Disease prediction model

Abstract

To predict a future onset of disease.SOLUTION: A disease prediction system comprises a prediction unit that inputs input information about a subject into a neural network to predict the future onset of disease in the subject. The input information includes first information having at least a plain X-ray image of the subject and second information about hormone-like substances in the subject. The neural network is generated through machine learning by using first learning information including the plain X-ray image and second learning information regarding hormone-like substances acquired at the time of acquisition of the first learning information or during a prescribed period before and after the acquisition as learning data, and using diagnostic results of the disease corresponding to the learning data as teacher data.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a disease prediction system. [Background technology]

[0002] Patent Document 1 describes a diagnostic support device for osteoporosis, which is one of the diseases. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2008-36068 Summary of the Invention [Problem to be solved by the invention]

[0004] In order to understand future disease risks, it is necessary to predict the future onset of these diseases. [Means for solving the problem]

[0005] A disease prediction system according to one embodiment of the present disclosure includes a prediction unit that inputs input information about a subject into a neural network to predict the subject's future onset of a disease, the input information including at least first information including a simple X-ray image of the subject and second information about hormone-like substances in the subject, and the neural network uses the first training information including the simple X-ray image and the second training information about hormone-like substances acquired at the time of acquisition of the first training information or within a predetermined period before and after acquisition as training data, and generates a diagnosis result of the disease corresponding to the training data by machine learning using the training data as teacher data. [Brief explanation of the drawings]

[0006] [Figure 1] 1 is a conceptual diagram schematically illustrating the configuration of a disease prediction system 1 according to the present disclosure. [Figure 2] 1 is a conceptual diagram schematically illustrating a configuration of a part of a disease prediction system 1 according to the present disclosure. [Figure 3] 1 is a conceptual diagram schematically illustrating a configuration of a part of a disease prediction system 1 according to the present disclosure. [Figure 4] 1 is a conceptual diagram schematically illustrating a configuration of a part of a disease prediction system 1 according to the present disclosure. [Figure 5] 1 is a conceptual diagram schematically illustrating a configuration of a part of a disease prediction system 1 according to the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0007] The disease prediction system 1 of the present disclosure can predict the onset of diseases in, for example, middle-aged and elderly people. Diseases in middle-aged and elderly people include, for example, osteoporosis, osteoarthritis, spondylosis, fractures, sarcopenia, frailty, menopausal disorders, erectile dysfunction (ED), and periodontal disease.

[0008] In the following, unless otherwise specified, osteoporosis will be used as an example of a disease.

[0009] FIG. 1 shows a conceptual configuration of a disease prediction system 1 of the present disclosure.

[0010] The disease prediction system 1 of the present disclosure includes a terminal device 2 and a prediction device 3. The terminal device 2 acquires subject data to be used for diagnosing a disease. The prediction device 3 can predict the future onset of a disease based on the data acquired by the terminal device 2.

[0011] As described above, the terminal device 2 can acquire data. The data acquired by the terminal device 2 is input to the prediction device 3 as input information I. The input information I includes data on the subject used for diagnosing a disease (first information I1) as well as information on the subject's hormone-like substance (second information I2). The terminal device 2 includes a first terminal device 21 that acquires the first information I1 and a second terminal device 22 that acquires the second information.

[0012] For example, when the disease prediction system 1 is used to diagnose osteoporosis, the first terminal device 21 may be, for example, a plain X-ray imaging device or a bone mass measuring device. In this case, the first information I1 may include a medical image such as a plain X-ray image.

[0013] The second terminal device 22 can acquire second information I2 that affects a disease related to the first information I1 acquired by the first terminal device 21. The second information I2 may be, for example, information on the presence or absence or concentration of a hormone-like substance. The second information I2 may be, for example, a measurement value of a hormone-like substance in the blood or urine of a subject. Specifically, when the first information I1 is related to osteoporosis, the second information I2 may be data indicating the presence or absence of non-steroidal estrogen. The second terminal device 22 may acquire, for example, a surface acoustic wave (SAW) Any inspection device using a Wave sensor may be used.

[0014] After acquiring the input information I, the terminal device 2 may transfer the input information I to the prediction device 3. For example, when acquiring a plain X-ray image of a subject, the terminal device 2 (first terminal device 21) is installed, for example, in an X-ray room and takes an X-ray photograph of the subject. Then, the image data is transferred from the terminal device 2 to the prediction device 3, and the future onset of the disease can be predicted via the prediction device 3.

[0015] It should be noted that the terminal device 2 does not have to directly transfer the input information I to the prediction device 3. In this case, for example, the input information I acquired by the terminal device 2 may be stored in a storage medium, and the input information I may be input to the prediction device 3 via the storage medium.

[0016] After acquiring the second input information I2, the second terminal device 22 may transfer the second input information I2 to the prediction device 3. For example, when a testing device using a SAW sensor transfers information on the concentration of a hormone-like substance in the blood or urine of a subject, the testing device may transfer the information as a numerical value of the concentration, or may transfer primary data information before conversion into concentration. The primary data may be, for example, information on phase changes, amplitude changes, frequency changes, etc. of the signal detected by the sensor.

[0017] FIG. 2 shows a conceptual configuration of the prediction device 3 according to this embodiment.

[0018] The prediction device 3 can predict the future onset of a disease in a subject from input information I input to the prediction device 3. For example, when the disease prediction system 1 predicts osteoporosis, the prediction device 3 can predict the future onset of osteoporosis in a subject from input information I including medical images acquired by the terminal device 2, and output the predicted prediction result O.

[0019] The prediction device 3 has an input unit 31, a control unit 34, an output unit 33, and a storage unit 35. The input unit 31, the output unit 33, the control unit 34, and the storage unit 35 are electrically connected to one another, for example, by a bus 60. The input unit 31 receives input information I from the terminal device 2. The control unit 34 executes a control program to predict the onset of a disease based on the input information I and on the prediction unit 32, which will be described later. The output unit 33 can output a prediction result O predicted by the prediction unit 32. The storage unit 35 stores the control program and various data, parameters, etc. required for control.

[0020] The prediction device 3 of the present disclosure has a plurality of electronic components and circuits. In other words, the prediction device 3 can have each component that constitutes the prediction device 3 formed by a plurality of electronic components and circuits. The plurality of electronic components may be, for example, active elements such as transistors or diodes, or passive elements such as capacitors, and may be formed by a conventionally known method.

[0021] As described above, the input unit 31 receives input information I. The input unit 31 may include a communication unit so that the input information I acquired by the terminal device 2 is directly input from the terminal device 2. The input unit 31 may also include an input device capable of inputting the input information I or other information. The input device may be, for example, a keyboard, a touch panel, or a mouse.

[0022] The control unit 34 can generally manage the operation of the prediction device 3 by controlling the other components of the prediction device 3. The control unit 34 can also be referred to as a control circuit. The control unit 34 includes at least one processor to provide control and processing power for performing various functions, as described in more detail below.

[0023] In one embodiment, a processor includes one or more circuits or units configured to perform one or more data computational procedures or processes, for example, by executing instructions stored in associated memory. In other embodiments, a processor may be firmware (e.g., discrete logic components) configured to perform one or more data computational procedures or processes.

[0024] According to various embodiments, the processor may include one or more processors, controllers, microprocessors, microcontrollers, application specific integrated circuits (ASICs), digital signal processors, programmable logic devices, field programmable gate arrays, or any combination of these devices or configurations, or other known combinations of devices and configurations, to perform the functions described below. In this example, the control unit 34 includes, for example, a CPU (Central Processing Unit).

[0025] The storage unit 35 includes a non-transitory recording medium such as a ROM (Read Only Memory) and a RAM (Random Access Memory) that can be read by the CPU of the control unit 34. A control program for controlling the prediction device 3 is stored in the storage unit 35. Various functions of the control unit 34 are realized by the CPU of the control unit 34 executing the control program in the storage unit 35. It can also be said that the control program is a prediction program for causing the computer device 1 to function as the prediction device 3.

[0026] In this example, the control unit 34 executes the control program in the storage unit 35, whereby the control unit 34 forms a prediction unit 32 capable of estimating the prediction result O. The prediction unit 32 includes, for example, a neural network. The control program can also be said to be a program for causing the computer device 1 to function as a neural network (prediction unit 32). An example of the configuration of the neural network will be described in detail later.

[0027] In addition to the control program, the memory unit 35 also stores trained parameters, estimation data (hereinafter also referred to as "input information"), training data, and teacher data related to the neural network. The training data and teacher data are data used when training the neural network. The trained parameters and estimation data are data used when the trained neural network estimates the onset of a disease.

[0028] The prediction unit 32 can predict the future onset of a disease in a subject from the input information I input to the input unit 31. The prediction unit 32 has AI (Artificial Intelligence). The AI ​​of the present disclosure may be, for example, a neural network.

[0029] Furthermore, the prediction unit 32 has undergone a learning process in advance. That is, by applying machine learning to the prediction unit 32 using learning data and teacher data, the prediction unit 32 can calculate a prediction result O from input information I. Note that the learning data or teacher data may be any data that corresponds to the input information I input to the prediction device 3 and the prediction result O output from the prediction device 3.

[0030] 3 and 4 show the concept of the configuration of the prediction unit 32 of the present disclosure.

[0031] The prediction unit 32 has a first neural network 321 and a second neural network 322. The first neural network 321 may be any neural network suitable for handling time-series information. For example, the first neural network 321 may be a ConvLSTM network that combines a CNN (Convolutional Neural Network) and an LSTM (Long Short-Term Memory). The second neural network 322 may be, for example, a convolutional network configured with a CNN.

[0032] 3 shows a conceptual diagram of the configuration of the first neural network 321 of the present disclosure. FIG. 4 shows a conceptual diagram of the configuration of the second neural network 322.

[0033] The first neural network 321 has an encoding unit E and a decoding unit D. The encoding unit E can extract features of time change and position information of the input information I. The decoding unit D can calculate new features based on the features extracted by the encoding unit E, the time change and initial value of the input information I.

[0034] The encoding unit E has multiple Convolutional Long Short-Term Memory (ConvLSTM) layers E1. The decoding unit D has multiple Convolutional Long Short-Term Memory (ConvLSTM) layers D1. Note that the multiple ConvLSTM layers E1 may each learn different content. The multiple ConvLSTM layers D1 may each learn different content. For example, one ConvLSTM layer may learn fine details such as pixel-by-pixel changes, while another ConvLSTM layer may learn rough details such as changes in the overall picture.

[0035] The second neural network 322 has a conversion unit C. The conversion unit C can convert the feature amount calculated by the decoding unit D into bone mass. The conversion unit C has multiple convolution layers C1, multiple pooling layers C2, and a fully connected layer C3. The fully connected layer C3 is located before the output unit 33, and multiple convolution layers C1 and multiple pooling layers C2 are arranged alternately.

[0036] During learning by the prediction unit 32, the learning data is input to the encoding unit E of the prediction unit 32, and the teacher data is compared with the output data output from the conversion unit C of the prediction unit 32.

[0037] The output unit 33 can display the prediction result O. The output unit 33 is, for example, a liquid crystal display or an organic EL display. The output unit 33 can display various information such as characters, symbols, and figures. The output unit 33 can display, for example, numbers or images.

[0038] <Examples of input information, learning data, and teacher data> The input information I includes first information I1 about the subject to be used for diagnosing the disease. The first information I1 may be, for example, test results or medical images of the subject. Specifically, for example, when the disease prediction system 1 predicts osteoporosis, the first information I1 may be the subject's bone mass, bone metabolic markers, X-ray images of the chest, lumbar region, or proximal femur, CT images, etc.

[0039] In the case of osteoporosis, the first information I1 specifically includes image data of a plain X-ray image of the subject's bones. The bones to be photographed are primarily cortical and cancellous bones derived from living organisms, but the bones to be photographed may also include artificial bones primarily composed of calcium phosphate or regenerated bones artificially produced by regenerative medicine or the like.

[0040] The input information I may also include second information on hormone-like substances that affect the disease associated with the first information I1. The second information I2 may be, for example, information on the presence or absence or concentration of hormone-like substances. For example, if the first information I1 is data related to osteoporosis, the second information I2 may be information on hormone-like substances that affect osteoporosis. Specifically, the second information I2 may be concentration information on at least one type of hormone, such as fibroblast growth factor 23 (FGF23), leptin, insulin, or sclerostin. Furthermore, if the subject is female, the second information I2 may be concentration information on at least one type of hormone, such as steroidal estrogens (e.g., estrone, estradiol, or estriol), non-steroidal estrogens, progesterone, or estrogen receptors. If the subject is male, the second information I2 may be concentration information on at least one type of hormone, such as testosterone or dihydrotestosterone, or androgen receptors.

[0041] For example, if the first information I1 is data related to sarcopenia, the second information I2 may be information about hormone-like substances that affect sarcopenia. Specifically, the second information I2 may be concentration information on at least one type of hormone, such as ghrelin, leptin, or adiponectin. In addition, if the subject is female, the second information I2 may be concentration information on at least one type of hormone, such as steroidal estrogens (e.g., estrone, estradiol, or estriol), nonsteroidal estrogens, progesterone, or estrogen receptors. If the subject is male, the second information I2 may be concentration information on at least one type of hormone, such as testosterone or dihydrotestosterone, or androgen receptors.

[0042] When the disease prediction system 1 predicts osteoarthritis or spondylosis, the first information I1 may be an X-ray image of the affected area. When predicting a fracture, the first information I1 may be bone mass, bone metabolic markers, a fracture history, or an X-ray image. When predicting sarcopenia, the first information I1 may be an X-ray image of the lower leg, muscle mass, grip strength, or walking speed. When predicting frailty, the first information I1 may be information related to grip strength, walking speed, activity level, oral function, fatigue, and sociability. When predicting menopausal symptoms, the first information I1 may be information related to facial flushing, sweating tendency, coldness of the face or limbs, shortness of breath, palpitations, difficulty falling asleep, depth of sleep, irritability, depression, headache, dizziness, nausea, fatigue, stiff shoulders, lower back pain, or pain in the limbs. When predicting erectile dysfunction, the first information I1 may be information related to the success or satisfaction of sexual intercourse. Furthermore, when predicting periodontal disease, the first information I1 may be information such as an X-ray image, a CT image, periodontal pocket depth, attachment level, and oral hygiene status.

[0043] The training data includes first training information of the same type as the first input information I1. For example, if the first input information I1 is a plain X-ray image, the training data may also include a plain X-ray image. Furthermore, if the first input information I1 is bone mass, the training data may also include bone mass.

[0044] The learning data takes into consideration changes over time. That is, the learning data may be a series of data obtained by examining the same person on different time axes. The learning data may also be a group of data from other people. When the learning data is image data, the image data may be a series of data obtained by photographing the same person or the same body part on different time axes. As a result, the prediction unit 32 that has undergone the learning process can predict the future onset of a disease.

[0045] The training data also includes second training information of the same type as the second input information I2. That is, the training data is information about the subject's hormone-like substances. The second training information may also be information about the time of acquisition of the first training information or a predetermined period before and after the acquisition. For example, if the subject is a woman, the second input information I2 may be concentration information about at least one type of substance, such as steroidal estrogens (e.g., estrone, estradiol, estriol), non-steroidal estrogens, progesterone, or estrogen receptors.

[0046] The learning data may be acquired using the terminal device 2, similar to the input information I. The first learning information may be acquired using the first terminal device 21, such as a simple X-ray imaging device. The second learning information may be acquired using the second terminal device 22, such as a SAW (Surface Acoustic Wave) sensor. For example, the learning data may be the concentration of non-steroidal estrogen, and may be acquired using the second terminal device 22, as described above.

[0047] The training data includes a diagnosis of a disease corresponding to the training data. For example, when predicting osteoporosis, the training data is an actual measurement of bone mass or a diagnosis of osteoporosis corresponding to each of the multiple training image data. The actual measurement of bone mass or the diagnosis of osteoporosis may be evaluated at approximately the same time as the training image data was captured.

[0048] In the case of bone mass, for example, the training data may be measured by, for example, a dual-energy X-ray absorptiometry (DEXA) method or an ultrasound method.

[0049] <Neural network learning example> The prediction unit 32 is optimized by machine learning using learning data and teacher data so that it can calculate a prediction result O from input information I. That is, the control unit 34 calculates the prediction result O from the input information I based on an approximation formula (prediction unit 32) optimized by machine learning, and machine learning is performed by adjusting parameters within the prediction unit 32 so that the difference between the pseudo-prediction result calculated from the learning data input to the input unit 31 and output from the output unit 33 and the teacher data is reduced. As a result, the prediction unit 32 can perform calculations on the input information I based on the learned parameters and output the prediction result O.

[0050] The parameter adjustment method may be, for example, backpropagation. The parameters include, for example, parameters used in the encoding unit E, decoding unit D, and transform unit C. Specifically, the parameters include weighting coefficients used in the ConvLSTM layers of the encoding unit E and decoding unit D, and the convolutional layer and fully connected layer of the transform unit C.

[0051] As described above, the disease prediction system 1 can predict the onset of a disease by outputting a prediction result O from input information I.

[0052] For example, a conventional osteoporosis diagnostic support device is disclosed that determines osteoporosis using features related to cortical bone. However, the conventional osteoporosis diagnostic support device diagnoses osteoporosis at present, but does not indicate the possibility of future onset of osteoporosis.

[0053] In contrast, the disease prediction system 1 according to the present invention can predict the future onset of a disease in the subject from the input information I. Therefore, it is possible to predict how the disease will progress after data acquisition from the input information I at the time of data acquisition. The disease prediction system 1 according to the present invention also has information (second information) related to hormone-like substances as input information I. For example, since the way the disease progresses changes depending on the presence or absence of hormone-like substances, the accuracy of future predictions can be further improved by inputting the second information.

[0054] The prediction result O of the disease prediction system 1 may be a prediction for a future date that is later than the acquisition date of the input information I. For example, the disease prediction system 1 may make a prediction for a period from 3 months to 50 years after the acquisition date of the input information I, and more preferably, from 6 months to 10 years after the acquisition date of the input information I.

[0055] The prediction result O may be output as a future numerical value. For example, when predicting bone mass, the prediction result O may be a numerical value expressed by at least one of the Young Adult Mean (YAM), T-score, and Z-score.

[0056] When predicting osteoarthritis or spondylosis, the prediction result O may be a KL (Kellgren-Lawrence) grade classification or the like. When predicting fracture, sarcopenia, frailty, menopausal disorder, erectile dysfunction, or periodontal disease, the prediction result O may be the presence or absence of onset. The presence or absence of onset may be determined based on a predetermined threshold value output from the prediction result.

[0057] The prediction result O may output the probability of future onset of a disease. The onset probability may be, for example, the probability of onset on a specific date. In this case, it may be output as "the probability of the disease in one year is 10%." It may also be output as the date of onset. In this case, it may be output as, for example, "there is a possibility that you will develop osteoporosis in seven years."

[0058] The prediction result O may be output as a secular change. For example, the transition of the prediction result O over time, such as the probability of onset one year later, the probability of onset five years later, the probability of onset ten years later, etc., may be output.

[0059] The disease prediction system 1 may output a prediction result O that predicts the onset of multiple diseases from one type of first information I1. In this case, for example, by inputting the concentration of nonsteroidal estrogen in urine as the first information I1, it is possible to simultaneously predict the onset of osteoporosis and menopausal disorders.

[0060] Furthermore, the disease prediction system 1 may output a prediction result O that predicts the onset of multiple diseases from one type of first information I1 and one type of second information I2. In this case, for example, by inputting an X-ray image as one type of first information I1 and a concentration of nonsteroidal estrogen in urine as one type of second information I2, it is possible to simultaneously predict the onset of multiple diseases, such as osteoporosis and osteoarthritis.

[0061] The input information I may include third information I3 having individual data of the subject. The third information I3 may be information relating to the subject's health condition. The individual data may include, for example, one or more types of information such as age information, sex information, height information, weight information, systolic blood pressure, total cholesterol, triglycerides, bad cholesterol (LDL-C, triglycerides), good cholesterol (HDL-cholesterol), insulin resistance index (HOMA-R index), blood glucose level, whether or not the subject has undergone menopause, and sperm count.

[0062] In this case, third learning information of the same type as the third information I3 may be learned as learning data. The third learning information may be information at the time of acquisition of the first learning information or information for a predetermined period before and after the acquisition.

[0063] The third information I3 may also include lifestyle information, which may include one or more types of information related to eating habits, drinking habits, smoking habits, or exercise habits such as walking speed or number of steps.

[0064] The input information I may include fourth information I4 regarding an intervention for the subject. The fourth information I4 may be information regarding planned (future) changes to the subject's lifestyle, etc. The fourth information I4 may include, for example, one or more types of information regarding changes to diet, lifestyle, weight, physical therapy, drug therapy, or planned supplements.

[0065] In this case, fourth learning information of the same type as the fourth information I4 may be learned as learning data. The fourth learning information may be information at the time of acquisition of the first learning information or information for a predetermined period before and after the acquisition.

[0066] Furthermore, when inputting information about supplements to be taken as the fourth information I4, the information about supplements may include one or more of information about taking calcium, vitamin D, vitamin K, branched-chain amino acids, flavonoids, or probiotics. The probiotic may be, for example, Bacillus subtilis C-3102 strain.

[0067] In the disease prediction system 1, the prediction result O may output a first result based on input information I selected from input information I excluding the fourth information I4, and a second result based on input information I including the fourth information I4 and at least the first input information I1.

[0068] In the disease prediction system 1, not only the prediction result O, which is a future prediction, but also the current diagnosis result may be output, so that changes in the disease over time can be compared.

[0069] The input information I may include fifth input information I5 containing information on bone metabolism markers of the subject. The bone metabolism information may be, for example, bone resorption capacity or bone formation capacity. These can be measured using at least one of bone resorption markers: type I collagen cross-linked N-telopeptide (NTX), type I collagen cross-linked C-telopeptide (CTX), tartrate-resistant acid phosphatase (TRACP-5b), and deoxypyridinoline (DPD); bone formation markers: bone alkaline phosphatase (BAP) and type I collagen cross-linked N-propeptide (P1NP); and bone-related matrix marker: undercarboxylated osteocalcin (ucOC). The bone resorption marker may be measured using serum or urine as a sample.

[0070] In this case, fifth training information of the same type as the fifth information I5 may be trained as training data. The fifth training information may be information at the time of acquisition of the first training information or information for a predetermined period before and after the acquisition.

[0071] The disease prediction system 1 may include a third terminal device that acquires fifth input information I5. Alternatively, the second terminal device 22 may be used to simultaneously measure the fifth input information I5 along with the second input information I2. For example, the bone metabolism marker DPD and the hormone-like substance non-steroidal estrogen concentrations may be simultaneously measured from a single urine sample from the subject. The second terminal device 22 may simultaneously measure two or more of the bone metabolism markers included in the first input information I1 and the hormone-like substance included in the second input information I2. Specifically, the bone metabolism markers DPD and CTX and the hormone-like substance steroidal estrogen and non-steroidal estrogen concentrations may be simultaneously measured from a single urine sample from the subject.

[0072] <Other embodiments> FIG. 5 shows a conceptual configuration of the prediction unit 32a using the linear regression model of the present disclosure.

[0073] In the above embodiment, an example in which a neural network is used as the prediction unit 32 has been described. However, the prediction unit 32 may be a linear regression model in which the input information is an explanatory variable, the prediction parameters are coefficients of the explanatory variables, and the prediction result is a target variable. The prediction parameters may be optimized by the least squares method using training data and training data. In other words, the training data may be used as an explanatory variable vector and the training data as a target variable vector, and an explanatory variable coefficient vector that minimizes the sum of squares of the error may be determined.

[0074] A prediction unit 32a using a linear regression model inputs input information into explanatory variables X1, X2,..., Xk to calculate a prediction result Y. The coefficients θ1, θ2..., θk of the explanatory variables, which are prediction parameters, may be optimized in advance by the least squares method using, for example, learning data and teacher data.

[0075] The prediction unit 32a using a linear regression model can easily predict diseases such as osteoporosis, fractures, sarcopenia, frailty, menopausal disorders, ED (Erectile Dysfunction), and periodontal disease.

[0076] The present invention is not limited to the above-described embodiment, but includes various modifications as long as they are consistent with the present invention. In addition, the embodiments of the present invention can be combined as appropriate.

[0077] For example, in the above example, when a neural network is used as the prediction unit 32, an example in which a ConvLSTM network is applied has been described, but the present invention is not limited to this. For example, the prediction unit 32 may use a recurrent neural network (RNN) or a generative adversarial network (GAN). The prediction unit 32 may also be a combination of multiple neural networks. Specifically, the prediction unit 32 may be a composite neural network that combines a ConvLSTM network and a convolutional neural network.

[0078] In the above examples, the prediction unit 32 includes a neural network and a linear regression model, but in the present invention, the prediction device 3 of the disease prediction system 1 may include multiple different prediction units. For example, the prediction device 3 may include both a prediction unit 32a using a linear regression model and a prediction unit 32b using a ConvLSTM neural network.

[0079] In this case, in the disease prediction system 1, the prediction unit 32a outputs a first prediction result O1. Furthermore, the prediction unit 32b outputs a second prediction result as a second prediction result O2. As a result, the first prediction result and the second prediction result can be compared as the prediction result O.

[0080] The disease prediction system 1 may output a third prediction result based on the first prediction result and the second prediction result as the prediction result O. As a result, for example, the result of correcting the first prediction result based on the second prediction result (third prediction result) can be set as the prediction result O.

[0081] [Other aspects] A disease prediction system according to one embodiment of the present disclosure includes an input unit to which input information having first information about a subject to be used in diagnosing a disease and second information about a hormone-like substance in the subject is input, and a control unit that predicts the future onset of the disease in the subject from the input information input to the input unit. [Explanation of symbols]

[0082] 1. Disease prediction system 2. Terminal Device 21 First terminal device 22 Second terminal device 3 Prediction Device 31 Input section 32 Prediction Department 32a Prediction Section 32b Prediction section 33 Output section 34 Control Unit 35 Storage section O Prediction results I Input information

Claims

1. a prediction unit that inputs input information about a subject into a neural network to predict future onset of a disease in the subject; the input information includes first information including at least a plain X-ray image of the subject and second information related to a hormone-like substance of the subject; A disease prediction system in which the neural network uses first learning information including a simple X-ray image and second learning information regarding hormone-like substances acquired at the time of acquisition of the first learning information or during a predetermined period before and after the acquisition as learning data, and generates diagnostic results of the disease corresponding to the learning data by machine learning as teacher data.

2. The disease prediction system of claim 1 , wherein the second information includes a measurement value of a hormone-like substance in the subject's blood or urine.

3. The disease prediction system according to claim 1 , wherein the prediction unit predicts the onset of two or more types of diseases among the diseases.

4. The disease prediction system according to claim 1 , wherein the second information is measured using a SAW (Surface Acoustic Wave) sensor.

5. The disease prediction system according to claim 1 , wherein the second learning information is acquired using a SAW (Surface Acoustic Wave) sensor.

6. The disease prediction system according to any one of claims 1 to 5, wherein the second information is information on the presence or absence or concentration of a hormone-like substance.

7. The disease prediction system according to any one of claims 1 to 6, wherein the learning data is a series of data on different time axes obtained by testing the same person.

8. The disease prediction system according to any one of claims 1 to 7, wherein the first learning information is a series of data of the same site photographed on different time axes.

9. The disease prediction system according to any one of claims 1 to 8, wherein the learning data is a group of data from other people.

10. The disease prediction system according to any one of claims 1 to 9, wherein the training data is bone mass.

11. The disease prediction system according to any one of claims 1 to 10, wherein the prediction unit outputs a numerical value represented by at least one of YAM (Young Adult Mean), T score, and Z score as a prediction result.

12. The disease prediction system according to any one of claims 1 to 10, wherein the prediction unit outputs a KL (Kellgren-Lawrence) grade classification as a prediction result.

13. The disease prediction system according to any one of claims 1 to 10, wherein the prediction unit outputs a future probability of onset of the disease as a prediction result.

14. The disease prediction system according to any one of claims 1 to 13, wherein the prediction unit simultaneously predicts the onset of a plurality of the diseases.

15. The disease prediction system according to any one of claims 1 to 14, further comprising an output unit that outputs the prediction result predicted by the prediction unit.

16. The disease prediction system according to any one of claims 1 to 15, wherein the prediction unit outputs changes over time.

17. The disease prediction system according to any one of claims 1 to 16, wherein the input information includes third information regarding the subject's health condition.

18. The disease prediction system according to claim 17 , wherein the third information includes at least one of age information, gender information, height information, and weight information.

19. The disease prediction system according to any one of claims 1 to 18, wherein the input information includes fourth information regarding an intervention for the subject.

20. The disease prediction system according to claim 19 , wherein the prediction unit predicts a first result based on input information excluding the fourth information and a second result based on input information including the fourth information.

21. The disease prediction system according to any one of claims 1 to 20, wherein the input information includes fifth information having bone metabolic marker information of the subject.

22. The disease prediction system according to any one of claims 1 to 21, wherein the system outputs a diagnosis result of the subject's current condition together with the diagnosis result.

Images