Disease diagnosis result determination device, disease diagnosis result determination method, and program

Abstract

This disease diagnosis result determination device is provided with: a brain examination result determination unit that determines a result of brain examination concerning a second subject on the basis of medical checkup data concerning the second subject and a first learning result obtained by performing learning in advance of a relationship between medical checkup data concerning a first subject and a result of brain examination concerning the first subject; and a disease diagnosis result determination unit that determines a disease diagnosis result concerning the second subject on the basis of a second learning result obtained by performing learning in advance of a relationship between a result of brain examination concerning a third subject and a disease diagnosis result concerning the third subject, and the result of brain examination concerning the second subject, determined by the brain examination result determination unit.

Description

Technical Field

[0001] The present invention relates to a disease diagnosis result determination device, a disease diagnosis result determination method, and a program. This application claims priority based on Japanese Patent Application No. 2020-171329 filed in Japan on October 9, 2020, and incorporates its content herein by reference.

Background Art

[0002] With the progress of aging, the number of dementia patients has increased rapidly and become a social problem. Early detection of cognitive impairment is important for the treatment and prevention of dementia, but in many cases, dementia is diagnosed after it has progressed. One of the reasons for this is that a large-scale screening test method for cognitive impairment has not been established. Currently, interview-type tests such as MMSE are used for screening tests. Since these interview-type tests are subjective examination methods, they need to be conducted one-on-one. Therefore, it is not possible to perform large-scale screening in a short time with these interview-type tests. Although imaging tests such as MRI have high diagnostic ability for dementia, they are not suitable for screening due to their high cost.

[0003] Conventionally, as a method for determining the risk of cognitive impairment, a method of determining the risk of cognitive impairment based on general blood test data and age using machine learning has been proposed (Patent Documents 1 and 2). In the methods for determining the risk of cognitive impairment described in Patent Documents 1 and 2, the relationship between age, the test results of general blood tests, and the score of the Mini-Mental State Examination (MMSE), which is an interview-type test of cognitive function, is learned by machine learning, and the MMSE score is determined based on age and the test results of general blood tests. In Patent Document 2, deep learning is used as machine learning, the age and the test results of general blood tests are input into the input layer, and the MMSE score is output from the output layer.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

[0005] In the risk determination methods for cognitive impairment described in Patent Documents 1 and 2, there are the following problems. When using deep learning as machine learning, the types of data (age, blood test data) input to the input layer of the neural network used in deep learning and the types of data (cognitive function evaluation by an interview-style test such as MMSE) output from the output layer are limited, and the prediction accuracy of cognitive impairment is low. An interview-style test such as MMSE is a subjective evaluation method, and there is a possibility that the evaluation of cognitive function is inaccurate. Also, in conventional deep learning, the process of predicting the data output from the output layer from the data input to the input layer is a black box, which has been an obstacle when applying deep learning to medicine. It has been required to be able to improve the determination accuracy of the diagnosis result of a disease.

[0006] The present invention has been made in view of the above points, and provides a disease diagnosis result determination device, a disease diagnosis result determination method, and a program that can improve the determination accuracy of the diagnosis result of a disease. [Means for Solving the Problems]

[0007] The present invention has been made to solve the above problems, and one aspect of the present invention is a first learning result in which the relationship between the health examination data regarding the first subject and the result of the brain examination regarding the first subject has been learned in advance, and based on the health examination data regarding the second subject, a brain examination result determination unit that determines the result of the brain examination regarding the second subject, a second learning result in which the relationship between the result of the brain examination regarding the third subject and the diagnosis result of the disease regarding the third subject has been learned in advance, and based on the result of the brain examination regarding the second subject determined by the brain examination result determination unit, a disease diagnosis result determination unit that determines the diagnosis result of the disease regarding the second subject, and a disease diagnosis result determination device comprising the same.

[0008] Further, one aspect of the present invention is that in the above disease diagnosis result determination device, the result of the brain examination consists of results of a plurality of types of brain examinations, the first learning result consists of a plurality of types of learning results corresponding to each of the plurality of types of brain examinations, and the brain examination result determination unit determines the results of the plurality of types of brain examinations regarding the second subject based on the plurality of types of learning results corresponding to each of the plurality of types of brain examinations and the health examination data regarding the second subject.

[0009] Further, one aspect of the present invention is that in the above disease diagnosis result determination device, the device further comprises a brain examination result acquisition unit that acquires the brain examination results obtained by previously performing some of the plurality of types of brain examinations regarding the second subject, and the disease diagnosis result determination unit performs the determination using the brain examination results acquired by the brain examination result acquisition unit instead of the results of the brain examinations regarding the second subject determined by the brain examination result determination unit for some of the plurality of types of brain examinations.

[0010] Further, one aspect of the present invention is that in the above disease diagnosis result determination device, the brain examination result determination unit determines the result of the brain examination regarding the second subject based on the first learning result and the health examination data regarding the second subject, using the statistical amounts of the health examination data regarding a plurality of fourth subjects as a part of the health examination data regarding the second subject.

[0011] Also, in one aspect of the present invention, in the above-described disease diagnosis result determination device, after the disease diagnosis result determination unit determines the diagnosis result, a brain examination result output unit that outputs a pair of the result of the brain examination regarding the second subject used for determining the diagnosis result and the diagnosis result is further provided.

[0012] Also, in one aspect of the present invention, based on a first learning result in which the relationship between the health examination data regarding the first subject and the result of the brain examination regarding the first subject has been learned in advance, and the health examination data regarding the second subject, a brain examination result determination step of determining the result of the brain examination regarding the second subject, a second learning result in which the relationship between the result of the brain examination regarding the third subject and the diagnosis result of the disease regarding the third subject has been learned in advance, and based on the result of the brain examination regarding the second subject determined by the brain examination result determination step, a disease diagnosis result determination step of determining the diagnosis result of the disease regarding the second subject, is a disease diagnosis result determination method.

[0013] Also, in one aspect of the present invention, a program for causing a computer to execute a brain examination result determination step of determining the result of the brain examination regarding the second subject based on a first learning result in which the relationship between the health examination data regarding the first subject and the result of the brain examination regarding the first subject has been learned in advance, and the health examination data regarding the second subject, a second learning result in which the relationship between the result of the brain examination regarding the third subject and the diagnosis result of the disease regarding the third subject has been learned in advance, and a disease diagnosis result determination step of determining the diagnosis result of the disease regarding the second subject based on the result of the brain examination regarding the second subject determined by the brain examination result determination step.

Advantages of the Invention

[0014] According to the present invention, the determination accuracy of the disease diagnosis result can be improved.

Brief Description of the Drawings

[0015]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Mode for Carrying Out the Invention

[0016] (Embodiment) Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 1 is a diagram showing an example of the configuration of a disease diagnosis result determination device 1 according to the present embodiment. When health examination data is input, the disease diagnosis result determination device 1 determines a diagnosis result of cognitive impairment based on first machine learning and second machine learning. In the present embodiment, the first machine learning and the second machine learning are deep learning as an example. In the disease diagnosis result determination device 1, a so-called two-stage deep learning similar to a medical diagnosis process is used. In the first machine learning, based on the relationship between systemic disorders and brain disorders, the results of brain examination methods (such as cognitive tests like the Mini Mental State Examination (MMSE), MRI (Magnetic Resonance Imaging) image diagnosis, etc.) are predicted from the health examination data. Next, in the second machine learning, a determination of the risk of cognitive impairment is made so that a doctor can integrally diagnose dementia from the results of the brain examination methods.

[0017] The disease diagnosis result determination device 1 includes a first learning result acquisition unit 2, a health examination data acquisition unit 3, a brain examination result determination unit 4, a second learning result acquisition unit 5, and a disease diagnosis result determination unit 6. The disease diagnosis result determination device 1 is, as an example, a personal computer (PC). Note that the disease diagnosis result determination devices 1, 1a, and 1b may be mobile terminal devices such as smartphones.

[0018] The first learning result acquisition unit 2 acquires a first learning result L1 supplied from the first learning result supply unit 7. The first learning result L1 is a result obtained by previously learning, by first machine learning, the relationship between the health examination data regarding the first subject P1 and the result of the brain examination regarding the first subject P1. The first machine learning is, as an example, deep learning. The first learning result L1 is the parameters (weights and biases) of the neural network used for the deep learning.

[0019] In this embodiment, the results of the brain examination consist of the results of multiple types of brain examinations. The first learning result L1 consists of one or more learning results corresponding to each of the multiple types of brain examinations. The multiple types of learning results are referred to as brain examination learning results LB-i (i = 1, 2, ···, n: n is the number of types of brain examinations). The results of the brain examination are, for example, the results of an interview-style test related to cognitive function such as MMSE, indicators related to image diagnosis (MRI, CT, PET, etc.) (indicators by VSRAD, BHQ (Brain Healthcare Quotient), etc.), lesion findings (cerebral infarction, brain atrophy, etc.), and one or more of dementia biomarkers (amyloid-β, etc.).

[0020] In this embodiment, as an example, the results of the brain examination are three: the score of MMSE, the GM-BHQ (Grey Matter Brain Healthcare Quotient) which is an indicator showing the volume of the cerebral gray matter, and the FA-BHQ (Fractional Anisotropy Brain Healthcare Quotinet) which is an indicator representing the degree of integration of nerve fibers in the cerebral white matter as the anisotropy ratio. The brain examination learning result LB-1 is the learning result corresponding to the score of MMSE. The brain examination learning result LB-2 is the learning result corresponding to GM-BHQ. The brain examination learning result LB-3 is the learning result corresponding to FA-BHQ.

[0021] The health examination data regarding the first subject P1 is also referred to as teacher health examination data. Also, the results of the brain examination regarding the first subject P1 are also referred to as teacher brain examination results. The first subject P1 is a subject for collecting teacher health examination data and teacher brain examination results. The first subject P1 is preferably a subject of a predetermined number or more in order to improve the determination accuracy by machine learning, but one or more subjects are sufficient.

[0022] The health examination data may be any of the examination items of various health examinations (medical check-ups) such as specific health examinations. Also, the health examination data may be data that can be obtained individually by the subject rather than at various medical institutions, such as data measured by a healthcare monitoring device or a body composition analyzer. The health examination data includes, for example, any one or more of age, the results of a general blood test, the measurement results of physical functions (any one or more of height, weight, abdominal circumference, BMI, blood pressure, pulse, etc.), medical history (any one or more of medication history, smoking habit), family history, information indicating subjective and objective symptoms (any one or more of physical findings, neurological findings, etc.), the examination results of physiological functions (any one or more of electrocardiogram, electroencephalogram, auditory examination, balance sense examination, visual examination, taste examination, etc.), the examination results of psychological functions (any one or more of intelligence quotient, state-trait anxiety inventory, etc.), other biological information (any one or more of fundus examination, digestive system examination, dental examination, speech and language, behavioral analysis, etc.), and information indicating traditional Chinese medical findings (any one or more of tongue diagnosis, pulse diagnosis, palpation, etc.).

[0023] The health examination data acquisition unit 3 acquires the health examination data MC supplied from the health examination data supply unit 9. The health examination data MC is the health examination data regarding the second subject P2. The second subject P2 is the subject whose disease diagnosis result is to be determined by the disease diagnosis result determination device 1. The data items included in the health examination data regarding the second subject P2 are the same as the data items included in the health examination data regarding the first subject P1 described above. For example, if the health examination data regarding the first subject P1 is a set of the age regarding the first subject P1, the results of a general blood test, and the results of an electroencephalogram examination, the health examination data regarding the second subject P2 is a set of the age regarding the second subject P2, the results of a general blood test, and the results of an electroencephalogram examination.

[0024] The brain examination result determination unit 4 determines the brain examination result determination result B based on the first learning result L1 and the health examination data MC. The brain examination result determination result B is the result of the brain examination regarding the second subject P2. As described above, since the first learning result L1 is the learning result based on the first machine learning, the brain examination result determination unit 4 makes a determination based on the first machine learning.

[0025] Here, the brain examination result determination unit 4 determines the results of multiple types of brain examinations for the second subject P2 based on the brain examination learning results LB-i, which are multiple types of learning results corresponding to the respective multiple types of brain examinations, and the health examination data MC. The brain examination result determination result B includes multiple brain examination result determination results B-i (i = 1, 2, ···, n: n is the number of types of brain examinations) corresponding to the respective multiple types of brain examinations. In this embodiment, the brain examination result determination result B consists of the brain examination result determination result B-1, which is the determination result of the MMSE score, the brain examination result determination result B-2, which is the determination result of GM-BHQ, and the brain examination result determination result B-3, which is the determination result of FA-BHQ.

[0026] The second learning result acquisition unit 5 acquires the second learning result L2 supplied from the second learning result supply unit 8. The second learning result L2 is a result obtained by pre-learning, by second machine learning, the relationship between the results of brain examinations for the third subject P3 and the diagnosis results of diseases for the third subject P3. The second machine learning is, for example, deep learning. The second learning result L2 is the parameters (weights and biases) of the neural network used for the deep learning.

[0027] The results of the brain examinations for the third subject P3 are also referred to as teacher brain examination results. Also, the diagnosis results of diseases for the third subject P3 are also referred to as teacher disease diagnosis results. The third subject P3 is a subject for collecting the teacher brain examination results and the teacher disease diagnosis results. The third subject P3 is preferably a subject of a predetermined number or more in order to improve the determination accuracy by machine learning, but may be one or more subjects. The teacher disease diagnosis result is the diagnosis result (cognitive impairment) of the disease diagnosed by a specialist from the results of the brain examinations for the third subject P3. The diagnosis result of the disease is classified into one of the classes of "normal", "mild cognitive impairment (MCI)", and "dementia" as an example.

[0028] The disease diagnosis result determination unit 6 determines the disease diagnosis result D based on the second learning result L2 and the brain examination result determination result B determined by the brain examination result determination unit 4. The disease diagnosis result D is the diagnosis result of the disease for the second subject P2. As described above, since the second learning result L2 is the learning result based on the second machine learning, the disease diagnosis result determination unit 6 makes a determination based on the second machine learning.

[0029] The first learning result supply unit 7 supplies the first learning result L1 to the disease diagnosis result determination device 1. The second learning result supply unit 8 supplies the second learning result L2 to the disease diagnosis result determination device 1. The health check data supply unit 9 supplies the health check data MC to the disease diagnosis result determination device 1. Each of the first learning result supply unit 7, the second learning result supply unit 8, and the health check data supply unit 9 may be, for example, a human interface device such as a keyboard, a tablet, or a scanner, or an information storage device such as a server, or a mobile terminal device such as a smartphone.

[0030] Note that a large number of cases in which health check data, brain examination results such as MRI, and diagnoses by specialists are set are collected from medical institutions and used as teacher health check data and teacher brain examination results for the first subject P1, and teacher brain examination results and teacher disease diagnosis results for the third subject P3.

[0031] The presentation unit 10 is, for example, a display or a printer, and presents the disease diagnosis result D output from the disease diagnosis result determination unit 6 provided in the disease diagnosis result determination device 1 by presentation means such as display or printing. Note that the presentation unit 10 may be a storage device such as a network server. In this case, the presentation unit 10 stores the disease diagnosis result D supplied from the disease diagnosis result determination unit 6 and supplies the stored disease diagnosis result D to other devices.

[0032] FIG. 2 is a diagram showing an example of the configuration of a neural network used for machine learning according to the present embodiment. The first neural network N1-1, the first neural network N1-2, and the first neural network N1-3 are neural networks for deep learning used for the first machine learning, respectively. The first machine learning is executed with the first neural network N1-1, the first neural network N1-2, and the first neural network N1-3 used in parallel. The second neural network N2 is a neural network for deep learning used for the second machine learning. In the disease diagnosis result determination device 1, the machine learning is executed in two stages of the first machine learning and the second machine learning. The second machine learning is executed using the output result of the first machine learning as an input. That is, the first machine learning and the second machine learning are executed in series.

[0033] Deep learning is a machine learning method using a multi-layer neural network (a neural network having two or more hidden layers). The first neural network N1-1, the first neural network N1-2, the first neural network N1-3, and the second neural network N2 are neural networks called feedforward networks, respectively. A feedforward network is a neural network in which information propagates in one direction from an input layer to an output layer.

[0034] When the health check data MC is input into the input layer of the first neural network N1-1, the first neural network N1-2, and the first neural network N1-3 respectively, the brain examination result determination result B is output from the output layer respectively. In the present embodiment, the configurations of the first neural network N1-1, the first neural network N1-2, and the first neural network N1-3 are common except for the number of units provided in the output layer. In the first neural network N1-1, the first neural network N1-2, and the first neural network N1-3, as an example, the number of hidden layers is 2 layers respectively, and the number of units in each hidden layer is 400. Note that the number of hidden layers and the number of units in each hidden layer may be different among the first neural network N1-1, the first neural network N1-2, and the first neural network N1-3.

[0035] The number of units provided in the input layer of each of the first neural network N1-1, the first neural network N1-2, and the first neural network N1-3 is determined according to the number of types of data included in the health check data MC and the representation of those data. Note that the data included in the health check data MC is represented as a numerical value or a value of a class classified according to the type.

[0036] The parameters (weights, biases) of the first neural network N1-1 are determined by performing learning based on the brain examination learning result LB-1. When the health check data MC is input into the input layer of the first neural network N1-1, the score of MMSE is output from the output layer. The output layer of the first neural network N1-2 includes the number of units corresponding to the representation of the score of MMSE. The score of MMSE is represented by a number from 0 to 30, for example. Accordingly, the output layer of the first neural network N1-1 includes 31 units.

[0037] The first neural network N1-2 has been determined by performing learning based on the brain examination learning result LB-2. When the health examination data MC is input to the input layer of the first neural network N1-2, GM-BHQ is output from the output layer. The output layer of the first neural network N1-2 includes a number of units corresponding to the representation of GM-BHQ.

[0038] The first neural network N1-3 has been determined by performing learning based on the brain examination learning result LB-3. When the health examination data MC is input to the input layer of the first neural network N1-3, FA-BHQ is output from the output layer. The output layer of the first neural network N1-3 includes a number of units corresponding to the representation of FA-BHQ. Note that the respective values of GM-BHQ and FA-BHQ are values corresponding to age. The respective values of GM-BHQ and FA-BHQ tend to decrease with age, and the average value for each age is 100. If the respective values of GM-BHQ and FA-BHQ are lower than the values corresponding to each age, it indicates a low brain health level. The values of GM-BHQ and FA-BHQ output from the output layer are respectively expressed as values corresponding to the age of the second subject P2.

[0039] The parameters (weights, biases) of the second neural network N2 have been determined by performing learning based on the second learning result L2. When the brain examination result determination result B is input to the input layer of the second neural network N2, the disease diagnosis result D is output from the output layer.

[0040] The outputs from the output layers of the first neural network N1-1, the first neural network N1-2, and the first neural network N1-3 are directly input to the input layer of the second neural network N2. Therefore, the number of units provided in the input layer of the second neural network N2 is equal to the total number of units provided in the input layers of the first neural network N1-1, the first neural network N1-2, and the first neural network N1-3.

[0041] The output layer of the second neural network N2 includes a number of units corresponding to the expression of the disease diagnosis result D. As described above, in this embodiment, since the disease diagnosis result D is classified and expressed into three classes, the output layer of the second neural network N2 includes three units. In the second neural network N2, as an example, the number of hidden layers is two layers, and the number of units in each hidden layer is 400. The configuration of the hidden layer of the second neural network N2 may be the same as or different from the configuration of the hidden layer of each of the first neural networks N1-1, N1-2, and N1-3.

[0042] In this embodiment, the types and values of the teacher's medical examination data are common among the brain examination learning results LB-i. That is, in this embodiment, the brain examination learning results LB-i are generated using common teacher's medical examination data among the brain examination learning results LB-i.

[0043] Note that the brain examination learning results LB-i may be respectively generated using different teacher's medical examination data among the brain examination learning results LB-i. For example, different values of the teacher's medical examination data may be used among the brain examination learning results LB-i to perform the first machine learning, and the brain examination learning results LB-i may be respectively generated. Also, among the brain examination learning results LB-i, the types or the number of types of the teacher's medical examination data may be different. Also, among the brain examination learning results LB-i, the number of pairs of the teacher's medical examination data and the teacher's brain examination results may be different.

[0044] For example, in the brain examination learning result LB-1 corresponding to the MMSE score, age and general blood test results are used as teacher health examination data, and 200 data may be used. In the brain examination learning result LB-2 corresponding to GM-BHQ, height, medication history, and electrocardiogram are used as teacher health examination data, and 1200 data may be used. In the brain examination learning result LB-3 corresponding to FA-BHQ, weight, abdominal circumference, smoking habit, and taste test are used, and 500 data may be used.

[0045] When the types or the number of types of teacher health examination data are different among the brain examination learning results LB-i, the number of units provided in the input layer of each of the first neural network N1-1, the first neural network N1-2, and the first neural network N1-3 differs according to the types or the number of types of teacher health examination data.

[0046] FIG. 3 is a diagram showing an example of a determination process by the disease diagnosis result determination device 1 according to the present embodiment. Step S10: The first learning result acquisition unit 2 acquires the first learning result L1 supplied from the first learning result supply unit 7. The first learning result acquisition unit 2 supplies the acquired first learning result L1 to the brain examination result determination unit 4.

[0047] Step S20: The health examination data acquisition unit 3 acquires the health examination data MC supplied from the health examination data supply unit 9. The health examination data acquisition unit 3 supplies the acquired health examination data MC to the brain examination result determination unit 4.

[0048] Step S30: The brain examination result determination unit 4 determines the brain examination result determination result B based on the first learning result L1 acquired by the first learning result acquisition unit 2 and the health examination data MC acquired by the health examination data acquisition unit 3. The brain examination result determination unit 4 supplies the determined brain examination result determination result B to the disease diagnosis result determination unit 6.

[0049] Step S40: The second learning result acquisition unit 5 acquires the second learning result L2 supplied from the second learning result supply unit 8. The second learning result acquisition unit 5 supplies the acquired second learning result L2 to the disease diagnosis result determination unit 6.

[0050] Step S50: The second learning result acquisition unit 5 acquires the brain examination result determination result B determined by the brain examination result determination unit 4.

[0051] Step S60: The disease diagnosis result determination unit 6 determines the disease diagnosis result D based on the second learning result L2 acquired by the second learning result acquisition unit 5 and the brain examination result determination result B determined by the brain examination result determination unit 4. Step S70: The disease diagnosis result determination unit 6 outputs the determined disease diagnosis result D to the presentation unit 10. With the above, the disease diagnosis result determination device 1 ends the determination process.

[0052] Note that when a part of the health check data MC cannot be acquired, the brain examination result determination unit 4 may perform the determination using the statistical amounts of the health check data regarding a plurality of fourth subjects P4. That is, when a part of the health check data MC cannot be acquired, the brain examination result determination unit 4 may use the statistical amounts of the health check data regarding a plurality of fourth subjects P4 as a part of the health check data MC, and determine the brain examination result determination result B based on the first learning result L1 and the health check data MC. The statistical amount of the health check data is, for example, the average value, the median value, etc. The fourth subject P4 is a subject who provides the health check data for interpolating the health check data MC when a part of the health check data MC is missing.

[0053] Note that in this embodiment, as an example, an example in the case where the disease is a cognitive disorder has been described, but it is not limited thereto. As the disease, in addition to the cognitive disorder, the diagnosis results of brain diseases such as depression may be determined. Further, the disease diagnosis result determination device 1 may perform the determination of the diagnosis results of diseases other than brain diseases.

[0054] As described above, the disease diagnosis result determination device 1 according to the present embodiment includes a brain examination result determination unit 4 and a disease diagnosis result determination unit 6. Based on the first learning result L1 in which the relationship between the health examination data regarding the first subject P1 and the result of the brain examination regarding the first subject P1 has been learned in advance, and the health examination data MC regarding the second subject P2, the brain examination result determination unit 4 determines the brain examination result regarding the second subject P2 (in the present embodiment, the brain examination result determination result B). Based on the second learning result L2 in which the relationship between the result of the brain examination regarding the third subject P3 and the diagnosis result of the disease regarding the third subject P3 has been learned in advance, and the result of the brain examination regarding the second subject P2 determined by the brain examination result determination unit 4 (in the present embodiment, the brain examination result determination result B), the disease diagnosis result determination unit 6 determines the disease diagnosis result regarding the second subject P2 (in the present embodiment, the disease diagnosis result D).

[0055] With this configuration, the disease diagnosis result determination device 1 according to the present embodiment can determine the result of the brain examination from the health examination data MC, and further determine the disease diagnosis result from the determined result of the brain examination. Therefore, the determination accuracy of the disease diagnosis result can be improved.

[0056] Here, in the disease diagnosis result determination device 1, the items included in the health examination data MC may be any items of various health examinations such as specific health examinations. Therefore, compared with the case where the determination is made based only on limited items such as age and blood data, the determination accuracy of the disease diagnosis result can be improved. In the disease diagnosis result determination device 1, any type of brain examination may be used for the result of the brain examination determined from the health examination data MC. Therefore, anatomical and objective indices of the brain such as indices related to MRI images can be used, and compared with the case where only indices based on subjective evaluation methods such as scores of questionnaire tests (such as MMSE) are used for the result of the brain examination, the determination accuracy of the disease diagnosis result can be improved. In the disease diagnosis result determination device 1, after determining the results of the brain examination from the health check data MC, the diagnosis result of the disease can be determined from the results of the brain examination so that a doctor can comprehensively diagnose dementia from the results of the brain examination method. Therefore, the determination process by machine learning (deep learning) becomes similar to the diagnosis by a specialist doctor, and the process by which the diagnosis result of the disease is determined by machine learning (deep learning) can be visualized.

[0057] In addition, in the disease diagnosis result determination device 1, since health check data is used, a systemic metabolic disorder that is a risk factor for dementia can be identified for each individual of the second subject P2, so there is an advantage that a customized lifestyle improvement program (such as diet and exercise therapy) can be provided. For this reason, in the disease diagnosis result determination device 1, efficient prevention of dementia is possible, and an improvement in incentives for lifestyle improvement is expected by using the disease diagnosis result determination device 1.

[0058] The disease diagnosis result determination device 1 aims to utilize regularly conducted health checks and determine the onset risk of diseases such as dementia with high accuracy by machine learning such as deep learning. In the conventional method, deep learning was used, with age and blood data used in the input layer and the score of the questionnaire test (MMSE) output from the output layer to predict cognitive impairment. In the conventional method, the items in the input layer and output layer were limited, and since the questionnaire test that outputs a score from the output layer is a subjective examination method, there were problems with the prediction accuracy of cognitive impairment. Furthermore, the prediction process from the input layer to the output layer of deep learning was black-boxed, which was an obstacle when applying deep learning to medical care.

[0059] In addition, in the disease diagnosis result determination device 1 according to the present embodiment, the brain examination results consist of the results of multiple types of brain examinations. The first learning result L1 consists of multiple types of learning results corresponding to each of the multiple types of brain examinations (in the present embodiment, the brain examination learning result LB-1, the brain examination learning result LB-2, and the brain examination learning result LB-3). The brain examination result determination unit 4 determines the results of multiple types of brain examinations for the second subject P2 based on the multiple types of learning results corresponding to each of the multiple types of brain examinations (in the present embodiment, the brain examination learning result LB-1, the brain examination learning result LB-2, and the brain examination learning result LB-3) and the health examination data MC regarding the second subject P2. With this configuration, since the disease diagnosis result determination device 1 according to the present embodiment can use the results of multiple types of brain examinations for determination, it can improve the determination accuracy of the disease diagnosis result compared to the case of using the result of one type of brain examination.

[0060] In addition, in the disease diagnosis result determination device 1 according to the present embodiment, the brain examination result determination unit 4 uses the statistical amounts of the health examination data regarding a plurality of fourth subjects P4 as a part of the health examination data MC regarding the second subject P2, and based on the first learning result L1 and the health examination data MC regarding the second subject P2, determines the brain examination result (in the present embodiment, the brain examination result determination result B) regarding the second subject P2. With this configuration, even when a part of the health examination data MC regarding the second subject P2 cannot be obtained, the disease diagnosis result determination device 1 according to the present embodiment can substitute a part of the health examination data MC with the statistical amounts of the health examination data regarding a plurality of fourth subjects P4, so that the disease diagnosis result can be determined even when a part of the health examination data MC regarding the second subject P2 cannot be obtained.

[0061] Hereinafter, a modification example of the present embodiment will be described. (Modification Example 1) In Modification 1, when the results obtained from some of the multiple types of brain examinations for the second subject P2 can be acquired in advance, regarding the results of such brain examinations, instead of the brain examination result determination result by the first machine learning, the case of determining the disease diagnosis result using the results of the actually performed brain examinations in advance will be described.

[0062] FIG. 4 is a diagram showing an example of the configuration of the disease diagnosis result determination device 1a according to this modification. Comparing the disease diagnosis result determination device 1a (FIG. 4) according to this modification with the disease diagnosis result determination device 1 (FIG. 1) according to the above-described embodiment, the disease diagnosis result determination unit 6a, the brain examination result acquisition unit 11a, and the brain examination result supply unit 12a are different. For the same configurations as those in the above-described embodiment, the same reference numerals are given, and the description of the same configurations and operations is omitted.

[0063] The disease diagnosis result determination device 1a includes a first learning result acquisition unit 2, a health check data acquisition unit 3, a brain examination result determination unit 4, a second learning result acquisition unit 5, a disease diagnosis result determination unit 6a, and a brain examination result acquisition unit 11a. The brain examination result acquisition unit 11a acquires the brain examination result E supplied from the brain examination result supply unit 12a. The brain examination result E is the result obtained from some of the multiple types of brain examinations for the second subject P2 that have been performed in advance.

[0064] The disease diagnosis result determination unit 6a makes a determination using the brain examination result E acquired by the brain examination result acquisition unit 11a instead of the brain examination result determination result B determined by the brain examination result determination unit 4 for some of the multiple types of brain examinations.

[0065] The brain examination result supply unit 12a supplies the brain examination result E to the disease diagnosis result determination device 1a. The brain examination result supply unit 12a may be, for example, a human interface device such as a keyboard, a tablet, or a scanner, or may be an information storage device such as a server.

[0066] FIG. 5 is a diagram showing an example of the determination process by the disease diagnosis result determination device 1a according to the present embodiment. Note that the processes of step S110 to step S140, step S160, and step S180 are the same as the processes of step S10 to step S40, step S50, and step S70 in FIG. 3, and thus the description thereof is omitted.

[0067] Step S150: The brain examination result acquisition unit 11a acquires the brain examination result E supplied from the brain examination result supply unit 12a. The brain examination result acquisition unit 11a supplies the acquired brain examination result E to the disease diagnosis result determination unit 6a. The brain examination result E is, for example, FA-BHQ among a plurality of types of brain examinations.

[0068] Step S170: The disease diagnosis result determination unit 6 determines the disease diagnosis result D based on the second learning result L2 acquired by the second learning result acquisition unit 5, the brain examination result determination result B determined by the brain examination result determination unit 4, and the brain examination result E acquired by the brain examination result acquisition unit 11a. Here, the disease diagnosis result determination unit 6a performs the determination using the brain examination result E acquired by the brain examination result acquisition unit 11a instead of the brain examination result determination result B determined by the brain examination result determination unit 4 for some of the plurality of types of brain examinations.

[0069] The disease diagnosis result determination unit 6 determines the disease diagnosis result D using, for example, the score of MMSE, GM-BHQ, and FA-BHQ acquired by the brain examination result acquisition unit 11a among the brain examination result determination results B determined by the brain examination result determination unit 4. Note that, for the brain examination result corresponding to the brain examination result E acquired by the brain examination result acquisition unit 11a among the brain examination result determination results B, the determination by the brain examination result determination unit 4 may be omitted.

[0070] As described above, the disease diagnosis result determination device 1a according to this modification example includes a brain examination result acquisition unit 11a. The brain examination result acquisition unit 11a acquires a brain examination result E obtained by previously performing some of a plurality of types of brain examinations on the second subject P2. The disease diagnosis result determination unit 6 performs determination using the brain examination result E acquired by the brain examination result acquisition unit 11a instead of the result of the brain examination regarding the second subject P2 (in this embodiment, the brain examination result determination result B) determined by the brain examination result determination unit 4 for some of the plurality of types of brain examinations. With this configuration, when the disease diagnosis result determination device 1a according to this modification example can acquire the brain examination result obtained by previously performing the brain examination, it can use the brain examination result for determination, so that the determination accuracy of the disease diagnosis result can be improved using the actual brain examination result. can be done.

[0071] As described above, in the disease diagnosis result determination device 1a, determination can be made using the health check data MC without actually performing a brain examination such as MRI image imaging. On the other hand, for example, when the brain examination result determination result B in the first machine learning indicates that the result of the brain examination is not good, the corresponding brain examination may be performed only on the subjects with high risk, and the result of the brain examination may be used for determination to improve the determination accuracy.

[0072] (Modification Example 2) There may be a case where it is desired to confirm through what process the disease diagnosis result by the disease diagnosis result determination device is determined. In Modification Example 2, the case where the brain examination result determination result is presented when it is desired to confirm the brain examination result determination result used for determining the disease diagnosis result will be described.

[0073] FIG. 6 is a diagram showing an example of the configuration of the disease diagnosis result determination device 1b according to this modified example. Comparing the disease diagnosis result determination device 1b (FIG. 6) according to this modified example with the disease diagnosis result determination device 1 (FIG. 1) according to the above-described embodiment, the brain examination result output unit 13b and the presentation unit 10b are different. For the same configurations as those in the above-described embodiment, the same reference numerals are given, and the description of the same configurations and operations is omitted.

[0074] The disease diagnosis result determination device 1b includes a first learning result acquisition unit 2, a health check data acquisition unit 3, a brain examination result determination unit 4, a second learning result acquisition unit 5, a disease diagnosis result determination unit 6, and a brain examination result output unit 13b. After the disease diagnosis result determination unit 6 determines the disease diagnosis result D, the brain examination result output unit 13b outputs determination process information R. The determination process information R is a pair of the result of the brain examination regarding the second subject P2 used for determining the disease diagnosis result D and the disease diagnosis result D.

[0075] The presentation unit 10b presents the determination process information R output from the brain examination result output unit 13b by presentation means such as display and printing.

[0076] FIG. 7 is a diagram showing an example of the determination process by the disease diagnosis result determination device 1b according to this embodiment. Since each process from step S210 to step S270 is the same as each process from step S10 to step S70 in FIG. 3, the description thereof is omitted.

[0077] Step S280: The brain examination result output unit 13b outputs the determination process information R to the presentation unit 10b. Here, the brain examination result output unit 13b outputs the determination process information R after the disease diagnosis result determination unit 6 determines the disease diagnosis result D in step 270.

[0078] As described above, the disease diagnosis result determination device 1b according to this modification example includes a brain examination result output unit 13b. After the disease diagnosis result determination unit 6 determines the diagnosis result of the disease (in this embodiment, the disease diagnosis result D), the brain examination result output unit 13b outputs a pair of the result of the brain examination regarding the second subject P2 (in this embodiment, the brain examination result determination result B) used to determine the diagnosis result (in this embodiment, the disease diagnosis result D) and the diagnosis result (in this embodiment, the disease diagnosis result D). With this configuration, in the disease diagnosis result determination device 1b according to this modification example, since the result of the brain examination used to determine the diagnosis result of the disease can be confirmed, the determination process can be confirmed. As described above, conventionally, the prediction process from the input layer to the output layer of deep learning has been black-boxed, which has been an obstacle when applying deep learning to medicine. In the disease diagnosis result determination device 1b, the process in which the output layer is estimated from the input layer of the deep learning method can be visualized.

[0079] In the above-described embodiment, an example in the case where the result of the brain examination consists of the results of a plurality of types of brain examinations has been described, but it is not limited to this. The result of the brain examination may be one type (for example, only the result of MMSE).

[0080] In the above-described embodiment, an example in the case where the first machine learning and the second machine learning are deep learning has been described, but it is not limited to this. As long as it is supervised learning, other machine learning may be used for the first machine learning and the second machine learning. Different types of machine learning may be used for the first machine learning and the second machine learning.

[0081] In the above-described embodiments, an example has been described in which the disease diagnosis result determination device 1 obtains the first learning result L1 and the second learning result L2 from the outside (the first learning result supply unit 7 and the second learning result supply unit 8), but the present invention is not limited to this. The disease diagnosis result determination device 1 may execute either the first machine learning or the second machine learning, or both. The disease diagnosis result determination device 1 may include either or both of the first learning unit and the second learning unit. The first learning unit executes the first machine learning and supplies the first learning result L1 to the first learning result acquisition unit 2. The second learning unit executes the second machine learning and supplies the second learning result L2 to the second learning result acquisition unit 5.

[0082] In the above-described embodiments, an example has been described in which the disease diagnosis result determination devices 1, 1a, and 1b include the brain examination result determination unit 4 and the disease diagnosis result determination unit 6, and the disease diagnosis result determination devices 1, 1a, and 1b execute the first machine learning and the second machine learning, but the present invention is not limited to this. The first machine learning and the second machine learning may be executed by different determination devices, respectively. In that case, for example, the determination device that executes the first machine learning includes the first learning result acquisition unit 2, the health check data acquisition unit 3, and the brain examination result determination unit 4. Also in that case, the determination device that executes the second machine learning includes the second learning result acquisition unit 5 and the disease diagnosis result determination unit 6, and obtains the brain examination result determination result from the determination device that executes the first machine learning to determine the disease diagnosis result.

[0083] Here, FIG. 8 is a diagram showing an example of a next-generation preventive medical model according to the present embodiment. In the conventional medical model, a person visited a medical institution and received treatment (insurance medical treatment) after getting sick. On the other hand, by using the disease diagnosis result determination devices 1, 1a, and 1b according to the above-described embodiments, since the disease diagnosis result can be determined based on the health check data that the subject receives every year in a healthy state, the subject can improve his / her lifestyle with reference to the disease diagnosis result in a healthy state. As a result, diseases can be prevented, the healthy life span can be extended, and medical costs can also be reduced.

[0084] FIG. 9 is a diagram showing an example of a health future prediction chart according to the present embodiment. The health future prediction chart is a plot of the disease risk based on the disease diagnosis results by the disease diagnosis result determination devices 1, 1a, and 1b according to the above-described embodiment against the time axis. The time indicated by this time axis is the time at the point when the health examination data was acquired. In the disease diagnosis result determination devices 1, 1a, and 1b according to the above-described embodiment, by changing the types of disease diagnosis results corresponding to the output layer of the neural network used for deep learning, various diseases can be predicted from the health examination data. Further, by continuously acquiring the health examination data over a long period of time and training the acquired health examination data, it becomes possible to predict the disease risks of various diseases. Since the disease risk changes every year due to lifestyle interventions, the visualization of the lifestyle intervention effect is possible with the health future prediction chart, and the incentive for the intervention can be increased.

[0085] (Example 1) Next, the embodiment of the above-described modification 2 will be described. In this modification, a feedforward type deep learning neural network as shown in FIG. 2 was used. In this modification, blood test data and age were used as the health examination data MC input to the input layer. In this modification, GM-BHQ by MRI was used as the brain examination result determination result B output from the output layer of the first neural network. Learning was performed by the first neural network using 1799 cases of brain MRI data. The prediction accuracy was verified by cross-validation.

[0086] The above-described brain MRI data includes a set of age, blood test data, and GM-BHQ. The brain MRI data is data showing the examination results by brain MRI for 1799 brain MRI patients with an average age of 62.0 ± 13.1 years. GM-BHQ, which is the BHQ of the cerebral gray matter (GM), was used as an index of brain atrophy indicating the degree of brain atrophy. GM-BHQ was measured based on the imaging results by a 3T-MR device.

[0087] The blood test data used as the health check data MC consists of the following 26 items. That is, the blood test data consists of 26 items: total protein (TP), albumin, A / G ratio, total bilirubin, GOT, GPT, γ-GTP, ALP, WBC, RBC, Hb, Ht, PLT, HCV, total cholesterol, triglyceride, HDL cholesterol, LDL cholesterol, calculated LDL cholesterol value, blood glucose, HbA1c, fibrinogen, urea nitrogen (BUN), creatinine (Cr), uric acid, and amylase.

[0088] Figure 10 is a diagram showing an example of the correlation between the measured value (correct label) and the predicted value of GM-BHQ according to this embodiment. Note that the measured value (correct label) and the predicted value of GM-BHQ correspond to the determination process information R output by the brain examination result output unit 13b provided in the disease diagnosis result determination device 1b (Fig. 6) described in Modification 2.

[0089] Figure 10(A) shows the correlation when age is included in the health check data MC input to the input layer. The measured value (correct label) of GM-BHQ is the value of GM-BHQ included in the above-described 1799 cases of brain MRI data. The predicted value of GM-BHQ is the value of GM-BHQ output by inputting the health check data MC to the first neural network trained using the brain MRI data.

[0090] The correlation coefficient for the correlation shown in Figure 10(A) is 0.70, and the p-value is less than 0.001, indicating a significant positive correlation between the measured value and the predicted value. For comparison, Figure 10(B) shows the correlation when age is not included in the health check data MC. The correlation coefficient for the correlation shown in Figure 10(B) is 0.58, and the p-value is less than 0.001, indicating a significant positive correlation between the measured value and the predicted value even when age is not included in the health check data MC.

[0091] In addition, the importance of the 26 items included in the blood test data used for the health examination data MC was calculated with respect to the prediction results. The importance is shown for the top 10 items in FIG. 11. The item of age had the highest importance with respect to the prediction results, followed by the item of blood urea nitrogen.

[0092] The above results indicate that by using deep learning (deep neural network using a feedforward network), not only cognitive function changes but also structural changes in the brain (brain atrophy) can be predicted from general blood test data, showing that the two-stage deep learning by the disease diagnosis result determination devices 1, 1a, and 1b in the embodiments can be applied to the diagnosis of dementia. Also, the above results suggest that the anatomical structure of the brain is closely related to systemic metabolic disorders. Predicting the structural changes and cognitive function of the brain based on general blood test data by two-stage deep learning may be applicable as a screening test for dementia.

[0093] Note that, a part of the disease diagnosis result determination devices 1, 1a, and 1b in the above-described embodiments, for example, the brain examination result determination unit 4, the disease diagnosis result determination unit 6, the brain examination result acquisition unit 11a, and the brain examination result output unit 13b may be realized by a computer. In that case, a program for realizing this control function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be read into a computer system and executed to be realized. Here, the "computer system" refers to a computer system built in the disease diagnosis result determination devices 1, 1a, and 1b, including hardware such as an OS and peripheral devices. Further, the "computer-readable recording medium" refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, etc., and a storage device such as a hard disk built in a computer system. Furthermore, the "computer-readable recording medium" also includes something that holds a program dynamically for a short time, like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line, and something that holds a program for a certain time, like a volatile memory inside a computer system that becomes a server or a client in that case. Also, the above program may be for realizing a part of the aforementioned functions, and may also be realized in combination with a program already recorded in a computer system for realizing the aforementioned functions. Also, a part or all of the disease diagnosis result determination devices 1, 1a, and 1b in the above-described embodiments may be realized as an integrated circuit such as an LSI (Large Scale Integration). Each functional block of the disease diagnosis result determination devices 1, 1a, and 1b may be made into a processor individually, or a part or all of them may be integrated and made into a processor. Also, the method of integrating into an integrated circuit is not limited to LSI, and may be realized by a dedicated circuit or a general-purpose processor. Also, when a technology for integrating into an integrated circuit that replaces LSI appears due to the progress of semiconductor technology, an integrated circuit using that technology may be used.

[0094] The above has described in detail an embodiment of the present invention with reference to the drawings. However, the specific configuration is not limited to the above, and various design changes and the like can be made without departing from the gist of the present invention.

Industrial Applicability

[0095] In the present invention, the first machine learning and the second machine learning are used. Since the algorithms of each machine learning (deep learning) use existing algorithms (such as feedforward neural networks), there is no need to develop the algorithm itself. In addition, by collaborating with the Japanese Brain Dock Society, it is not difficult to collect a large number of cases that combine health examination data, brain examination results such as MRI, and diagnoses by specialists. By using the present invention, an inexpensive large-scale screening test for diseases such as dementia using health examination data becomes possible. Since only health examination data is used in the present invention, there is no need for blood sampling, and there is an advantage that the risk of diseases such as dementia can be determined in daily life, such as using it in non-medical facilities such as sports gyms or using a smartphone. Taking advantage of such advantages and combining the present invention with recently released dementia insurance enables early detection of dementia. Furthermore, by combining it with specific health examinations and specific health guidance (for insurance subscribers and dependents aged 40 to less than 75 years old), it becomes possible to suppress or delay the onset of dementia at the municipal to national level, and it is considered to bring innovation to the countermeasures against dementia, which has become a social problem.

Explanation of Reference Numerals

[0096] 1, 1a, 1b... Disease diagnosis result determination device, 4... Brain examination result determination unit, 6... Disease diagnosis result determination unit, L1... First learning result, L2... Second learning result, MC... Health examination data, B... Brain examination result determination result, D... Disease diagnosis result

Claims

1. Based on a first learning result in which the relationship between the medical examination data regarding the first subject and the result of the brain examination regarding the first subject has been learned in advance, and the medical examination data regarding the second subject, a brain examination result determination unit that determines the result of the brain examination regarding the second subject; Based on a second learning result in which the relationship between the result of the brain examination regarding the third subject and the diagnosis result of the disease regarding the third subject has been learned in advance, and the result of the brain examination regarding the second subject determined by the brain examination result determination unit, a disease diagnosis result determination unit that determines the diagnosis result of the disease regarding the second subject; A disease diagnosis result determination device comprising the above.

2. The result of the brain examination consists of the results of multiple types of brain examinations, The first learning result consists of multiple types of learning results corresponding to each of the multiple types of brain examinations, The brain examination result determination unit determines the results of the multiple types of brain examinations regarding the second subject based on the multiple types of learning results corresponding to each of the multiple types of brain examinations and the medical examination data regarding the second subject. The disease diagnosis result determination device according to Claim 1.

3. Further comprising a brain examination result acquisition unit that acquires the brain examination results obtained by previously performing some of the multiple types of brain examinations regarding the second subject, For some of the multiple types of brain examinations, the disease diagnosis result determination unit performs the determination using the brain examination results acquired by the brain examination result acquisition unit instead of the brain examination results regarding the second subject determined by the brain examination result determination unit. The disease diagnosis result determination device according to Claim 2.

4. The brain examination result determination unit uses the statistical quantity of the medical examination data regarding a plurality of fourth subjects as a part of the medical examination data regarding the second subject, and based on the first learning result and the medical examination data regarding the second subject, determines the result of the brain examination regarding the second subject. The disease diagnosis result determination device according to any one of Claims 1 to 3.

5. After the disease diagnosis result determination unit determines the diagnosis result, a brain examination result output unit that outputs a pair of the result of the brain examination regarding the second subject used to determine the diagnosis result and the diagnosis result. The disease diagnosis result determination device according to any one of Claims 1 to 4, further comprising the above.

6. A disease diagnosis result determination method executed by a computer, a brain examination result determination step in which the computer determines a brain examination result for the second subject based on a first learning result in which a relationship between a medical examination data for a first subject and a result of a brain examination for the first subject has been learned in advance, and the medical examination data for the second subject; a disease diagnosis result determination step in which the computer determines a disease diagnosis result for the second subject based on a second learning result in which a relationship between a result of a brain examination for a third subject and a disease diagnosis result for the third subject has been learned in advance, and the brain examination result for the second subject determined in the brain examination result determination step; A disease diagnosis result determination method comprising: **Claim 7** A program for causing a computer to execute a brain examination result determination step in which the computer determines a brain examination result for the second subject based on a first learning result in which a relationship between a medical examination data for a first subject and a result of a brain examination for the first subject has been learned in advance, and the medical examination data for the second subject; execute a disease diagnosis result determination step in which the computer determines a disease diagnosis result for the second subject based on a second learning result in which a relationship between a result of a brain examination for a third subject and a disease diagnosis result for the third subject has been learned in advance, and the brain examination result for the second subject determined in the brain examination result determination step; ​

Images