is the independent variable x n to the dependent variable P region1 This corresponds to a model for obtaining

[0088] The right-hand side of Equation 1 contains the coefficient a n and the variable x n where the coefficient a n is obtained from the disease estimation formula table 205, and the variable x n is obtained for each test item 701, for example, from the score that the subject gets by answering questions.

[0089] The disease estimation formula table 205 contains the coefficient a used for each estimation formula. n The intercept a0 is set uniquely for each combination of brain region, subject target, and environment (conveniently referred to as "test type") 702 to 706. In other words, the intercept a0 generally takes a different value for each test type 702 to 706. In addition, the coefficient a of the estimation equation corresponding to each test item 701 for each test type 702 to 706 is n also generally have different values.

[0090] The "0" in the disease estimation formula table 205 is the coefficient a of the estimation formula. n is 0, and therefore in this case, the score obtained from the corresponding test item 701 does not contribute to the calculation result of the estimated equation. For example, the variable obtained from "TestA Item1" does not contribute to the calculation result of the estimated equation for "Area 1 / Patient / Clinical" of the test type 702.

[0091] A specific brain area corresponds to a combination of test items 701. For example, the test type 703 "Area 1 / Patient / Non-clinical" corresponds to "TestA Item1", "TestA Item2", and so on. nFor the coefficient a specified in the lesion estimation formula table 205 n Applying this, Equation 1 is constructed.

[0092] For example, in the estimation equation for “Area 1 / Patient / Non-clinical” of test type 703, the variable x1 in equation 1 is given a variable based on the score of “TestA Item1”, and the coefficient a1 is the coefficient a specified by “TestA Item1” and “Area 1 / Patient / Non-clinical”. n Similarly, the variable x2 is given a variable based on the score of "TestA Item2", and the coefficient a2 is the coefficient a specified for "TestA Item2" and "Domain 1 / Patient / Non-Clinical". n The intercept a0 is the a0 specified in "Intercept" and "Domain 1 / Patient / Non-clinical". Note that the number of n in Equation 1 may differ depending on the estimation formula.

[0093] The contents of the lesion estimation formula table 205 are prepared by empirically obtaining estimation formulas based on diagnostic images of the brain identified by, for example, MRI and test results, and the contents of the lesion estimation formula table 205 are added based on estimation formulas developed by the formula development unit 206. Thus, the lesion estimation formula table 205 can be updated and added as estimation conditions change.

[0094] The granularity of the brain regions in Fig. 7 may be mixed. Conversion of lesion risk between granularities is possible by the method shown in Fig. 1. As can be seen from Figs. 6 and 7, in this embodiment, subject characteristics and examination sites are accepted as information for estimating the purpose, and tests and algorithms suited thereto can be selected.

[0095] 8 shows an example of the variable test estimation equation table 207 used in step S406. The variable test estimation equation table 207 shows an estimation equation for estimating the appropriate test item variable from the variables of other available test items. The estimation equation includes coefficients b of n independent variables for estimating m dependent variables. n,m Also, the intercept b0 of the equation is shown.

[0096] The variable test estimation formula table 207 is generated in advance and stored in a database by analyzing relationships, for example by evaluating the correlation between scores of different tests.

[0097] For example, in the lesion estimation formula table 205 in FIG. 7, the estimation formula for "Area 1 / Patient / Non-clinical" in the test type 703 requires the score of "TestA Item2". However, in the subject-environment-test table 203 in FIG. 6, for example, if TestA Item2 is not appropriate for "Patient / Non-clinical" due to test time restrictions, it is desirable not to use the score of "TestA Item2". Therefore, the score of an inappropriate test is substituted with the score of an appropriate test. For example, according to the variable test estimation formula table 207 (FIG. 8), "TestA Item2" (801) can be replaced with at least one of "TestA Item1", "TestC Item1", and "TestC Item2". It is desirable to make the estimation formula independent for each item (non-multivariate) for each test in order to increase the replacement rate of variables that cannot be used.

[0098] Now, let us change Equation 1 to the following Equation 2: n Let us represent this as "TestA Itemn".

[0099]

number

[0100] Here, "TestA Itemn" (801) is a variable obtained from the test item. Here, consider the case where TestA Item2 cannot be used in the second term in parentheses on the right side of Equation 2, "a2(TestA Item2)."

[0101] In this case, the replacement becomes Equation 3.

[0102]

number

[0103] The variable (TestA Item2) is (b TestA Item2, TestC Item1 From the table in Figure 8, the coefficient b for the conversion of TestC Item1 and TestA Item2 is TestA Item2, TestC Item1 and the intercept b0 can be obtained. This substitution formula is just one example, and other formulas may be used by using the correlation between scores obtained from different test items.

[0104] As described above, the estimation formula is corrected in step S407, and estimation becomes possible according to the flow in FIG. 4B.

[0105] 9 shows an example of the subject database 204. A subject ID 901, gender 902, and age 903 are recorded. If a subject has previously undergone an examination, the system environment 904 (clinical or non-clinical site) in which the examination was performed, examination date and time 905, lesion data (e.g., estimated results or imaging results) 906, 907 for each region, and items 908, 909, and 910 related to the test results (e.g., scores) are stored in the database. A subject whose test results are recorded can be treated as a registered subject in FIG. 5B, and the same test can be used in the next examination.

[0106] The subject database 204 is used to retrieve the subject's estimation history and collect subject data to develop new estimation equations in steps S408 and S409, if necessary.

[0107] Fig. 10 shows the process flow of the subject data selection process S408 by the formula development unit 206. The formula development unit 206 executes two main processes: data selection S408 and modeling S409. Fig. 10 shows the data selection process.

[0108] Data selection S408 and modeling S409 are performed when, first, variables (scores) that cannot be used in the selected estimating equation are used (S405) and the scores cannot be replaced (S406), and second, there are brain regions that cannot be covered by the selected estimating equation (S404).

[0109] In this case, the formula development unit 206 develops a new estimation formula for making the estimation using the variables available in the appropriate test list.

[0110] First, all subject data is checked in the subject database 204, and if the subject data does not contain variables (scores) related to the appropriate test list, the missing variables are estimated to the extent possible based on the variable test estimation equation table 207. To this end, the missing variables may be estimated from the variables included in the subject data by the method described in Fig. 8 (S1001).

[0111] Next, subject data is extracted that has all the variables (scores) associated with the appropriate test list (S1002).

[0112] Through calculations using the extracted subject data, the sparsity of the lesions in each of the subject data is evaluated (S1003), and the occurrence of the lesions is evaluated (S1004).

[0113] Based on the lesion sparsity and lesion occurrence data, a suitable dataset is extracted for developing an estimation model (S1005).

[0114] In order to create a model suited to the purpose of lesion estimation, subject data is selected according to the data characteristics and used to learn the estimation model. This improves the suitability of the model for the estimation purpose. The estimation purpose is determined according to the input in S401 (FIG. 4A) (will be described in detail later in FIG. 17).

[0115] Figure 11 illustrates the concept of lesion sparsity calculation. Sparsity indicates the density of lesions per region. In this calculation, two matrices are calculated: a distance matrix 1101 and a lesion occurrence matrix 1102. Both matrices have kxk arrays, where k is the number of brain regions (e.g., regions R1, R2, R3, R4, R5, etc.). The distance matrix 1101 and the lesion occurrence matrix 1102 are diagonally symmetric matrices.

[0116] Distance matrix 1101 indicates the distance between brain regions. Each element of distance matrix 1101 indicates the physical distance between regions (R1-R1, R1-R2, etc.). The values ​​are normalized, with the furthest distance being 1. That is, the physical distance can be defined by the distance between the centers of points of the regions, or the average distance between the voxel-based coordinates of the related regions. The voxel-based coordinates can be obtained, for example, via magnetic resonance imaging, computed tomography, and any imaging technique.

[0117] Lesion occurrence matrix 1102 indicates the occurrence or nonoccurrence of lesions in regions (for example, R1-R1, R1-R2, etc.). Each element of lesion occurrence matrix 1102 indicates the occurrence or nonoccurrence of lesions in two regions by using "1" when a lesion occurs in both regions (logical AND) and "NaN" when a lesion occurs in either region (logical OR).

[0118] The data is not limited to lesion distance and occurrence, but may also include lesion location, and the definition of interlesion occurrence may be weighted by lesion severity.

[0119] Next, an element-wise multiplication (Hadamard product; ○) of the distance matrix 1101 and the lesion occurrence matrix 1102 is performed to generate a product matrix 1103. The measure of lesion sparsity is quantified to a value in the range 0 to 1 by averaging over all non-NaN elements. This parameter represents the measure of lesion sparsity, i.e., how close or far the lesion regions are from each other. Thus, the sparsity of each subject's data is calculated.

[0120] 12 is a graph showing the relationship between the lesion occurrence probability (horizontal axis) and the lesion sparsity (vertical axis) of the subject data. The lesion occurrence probability 1201 shown on the horizontal axis is defined as the ratio of the number of lesion areas to the total number of areas. When a lesion occurs in all areas, the lesion occurrence probability is 1. When there is no lesion at all, the lesion occurrence probability is 0. The lesion sparsity 1202 shown on the vertical axis is obtained by the calculation in FIG. 11.

[0121] Since at least two lesioned regions are required to calculate sparsity, the sparsity cannot be 0. A high sparsity value (up to 1) indicates that the lesions that occur are far apart from each other.

[0122] The relationship between lesion occurrence and lesion sparsity is visualized within the range of lower bound 1203 and upper bound 1204. Each subject data is scattered within these ranges. As the probability of lesion occurrence increases, the distribution of the subject data sparsity narrows and converges around the medium sparsity (0.5).

[0123] The selection of subject data from the database for the purpose of developing a new lesion estimation equation is performed taking into consideration its suitability for the model. The suitability for the model can be evaluated, for example, by two parameters 1200: the probability of lesion occurrence 1201 and the sparsity 1202 of the lesion.

[0124] The subject data sets may be classified and selected based on criteria such as regions of low occurrence-low sparsity 1205, regions of medium occurrence-medium sparsity 1206, regions of high occurrence-medium sparsity 1207, etc.

[0125] The classification of the regions of the subject data set can be performed by a known unsupervised clustering method. Each region may be a separate region 1205, 1206 or may be an overlapping region 1206, 1207.

[0126] For example, in Figure 6, TestA, which is suitable for screening purposes, is selected for "Clinical" under "Normal / Asymptomatic". In order to estimate the pathology risk of asymptomatic people, the appropriateness of the model is improved by using subject data with a low incidence of pathology. Therefore, when "Clinical" is selected under "Normal / Asymptomatic" in S401 (Figure 4A), the low incidence-low to high sparsity region 1208 is selected.

[0127] In addition, in the case of "history of stroke" and "clinical" in Figure 6, TestB, which is suitable for estimating the risk around the gyrus with lesions with high resolution, is selected. The subject dataset with high incidence of lesions and medium sparsity is suitable for estimating the risk of the gyrus with lesions. In this case, the region 1209 with high incidence and medium sparsity is selected.

[0128] As described above, in this embodiment, subject data is classified based on two parameters: the occurrence of a lesion and the sparseness (sparsity) of the lesion.

[0129] 13 is a diagram showing a process flow S409 of mathematical modeling by the mathematical formula development unit 206. In this process, a model (corresponding to Equation 1) for deriving the probability of a lesion risk from the score is created.

[0130] From the dataset with a given occurrence probability and sparsity extracted by the principle of Figure 12, the lesion data for each subject (dependent variable) and the independent variable x based on the score of the appropriate test list are n The lesion data includes the presence or absence of a lesion, the level of the lesion, the brain region of the lesion, and the like.

[0131] As explained above, the mathematical modeling process flow S409 is executed when, first, variables (scores) that cannot be used in the selected estimation formula are used (S405) and the scores cannot be replaced (S406), or when, second, there is a brain area that cannot be covered by the selected estimation formula (S404).

[0132] Since an estimated formula is prepared for each brain region, in the first case, an estimated formula is developed by focusing on the brain region targeted by the estimated formula that cannot be used as a result of the inability to prepare variables, whereas in the second case, an estimated formula is developed by focusing on the uncovered brain region.

[0133] Variables from the appropriate test list in Equation 1 are trained to correspond to the lesion status of each region (S1302).

[0134] A training process is performed to model the probability of pathology risk from the scores, for example in the form of Equation 1. For example, the input for learning is the independent variables x n and lesion class (dependent variable: absent vs. present or low vs. high).

[0135] In training, for example, if there is a lesion in a brain region, the values ​​of each variable are learned when the coefficients are fixed. It is also possible to train the coefficients, but it should be noted that the coefficients may be 0 in the estimation formula.

[0136] The trained model is verified (S1303) and evaluated (S1304). For verification, a dataset different from that used for training is usually used. The evaluation items may be evaluation of desired specifications such as estimation accuracy, specificity, or detection sensitivity.

[0137] Next, the performance of the model (eg, estimation accuracy, specificity, or sensitivity) is checked to see if the performance is satisfactory (S1305).

[0138] According to the above process flow, the data used for modeling is selected from a data set with a predetermined incidence rate and sparsity, so that a model according to the purpose is generated. The purpose is set according to the input in S401 (e.g., emphasis on accuracy or sensitivity). As described above, the estimation formula for each region can optimize the estimation performance (accuracy, specificity, sensitivity, etc.).

[0139] 14 is an example of a user interface developed by the test development unit 209. The user interface operates to input test variables or scores that are included in the estimation formula 1400.

[0140] In this user interface, the input items for a test can be one or more variables. After the subject performs the items related to a particular test, the variable values ​​1401, 1402, 1403, 1404 can be entered. The inputs can be made directly by a user (such as a medical professional) or the inputs can be converted by the system according to the subject's input. The estimation process can then be executed by pressing the "Estimate" button 1405.

[0141] The estimation environment user interface has appropriate variable entries. The variables correspond to items per test. It is not necessary to run all tests. This can reduce the estimation time.

[0142] FIG. 15 shows the estimation process by the estimation resolution unit 302.

[0143] First, the unit receives variable input from the user interface 1400 (FIG. 14) (S1501).

[0144] The variables are applied to the selected, and if necessary modified or developed, estimation risk is calculated for each brain region (S1502).

[0145] Thereafter, the estimated resolution is determined according to the input that has been set. The resolution can be appropriately specified by the user. It can also be determined in advance, for example, according to the subject classification 508. For example, high resolution is set for those with a medical history 508a, and low resolution for those who are normal / asymptomatic 508b (S1503). The adjustment of the lesion risk according to the resolution is performed as described in FIG. 1.

[0146] The estimation result is displayed on the display unit 210 (S1504).

[0147] Depending on the subject target and the purpose of the estimation, the resolution of the estimation can be appropriate. For example, in normal / asymptomatic 508b, recognition of the onset of a pathology is a priority. Thus, estimation at a hemispheric or lobar resolution may be sufficient. On the other hand, a higher estimation resolution (e.g., at the lobar or gyral level) is required to aid in the rehabilitation of the patient.

[0148] 16A and 16B show examples of displaying the estimation results. As for display types, a user interface 1600a (FIG. 16A) for estimation results for a new subject and a user interface 1600b (FIG. 16B) for estimation results for a registered subject are prepared.

[0149] 16A, now for a new subject, the only result displayed will be the current estimate 1601. The estimate results are visualized by region, either at the gyrus, lobe, or hemisphere level.

[0150] In this example, an estimate of damaged leaves is displayed. The degree of disease risk (e.g., from low to high) is indicated by a color map 1602.

[0151] Areas that are not stained (parietal, occipital, cerebellar lobes, etc.) do not indicate a risk of lesion. In this example, the risk of lesion in the frontal, temporal, and sublobes increases gradually (1603<1604<1605).

[0152] Referring to FIG. 16B, if the subject is a previously enrolled subject, the current estimate 1601 and the previous estimate 1606 are displayed.

[0153] In this example, the risk in the frontal lobe decreases (1607>1603), the risk in the sublobe increases (1609<1605), and the risk in the temporal lobe remains the same (1608=1604).

[0154] The results of the pathology risk estimation are displayed with different degrees of estimated probability. Furthermore, the estimation results can be visualized with different spatial resolutions depending on what has been determined.

[0155] Fig. 17 is a table showing an example of the conditions of the subject target, the site of use, the purpose, the appropriate estimation model, and the data set appropriate for learning the estimation model in this embodiment. Although Fig. 17 also shows the conditions of the site of use and the purpose, in the disease risk estimation, it is not necessary to input the site of use and the purpose, and it is sufficient to input the subject target.

[0156] For example, when the subject target is "patient / clinical", it is assumed that the patient is examined in a rehabilitation facility or hospital. In this case, typical applications and purposes are to measure the rehabilitation effect or identify the lesion site, and the performance of the estimation model required for this is high resolution and high accuracy. Therefore, when "patient / clinical" is input in S401 of FIG. 4A, in S402, a test list that can obtain high resolution and high accuracy results is selected in the subject-environment-test table 203. In S403, an estimation formula in which the subject target is "patient / clinical" for each brain site is selected from the lesion estimation formula table 205. At this time, an estimation formula that can use the score obtained from the selected test as much as possible is selected. As described above, the estimation algorithm selection unit (also shown as the estimation formula selection unit 202) may use the table shown in FIG. 17 when determining the required resolution and estimation performance from the purpose of estimation. Here, it is necessary to cover all areas of the brain for the examination, but when the variables used in the selected estimation formula cannot be obtained from the selected test, or when an estimation formula that covers all brain areas has not been obtained, an estimation formula using variables obtained from the selected test or variables that can be converted from it is developed. When developing the estimation equation, a dataset with high pathology incidence and moderate sparseness is extracted from the subject database 204 for use in a high-resolution, highly accurate model.

[0157] As explained above, in the past, it was difficult to obtain data from patients with appropriate lesion distribution in all brain regions, and the estimation ability for detecting true lesions was insufficient, resulting in false negative results. In addition, there was a problem that the burden on subjects increased when attempting to satisfy both resolution and sensitivity with high quality.

[0158] According to this embodiment, it is possible to select a test and an algorithm having a resolution and detection sensitivity suitable for the inspection purpose. Furthermore, if the variables obtained from the test do not match the variables used in the algorithm, the variables can be converted to variables obtained from another test or a new algorithm can be developed. This makes it possible to perform an inspection suited to the inspection purpose. [Explanation of symbols]

[0159] An input receiving unit 201, an estimated formula selection unit 202, a subject-environment-test table 203, a subject database 204, and a lesion estimated formula table 205, a formula development unit 206, a variable test estimated formula table 207, a formula input extraction unit 208, a test development unit 209, and a display unit 210.

Claims

1. An information processing device for selecting a test and an algorithm to be used for estimating a risk of a lesion in a subject's brain from among previously prepared tests and algorithms, an input receiving unit for receiving an input of a purpose of the estimation or information for estimating the purpose of the estimation; an estimation algorithm selection unit that selects an algorithm to be used for the estimation based on an input from the input receiving unit, and selects a test for obtaining a variable reflecting a brain state of the subject based on the input from the input receiving unit; An information processing device comprising:

2. The estimation algorithm selection unit is selecting the test based on the input of the input receiving unit, and then selecting an algorithm based on the input of the input receiving unit that uses variables resulting from the selected test.

2. The information processing device according to claim 1.

3. The algorithm is prepared in advance for each brain region, the estimation algorithm selection unit determines the required resolution and estimation performance from the purpose of the estimation and selects an algorithm for every region of the brain; 3. The information processing device according to claim 2.

4. The estimation algorithm selection unit is if all of the variables used by the selected algorithm are not available from the selected test, replacing the variables not available from the selected test with variables available from the selected test; 4. The information processing device according to claim 3.

5. a variable test estimation equation table storing rules for converting variables obtained from a first test into variables obtained from a second test, and the estimation algorithm selection unit refers to the variable test estimation equation table to replace variables not obtained from a selected test with variables obtained from the selected test; 5. The information processing device according to claim 4.

6. It also has a formula development unit. When the estimation algorithm selection unit selects an algorithm using variables obtained from the selected test, but is unable to select an algorithm for all brain regions, or when the estimation algorithm selection unit converts variables obtained from the selected test into variables used by the selected algorithm, but is unable to obtain all of the variables used by the selected algorithm, the formula development unit newly develops an algorithm using variables obtained from the selected test.

6. The information processing device according to claim 5.

7. The formula development unit Refer to a subject database that stores the subject's past data, such as the brain state of the subject and variables reflecting the brain state of the subject obtained by the test; extracting a data set including variables from the selected tests from the subject database based on incidence and sparsity of pathology; Develop new algorithms using the extracted data set; 7. The information processing device according to claim 6.

8. In addition, we have a test development unit. the test development unit creates an interface for obtaining variables reflecting the brain state of the subject to be input into the algorithm based on the selected algorithm; 2. The information processing device according to claim 1.

9. The input receiving unit receives a subject's characteristics and an examination site as information for estimating the purpose.

2. The information processing device according to claim 1.

10. A method for selecting a test and an algorithm to be used for estimating a risk of a lesion in a subject's brain from among pre-prepared tests and algorithms using an information processing device, comprising: The information processing device includes an input receiving unit and an estimation algorithm selecting unit; A first step in which the input receiving unit accepts input of the purpose of the estimation or information for estimating the purpose of the estimation; a second step in which the estimation algorithm selection unit selects an algorithm to be used for the estimation based on the input, and selects a test for obtaining a variable reflecting a brain state of the subject based on the input; A brain lesion estimation method to perform.

11. The second step includes: selecting the test based on the inputs; and then selecting an algorithm based on the inputs that uses variables from the selected test. The method for estimating a brain lesion according to claim 10.

12. The algorithm is prepared in advance for each brain region, The second step is to select an algorithm for every region of the brain. The method for estimating a brain lesion according to claim 11.

13. If in the second step, the estimation algorithm selection unit performs a third step of replacing variables not obtained from the selected test with variables obtained from the selected test, the variables used by the selected algorithm being not obtained from the selected test; In the third step, converting the variables resulting from the selected test into variables used by the selected algorithm by referring to a variable test estimator table, the variable test estimator table storing rules for converting variables resulting from a first test into variables resulting from a second test; The method for estimating a brain lesion according to claim 12.

14. The information processing device further comprises a formula development unit, If in the second step an algorithm is selected that uses variables obtained from the selected test, but it is not possible to select an algorithm for all brain regions, or if in the second step variables obtained from the selected test are converted into variables used by the selected algorithm, but it is not possible to obtain all variables used by the selected algorithm, a fourth step in which the formula development unit refers to a subject database storing the subject's past data, which includes the brain state of the subject and variables reflecting the brain state of the subject obtained by the test, and extracts a data set including variables obtained from the selected test from the subject database based on the incidence rate and sparsity of a lesion, and uses the extracted data set to newly develop an algorithm; The method for estimating a brain lesion according to claim 13.

15. The lesion incidence and sparsity conditions are set based on the input. The method for estimating a brain lesion according to claim 14.

Images