Estimation system, estimation device, and control program

Abstract

To provide an estimation system, estimation device, and control program capable of improving the reliability of estimation results.SOLUTION: In a disease estimation system, an estimation device (3) comprises an approximator (32) that outputs an estimation result related to a disease of a patient and reference data indicating a basis for the estimation result from first data of the patient, and an output section (33) that outputs the estimation result and the reference data. The approximator (32) is learned using learning data including second data of the same type as the first data, and teacher data including diagnostic results corresponding to the learning data. The approximator (32) outputs reference data including third data retrieved from the second data included in the learning data.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to an estimation system, an estimation device, and a control program. [Background technology]

[0002] Conventionally, a diagnosis support device including a neural network has been known (Patent Document 1). [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] Such a diagnosis support device is required to have improved reliability. [Means for solving the problem]

[0005] An estimation system according to one embodiment of the present disclosure includes an approximator that outputs, from first data of a patient, an estimation result related to the patient's disease and reference data showing the basis of the estimation result, and an output unit that outputs the estimation result and the reference data, wherein the approximator is trained using training data including second data of the same type as the first data and teacher data including a diagnostic result corresponding to the training data, and the approximator outputs the reference data including third data searched from the second data included in the training data.

[0006] Furthermore, an estimation device according to one aspect of the present disclosure includes an approximator that outputs, from first data of a patient, an estimation result related to the patient's disease and reference data showing the basis of the estimation result, and an output unit that outputs the estimation result and the reference data, wherein the approximator is trained using training data including second data of the same type as the first data and teacher data including a diagnostic result corresponding to the training data, and the approximator outputs the reference data including third data searched from the second data included in the training data.

[0007] Furthermore, a control program according to one aspect of the present disclosure is a control program that causes a computer to execute the steps of causing an approximator to output, from first data of a patient, an estimation result related to the patient's disease and reference data showing the basis of the estimation result, and an output step of outputting the estimation result and the reference data, wherein the approximator learns using training data including second data of the same type as the first data and teacher data including a diagnostic result corresponding to the training data, and the approximator outputs the reference data including third data searched from the second data included in the training data. [Effects of the Invention]

[0008] The user can check the reliability of the estimation result, thereby improving the reliability of the estimation result. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram illustrating an example of the configuration of a disease prediction system. [Figure 2] FIG. 1 is a diagram illustrating an example of a partial configuration of a disease prediction system. [Figure 3] FIG. 1 is a diagram illustrating an example of a partial configuration of a disease prediction system. [Figure 4] FIG. 1 is a diagram illustrating an example of a partial configuration of a disease prediction system. [Figure 5] FIG. 1 is a diagram illustrating an example of a partial configuration of a disease prediction system. [Figure 6]FIG. 1 is a diagram illustrating an example of the operation of a part of a disease prediction system. [Figure 7] FIG. 1 is a diagram illustrating an example of the operation of a part of a disease prediction system. DETAILED DESCRIPTION OF THE INVENTION

[0010] The disease inference system 1 of the present disclosure can infer a disease from input data I using AI (Artificial Intelligence). The input data I may be, for example, image data. The disease inferred by the present invention may be any disease that a doctor can diagnose based on the image data. For example, it may be cervical cancer or a skin disease.

[0011] FIG. 1 conceptually shows the configuration of a disease inference system 1 of the present disclosure.

[0012] The disease inference system 1 of the present disclosure includes a terminal device 2 and an inference device 3. The terminal device 2 acquires patient input data I used for diagnosing a disease. The inference device 3 can infer the type and location of the disease based on the data acquired by the terminal device 2.

[0013] The terminal device 2 can acquire image data. The input data I acquired by the terminal device 2 is input to the estimation device 3. In the case of cervical cancer, the terminal device 2 may be, for example, a device capable of capturing colposcopy images. In the case of skin diseases, the terminal device 2 may be, for example, a device capable of capturing dermoscopy images. In the case of cervical cancer, the input data I may be, for example, a colposcopy image, and in the case of skin diseases, the input data I may be, for example, a dermoscopy image.

[0014] The terminal device 2 may have a communication unit 21 and may transfer the input data I to the estimating device 3 after acquiring the input data I.

[0015] The terminal device 2 does not have to directly transfer the input data I to the estimation device 3. In this case, for example, the input data I acquired by the terminal device 2 may be stored in a storage medium, and the input data I may be input to the estimation device 3 via the storage medium.

[0016] FIG. 2 conceptually shows the configuration of the estimation device 3 according to this embodiment.

[0017] The estimation device 3 can estimate the patient's disease from the input data I input to the estimation device 3. For example, when the disease estimation system 1 estimates cervical cancer, the estimation device 3 can estimate the type of cervical cancer and the location of the lesion in the patient from the input data I acquired by the terminal device 2, and output the estimated estimation result O.

[0018] The estimation device 3 has an input unit 31, an approximator 32, and an output unit 33. The input unit 31 receives input data I from the terminal device 2. The approximator 32 performs a predetermined calculation on the input data I to estimate the disease classification and the lesion site. The output unit 33 outputs the estimation result O estimated by the approximator 32.

[0019] As described above, the input data I is input to the input unit 31. The input unit 31 of the present disclosure may be any communication device that can be connected to the communication unit 21 of the terminal device 2 via a network. The input unit 31 may also be a port, such as a USB port, that can receive the input data I via a storage medium. The input unit 31 may also be provided with an input device that can input information other than the input data I. The input device may be, for example, a keyboard, a touch panel, a mouse, or a USB port.

[0020] The approximator 32 can predict a patient's disease from the input data I input to the input unit 31. The approximator 32 is a so-called AI (Artificial Intelligence). Specifically, the approximator 32 has hardware configured with a plurality of electronic components and circuits to function as an AI.

[0021] As described above, the approximator 32 has a plurality of electronic components and circuits. In other words, a portion of the approximator 32 is composed of 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.

[0022] Furthermore, the learned approximator 32 performs calculations according to a trained model that has already been trained when inferring a disease. That is, the disease inference system 1 uses the training data through the approximator 32 to acquire in advance parameters and the like required for the calculations of the approximator 32. As a result, the approximator 32 can calculate an inference result O from input data I. Note that the training data or training data may be data that corresponds to the input data I input to the estimation device 3 and the inference result O output from the estimation device 3.

[0023] FIG. 3 conceptually illustrates the processing performed by the approximator 32 of the present disclosure.

[0024] The approximator 32 according to this embodiment is, for example, a CNN. The approximator 32 includes, as functional units, an input layer 3a, a hidden layer 3b, and an output layer 3c. The hidden layer 3b is also called, for example, an intermediate layer. The input layer 3a, the hidden layer 3b, and the output layer 3c each include a plurality of nodes 3d and a plurality of edges 3e connecting the plurality of nodes 3d together, and together they form a so-called neural network. As described above, each of the functional units can be realized by a plurality of electronic components, circuits, and the like included in the approximator 32.

[0025] Note that the connection relationships between the plurality of nodes 3d and the plurality of edges 3e are optimized for disease estimation by applying the trained model to the approximator 32. That is, the trained approximator 32 can perform calculations using parameters suitable for disease estimation.

[0026] The input layer 3a is a layer that receives input data I. Each node 3d in the input layer 3a receives data that constitutes the input data I. For example, if the input data I is image data, a numerical value indicating the color or shade of each pixel of the image data is input to each node 3d.

[0027] 4 and 5 conceptually show the configuration of a portion of the hidden layer 3b of the present disclosure.

[0028] Hidden layer 3b can perform calculations based on the information of the image data input to input layer 3a. Hidden layer 3b has convolutional layer 3b1, pooling layer 3b2, and fully connected layer 3b3. Hidden layer 3b can extract features from the data input to input layer 3a using convolutional layer 3b1, pooling layer 3b2, and fully connected layer 3b3.

[0029] The convolutional layer 3b1 can convolve the data by filtering the data input to each node 3d in the immediately preceding layer. In other words, it can reduce the data while maintaining the features of the input data I. Specifically, it performs calculations on the numerical values ​​input to each node 3d in the immediately preceding layer based on the filter coefficients obtained in the learning process, thereby reducing the data while maintaining the features. Each of the multiple convolutional layers 3b1 performs calculations based on its own filter coefficients.

[0030] The pooling layer 3b2 is located after the convolutional layer 3b1. The pooling layer 3b2 can reduce the dimensionality of the data in the convolutional layer 3b1 while maintaining the characteristics of the data in the convolutional layer 3b1 based on the numerical values ​​of each node 3d calculated in the immediately preceding convolutional layer 3b1. For example, the pooling layer 3b2 can extract the largest value within a given range from the numerical values ​​of each node 3d in the immediately preceding convolutional layer 3b1.

[0031] One convolutional layer 3b1 and one pooling layer 3b2 form one set. Alternatively, the hidden layer 3b may have multiple convolutional layers 3b1 and multiple pooling layers 3b2. In this case, the multiple convolutional layers 3b1 and multiple pooling layers 3b2 are arranged alternately, forming multiple sets of convolutional layers 3b1 and pooling layers 3b2 in the hidden layer 3b, thereby gradually narrowing down the features of the input data I. Note that the number of multiple convolutional layers 3b1 and multiple pooling layers 3b2 may be any number that allows the approximator 32 to function with the required accuracy.

[0032] The fully connected layer 3b3 can extract features of the input data I from the data input to each node 3d of the immediately preceding pooling layer 3b2. Specifically, it can extract features by performing calculations on the numerical values ​​input to each node 3d of the immediately preceding layer based on the weighting coefficients obtained in the learning process.

[0033] The output layer 3c can output an estimation result O based on the numerical value of each node 3d calculated in the fully connected layer 3b3. Specifically, for example, the degree of possibility of disease can be output based on the numerical value pattern of each node 3d in the fully connected layer 3b3.

[0034] The output unit 33 can display the estimation result O estimated by the output layer 3c. The output unit 33 is, for example, a liquid crystal display or an organic EL display. The output unit 33 can display various types of information such as characters, symbols, and figures. The output unit 33 can display, for example, numbers or images.

[0035] The estimation result O is output as, for example, a character string.

[0036] The estimation device 3 further includes a control unit 34. The control unit 34 can comprehensively manage the operation of the estimation device 3 by controlling the other components of the estimation device 3. That is, the control unit 34 can control the calculations of the approximator 32, etc. The control unit 34 includes multiple electronic components and circuits, and constitutes, for example, a CPU (Central Processing Unit). The multiple 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.

[0037] The estimation device 3 further includes a storage unit 35. The storage unit 35 can store programs for controlling the estimation device 3. The storage unit 35 may be, for example, a read-only memory (ROM) or a hard-disk drive (HDD). The storage unit 35 may also store input data I, calculation results of the approximator on the learning data, teacher data, and a learned model (learned parameters).

[0038] <Examples of input data, learning data, and teacher data> The input data I may be data such as patient image data used for diagnosing the disease. The input data I may be, for example, a colposcopy image. From the colposcopy image, for example, a classification of cervical cancer can be estimated. Furthermore, if the input data I is, for example, a dermoscopy image, a skin disease can be estimated.

[0039] The input data I is an image of the area to be diagnosed. For example, if cervical cancer is to be inferred, the input data I should be an image of the cervix. If a skin disease is to be inferred, the input data I should be an image of the skin including the lesion.

[0040] The training data includes the same type of data as the first input data I1. For example, if the first input data I1 is a colposcopy image, the training data may also include a colposcopy image.

[0041] The training data includes a diagnosis of a disease corresponding to the training data. For example, when estimating cervical cancer, the training data includes a diagnosis corresponding to each of a plurality of training image data. The diagnosis may be evaluated at approximately the same time as the training image data was captured.

[0042] For example, in the case of cervical cancer, the training data may be a doctor's diagnosis.

[0043] <Neural network learning example> The approximator 32 is optimized by machine learning using training data and teacher data so that it can calculate an estimation result O from input data I. That is, before the learning process, the approximator 32 is trained by machine learning by calculating a pseudo estimation result from training data based on an unlearned mathematical model and adjusting parameters within the approximator 32 so that the difference between the pseudo estimation result and the teacher data becomes small. As a result, the approximator 32 can perform a calculation on the input data I based on the trained model and output an estimation result O.

[0044] The parameter adjustment method may be, for example, backpropagation. The parameters may be, for example, filter coefficients applied to the operations performed in the multiple convolutional layers 3b1, or weighting coefficients applied to the operations performed in the fully connected layer 3b3.

[0045] FIG. 6 shows the calculation sequence of the approximator 32 according to this embodiment.

[0046] In the disease inference system 1, a disease can be inferred by performing calculations in the approximator 32 based on input data I. Furthermore, in the disease inference system 1 according to the present invention, the approximator 32 can search for an inference result O of the inferred disease and reference data R similar to the inference result O as the basis for the inference result O, and output the search results.

[0047] For example, conventional diagnostic support devices use neural networks to assist in disease diagnosis, but they simply output the results of the neural network estimation, making it impossible to confirm the reliability of the estimation results. In other words, since it is difficult for humans to understand the process by which results are derived in diagnostic support devices using neural networks (a so-called black box), diagnostic support devices are required to show users the basis for deriving the results. Furthermore, they are also required to detect and notify users when data that is not relevant to the diagnosis is input.

[0048] In contrast, in the disease inference system 1 according to the present invention, as described above, the approximator 32 searches for the inference result O of the inferred disease and similar reference data R, and the output unit O outputs the search result together with the inference result O. That is, the disease inference system 1 according to the present invention can output the reference data R, for example, if similar reference data R is available, and on the other hand, if similar reference data R is not available, it can output a message indicating that there is no reference data R. Therefore, the reliability of the inference result O of the disease inference system 1 can be confirmed.

[0049] FIG. 7 shows in more detail the order of calculations after the estimation result O of the approximator 32 according to this embodiment is calculated.

[0050] In the disease inference system 1 according to this embodiment, the approximator 32 may search for reference data R from the training data. Specifically, the approximator 32 may search for training data similar to the inference result O by comparing the numerical values ​​of each node 3d in the fully connected layer 3b3 during the calculation of the input data I with the numerical values ​​of each node 3d in the fully connected layer 3b3 of the training data during training.

[0051] That is, in this case, in the disease inference system 1 according to this embodiment, after the learning process is completed, the approximator 32 performs calculation processing again using the learning data, and the numerical values ​​calculated in the process and information on the teacher data are stored in the storage unit 35. As a result, the numerical values ​​calculated by the approximator 32 in the process of calculation processing for the data input to the input unit 31 and the teacher data that have a high degree of coincidence with the numerical values ​​stored in the storage unit 35 can be displayed to the user together with the inference result O. The user can confirm the reliability of the inference result by visually comparing the displayed teacher data with the input data I.

[0052] The similarity determination may be made by comparing the numerical values ​​of each node 3d. For example, when comparing each node 3d, if 80% or more of the nodes 3d to be compared have a matching rate of ±20% or less for each numerical value of each node 3d, they may be determined to match.

[0053] The approximator 32 may also search for training data similar to the estimation result O by comparing the numerical values ​​of each node 3d in the fully connected layer 3b3 and each pooling layer 3b2 when calculating the input data I with the numerical values ​​of the training data during training.

[0054] The approximator 32 outputs the learning data as the reference data R, and may also output teacher data related to the learning data that is output.

[0055] The approximator 32 may output the learning data and the teacher data as the reference data R, and may also output personal data accompanying the learning data. The personal data may be, for example, age or gender.

[0056] The approximator 32 may output the learning data and the teacher data as the reference data R, and may also output diagnostic information about the teacher data. The diagnostic information may be information about the person diagnosing the problem, the diagnostic device, etc.

[0057] The approximator 32 may output the estimation result O as the same type of data as the input data I. That is, for example, if the input data I is image data, the image data may be output as the estimation result O. In this case, the approximator 32 needs to be trained in order to output the image data as the estimation result O. In this case, for example, a ConvLSTM network that combines a CNN (Convolutional Neural Network) and an LSTM (Long Short-Term Memory) may be applied as the approximator 32.

[0058] The approximator 32 may output image data of the learning data as the search result O', which makes it easier to compare the input data I and the learning data.

[0059] When the estimation result O and the search result O' are output as image data, the approximator 32 may emphasize the difference between the estimation result O and the search result O'.

[0060] The present invention is not limited to the above-described embodiments, 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.

[0061] For example, in the above example, when a neural network is used as the approximator 32, an example in which a CNN is applied has been described, but the present invention is not limited to this. For example, the approximator 32 may use a ConvLSTM network that combines a CNN (Convolutional Neural Network) and an LSTM (Long Short-Term Memory), an RNN (Recurrent Neural Network), or a GAN (Generative Adversarial Network). The approximator 32 may also combine multiple neural networks. Specifically, it may be a composite neural network that combines a ConvLSTM network and a convolutional neural network.

[0062] [Other Aspect 1] A disease inference system according to one aspect of the present disclosure is a system capable of inferring a disease from an image. The disease inference system includes an input unit and an approximator. The input unit is capable of inputting input data. The approximator is subjected to a learning process using training data including data of the same type as the input data and teacher data including diagnostic results related to the training data. The approximator is capable of searching for an inference result of the disease inferred from the input data input to the input unit and reference data similar to the inference result, and outputting the search results.

[0063] [Other Aspect 2] The present invention also includes the following aspects.

[0064] An estimation device according to one aspect of the present disclosure includes an input unit capable of receiving input data, and an approximator capable of outputting an estimation result related to a lesion site from the input data, wherein the approximator is trained using training data including data of the same type as the input data and teacher data including a diagnostic result related to the training data, and outputs the estimation result and reference data similar to the estimation result, which is a search result searched as the basis for the estimation result.

[0065] Furthermore, a control program according to one aspect of the present disclosure causes a computer to execute an input step of inputting the input data into an approximator that has undergone a learning process using learning data including data of the same type as the input data and teacher data including diagnostic results related to the learning data, and that is capable of estimating an inference result related to a lesion site, and an output step of outputting the inference result and reference data similar to the inference result, which are search results searched as the basis for the inference result.

[0066] [Other Aspect 3] The present invention also includes the following aspects.

[0067] An estimation system according to one embodiment of the present disclosure includes an approximator that outputs an estimation result related to a disease from first data, and an output unit that outputs the estimation result and reference data showing the basis for the estimation result, and the approximator is trained using training data including second data of the same type as the first data, and teacher data including a diagnostic result related to the training data.

[0068] In addition, an estimation device according to one embodiment of the present disclosure includes an approximator that outputs an estimation result related to a disease from first data, and an output unit that outputs the estimation result and reference data showing the basis for the estimation result, and the approximator is trained using training data including second data of the same type as the first data, and teacher data including a diagnostic result related to the training data.

[0069] Furthermore, a control program according to one aspect of the present disclosure is a control program that causes a computer to execute the steps of causing an approximator to output an inference result related to a disease from the first data, and an output step of outputting the inference result and reference data showing the basis for the inference result, wherein the approximator is trained using training data including second data of the same type as the first data, and teacher data including a diagnostic result related to the training data. [Explanation of symbols]

[0070] 1 Disease Prediction System (Prediction System) 2. Terminal Device 21 Communications Department 3 Estimation device 31 Input section 32 Approximator 33 Output section 34 Control Unit 35 Storage section 3a Input layer 3b Hidden layer 3b1 Convolutional Layer 3b2 pooling layer 3b3 fully connected layer 3c output layer I Input data (first data) O Estimation result O´ Search results R Reference Data

Claims

1. an approximator that outputs, from first data of a patient, an estimation result related to the patient's disease and reference data indicating the basis of the estimation result; an output unit that outputs the estimation result and the reference data; Equipped with the approximator is trained using training data including second data of the same type as the first data, and teacher data including a diagnosis result corresponding to the training data, the approximator outputs the reference data including third data searched from the second data included in the learning data. Estimation system.

2. The estimation system according to claim 1 , wherein the reference data includes training data related to the third data.

3. The estimation system according to claim 1 , wherein the reference data includes personal data associated with the third data.

4. The estimation system according to claim 1 , wherein the reference data includes diagnostic information of training data related to the third data.

5. The estimation system according to claim 1 , wherein the first data is image data.

6. The estimation system according to claim 1 , wherein the first data includes an image of a region to be diagnosed.

7. The estimation system according to claim 1 , wherein the estimation result is image data.

8. the estimation result is image data of the same type as the first data, the third data is image data, The estimation system of claim 7 , wherein the approximator emphasizes differences between the estimation result and the reference data.

9. an approximator that outputs, from first data of a patient, an estimation result related to the patient's disease and reference data indicating the basis of the estimation result; an output unit that outputs the estimation result and the reference data; Equipped with the approximator is trained using training data including second data of the same type as the first data, and teacher data including a diagnosis result corresponding to the training data, the approximator outputs the reference data including third data searched from the second data included in the learning data. Estimation device.

10. On the computer, causing the approximator to output, from the first data of the patient, an estimation result related to the patient's disease and reference data indicating a basis for the estimation result; an output step of outputting the estimation result and the reference data; A control program for executing the approximator is trained using training data including second data of the same type as the first data, and teacher data including a diagnosis result corresponding to the training data, the approximator outputs the reference data including third data searched from the second data included in the learning data. Control program.

Images