DIAGNOSIS SUPPORT DEVICE, COMPUTER PROGRAM, AND METHOD FOR ESTIMATING GAS DIFFUSION CAPACITY

Abstract

To provide a technique for estimating the gas diffusion ability of a subject without requiring a detailed respiratory function inspection.SOLUTION: In a diagnosis support system 10, a diagnosis support device 14 stores a model which is constructed according to machine learning based on gas diffusion abilities being results of basic respiratory function inspections of multiple subjects and results of detailed respiratory function inspections of multiple subjects, for the multiple subjects having been subjected to the basic respiratory function inspections and the detailed respiratory function inspections, and which outputs a gas diffusion ability value corresponding to a result of the basic respiration function inspection that has been input. The diagnosis support device 14 receives a result of the basic respiratory inspection of a subject, inputs the inspection result to the model, and acquires a gas diffusion ability value that is output from the model. The diagnosis support device 14 outputs information related to the gas diffusion ability value that has been output from the model.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present disclosure relates to data processing technology, and in particular to a diagnosis support device, a computer program, and a method for estimating gas diffusion capacity. [Background technology]

[0002] Spirograms are known as a basic respiratory function test. Spirograms measure the inhaled and exhaled volumes of a subject taking a deep breath to examine the subject's breathing ability. On the other hand, detailed respiratory function tests are sometimes performed to accurately identify respiratory diseases and correctly evaluate the condition. Detailed respiratory function tests measure gas diffusion capacity. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Special Publication No. 2016-526466 Summary of the Invention [Problem to be solved by the invention]

[0004] Precision respiratory function testing requires relatively large dedicated equipment, takes a long time (about 1 hour), places a heavy burden on the subject, and is associated with a high risk of infection, making it difficult to perform frequently. This has led to problems such as delays in referrals from non-respiratory medicine doctors and private practice physicians to specialists, and even respiratory medicine doctors are slow to understand the progression of the disease.

[0005] The present disclosure has been made in light of these circumstances, and one object of the present disclosure is to provide a technique for estimating a subject's gas diffusion capacity without requiring a precise respiratory function test. [Means for solving the problem]

[0006] In order to solve the above problems, a diagnostic support device of one embodiment of the present disclosure is a model constructed by machine learning based on the results of a first respiratory function test of multiple subjects who have undergone a first respiratory function test that is relatively easy to perform and a second respiratory function test that is relatively difficult to perform, and the gas diffusion capacity that is the result of the second respiratory function test of the multiple subjects, and is a diagnostic support device that can access a memory unit that stores the model and outputs a value of gas diffusion capacity corresponding to the input result of the first respiratory function test, and is equipped with a reception unit that receives the result of the first respiratory function test of a certain subject, an acquisition unit that inputs the result of the first respiratory function test into the model and acquires the value of gas diffusion capacity output from the model, and an output unit that outputs information regarding the acquired value of gas diffusion capacity.

[0007] Another aspect of the present disclosure is a gas diffusion capacity estimation method. This method is a model constructed by machine learning based on the results of the first respiratory function test of a plurality of subjects who have undergone a first respiratory function test that is relatively easy to perform and a second respiratory function test that is relatively difficult to perform, and the gas diffusion capacity that is the result of the second respiratory function test of the plurality of subjects, and a computer that can access a storage unit that stores the model that outputs a value of gas diffusion capacity corresponding to the input result of the first respiratory function test performs the steps of accepting the result of the first respiratory function test of a certain subject, inputting the result of the first respiratory function test into the model to obtain the value of gas diffusion capacity output from the model, and outputting information on the obtained value of gas diffusion capacity.

[0008] Any combination of the above components and any conversion of the expressions of the present disclosure between a system, a computer program, a recording medium storing a computer program, etc. are also valid as aspects of the present disclosure. Effect of the Invention

[0009] According to the present disclosure, a technique can be provided for estimating a subject's gas diffusion capacity without requiring a precise respiratory function test. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a diagram illustrating a configuration of a diagnosis support system according to a first embodiment. [Diagram 2] 2(a) and 2(b) are diagrams showing an example of pre-processing for basic data of subjects. [Diagram 3] 3(a) and 3(b) are diagrams showing an example of preprocessing of the results of a precise respiratory function test of a subject. [Figure 4] FIG. 2 is a diagram showing a network configuration of a pathological condition index derivation model according to the first embodiment. [Diagram 5] FIG. 13 is a diagram showing an example of pre-processing for an X-ray image of a subject. [Figure 6] FIG. 13 is a diagram showing an example of pre-processing for an X-ray image of a subject. [Figure 7] FIG. 13 is a diagram showing a network configuration of a pathological condition index derivation model according to a second embodiment. [Figure 8] 8(a) and 8(b) are diagrams showing the estimation results of the pathological index derivation model. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] In the embodiment, a diagnosis support system is proposed, which is an information processing system that supports the diagnosis of respiratory diseases. In the diagnosis support system of the embodiment, the gas diffusion capacity of a subject, which is the result of a detailed respiratory function test, is estimated based on the result of a basic respiratory function test (hereinafter also referred to as a "basic respiratory function test"). The basic respiratory function test can be said to be a respiratory function test that is relatively easy to perform, and is also called a spirogram. The detailed respiratory function test can be said to be a respiratory function test that is relatively difficult to perform. Gas diffusion capacity is also called "pulmonary diffusion capacity" or "DLCO", and is the capacity to transport gas from the alveoli to red blood cells via the alveolar epithelium and capillary endothelium. In the embodiment, DLCO% indicating the gas transport ratio is used as an index value of gas diffusion capacity. The larger the DLCO%, the better the gas diffusion capacity is judged to be.

[0012] <First Example> 1 shows the configuration of a diagnosis support system 10 of the first embodiment. The diagnosis support system 10 includes a model generating device 12, a diagnosis support device 14, and a user device 16. Each of these devices may be stored in a single housing, and may be connected to each other via a communication network such as a LAN, a WAN, or the Internet so that they can transmit and receive data to each other.

[0013] The model generating device 12 is an information processing device (in other words, a computer) that generates a model (hereinafter also referred to as a "pathological index derivation model") that outputs a value of gas diffusion capacity (DLCO%) corresponding to the input result of the basic respiratory function test based on the results of the basic respiratory function test for multiple subjects and the value of gas diffusion capacity (DLCO%) obtained as a result of the detailed respiratory function test for the multiple subjects by machine learning. The pathological index derivation model may be a mathematical model, and in this embodiment, it is a neural network.

[0014] The diagnostic support device 14 is an information processing device that estimates a gas diffusion capacity (DLCO%) corresponding to the result of the basic respiratory function test of the subject using the pathological index derivation model generated by the model generation device 12, and outputs information related to the gas diffusion capacity (DLCO%). The user device 16 is an information processing device that inputs the result of the basic respiratory function test of the subject to the diagnostic support device 14, and displays information related to DLCO (DLCO%) output from the diagnostic support device 14 on a display device. The user device 16 may be a PC, tablet terminal, or smartphone operated by a medical professional such as a doctor or clinical laboratory technician.

[0015] The model generating device 12 of FIG. 1 will now be described in detail. 1 includes a block diagram showing the functional blocks of the model generating device 12. Each block shown in the block diagram in this specification can be realized in terms of hardware by elements such as a computer processor, CPU, and memory, electronic circuits, and mechanical devices, and in terms of software by a computer program, etc., but here, functional blocks realized by cooperation between them are depicted. Therefore, it will be understood by those skilled in the art that these functional blocks can be realized in various forms by combinations of hardware and software.

[0016] The model generating device 12 includes an inspection information storage unit 20, a reading unit 22, a preprocessing unit 24, a model generating unit 26, and a model output unit 28. A computer program (e.g., a model generating program) in which the functions of at least some of the multiple functional blocks are implemented may be stored in a non-transient predetermined recording medium and installed in the storage of the model generating device 12 via the recording medium. Alternatively, the model generating program may be downloaded via a network and installed in the storage of the model generating device 12. The CPU of the model generating device 12 may fulfill the functions of the multiple functional blocks shown in FIG. 1 by reading the model generating program into a main memory and executing it.

[0017] The test information storage unit 20 stores the basic data of each subject and the result of the precise respiratory function test (DLCO% in this embodiment) for each subject who underwent both the basic respiratory function test and the precise respiratory function test, in association with each other. The basic data includes four items of data (height, weight, age, and sex) that indicate physical information. The basic data further includes 19 items of data obtained as a result of the basic respiratory function test.

[0018] The 19 items of data from the basic respiratory function test were VC (vital capacity, unit L (liters)), TV (tidal volume, unit L), ERV (expiratory reserve, unit L), IRV (inspiratory reserve, unit L), IC (maximum inspiratory volume, unit L), FRC (functional residual capacity, unit L), RV (residual volume, unit L), TLC (total lung capacity, unit L), RV / TLC (residual volume, unit %), FVC (forced vital capacity, unit L), FEV1.0 (forced expiratory capacity in 1 second), and These include FEV1.0%(G) (Gaenlser's 1 second rate, %), FEV1.0%(T) (Tiffenaeu's 1 second rate, %), MMEF (maximum expiratory flow ratio, L / sec), AT (inflection point (anaerobic threshold), %), PEF (maximum expiratory flow rate, L / sec), V50 (50% expiratory rate, L / sec), V25 (25% expiratory rate, L / sec), and V25 / height (L / sec / meter).

[0019] The reading unit 22 reads basic data of a plurality of subjects and the results of the precise respiratory function test (DLCO%) stored in the test information storage unit 20. The preprocessing unit 24 performs preprocessing on the basic data read by the reading unit 22 and preprocessing on the results of the precise respiratory function test (DLCO%).

[0020] Specifically, the pre-processing unit 24 standardizes the average value and standard deviation of each item of basic data (4 items of physical information + 19 items of basic respiratory function test results) of 23 items of subjects, and obtains the standardized results as learning data. Specifically, the pre-processing unit 24 converts the value of each item of data so that the average value of the data of each item is 0 and the standard deviation is 1.

[0021] Figures 2(a) and 2(b) show examples of preprocessing of a subject's basic data. Figure 2(a) shows an example of the distribution of the original basic data. The horizontal axis of Figure 2(a) is the data value (e.g., the measurement value in a basic respiratory function test), and the vertical axis of Figure 2(a) is the number of data (frequency). Figure 2(b) shows the distribution of the data after preprocessing. The horizontal axis of Figure 2(b) is the deviation, and the vertical axis of Figure 2(b) is the number of data (frequency). By standardizing the average value and standard deviation of the data for each item of basic data, it is possible to correct the disparity in the data for each item and improve the estimation accuracy of the pathology index derivation model.

[0022] In addition, the preprocessing unit 24 obtains the natural logarithm of the value of gas diffusion capacity (DLCO%), which is the result of the precise respiratory function test, as learning data. Figures 3(a) and 3(b) show examples of preprocessing for gas diffusion capacity (DLCO%), which is the result of the precise respiratory function test of the subject. Figure 3(a) shows an example of the distribution of the original gas diffusion capacity. The horizontal axis of Figure 3(a) is the value of gas diffusion capacity (DLCO%) (measured value in the precise respiratory function test), and the vertical axis of Figure 3(a) is the number of data (frequency). Figure 3(b) shows the distribution of data after preprocessing. The horizontal axis of Figure 3(b) is the natural logarithm of the value of gas diffusion capacity (DLCO%) (measured value in the precise respiratory function test), and the vertical axis of Figure 3(b) is the number of data (frequency). The preprocessing makes the distribution of gas diffusion capacity closer to a normal distribution, thereby improving the estimation accuracy of the pathological index derivation model.

[0023] 1, the model generating unit 26 generates a pathological condition index derivation model by machine learning based on the basic data (basic data after preprocessing) of each of the multiple subjects and the gas diffusion capacity (value of the gas diffusion capacity after preprocessing) of each of the multiple subjects. In the first embodiment, the model generating unit 26 constructs a neural network that outputs a gas diffusion capacity (DLCO%) corresponding to the input basic data (basic data after preprocessing) as the pathological condition index derivation model.

[0024] FIG. 4 shows the network configuration of the pathology index derivation model of the first embodiment. The pathology index derivation model of the first embodiment is a 10-layer neural network including an input layer, a fully connected layer, a batch normalization layer, an activation layer, and a dropout layer. The input layer 40 of FIG. 4 accepts the input of 23 items of basic data after preprocessing. In the next fully connected layer, 23 neurons corresponding to the input 23 items of basic data are fully connected to 1024 neurons. The last activation layer 42 can be said to be an output layer, and outputs one DLCO% index value. The activation function used in the activation layers other than the last one of the multiple activation layers arranged in the pathology index derivation model is "relu", and the activation function used in the last activation layer 42 is "linear". The dropout rate of the dropout layer is 0.5.

[0025] The model generation unit 26 may generate a pathology index derivation model (neural network) by using a known library or framework such as Keras. The network configuration of the pathology index derivation model, the weights and biases between neurons, the type of activation function, and the dropout rate may be appropriately set or selected based on the developer's knowledge, experiments using validation data and test data, and the like.

[0026] Returning to FIG. 1, the model output unit 28 transmits data of the pathological index derivation model generated by the model generation unit 26 to the diagnosis support device 14, and stores the data of the pathological index derivation model in the model storage unit 30 of the diagnosis support device 14.

[0027] Next, the diagnosis support device 14 in FIG. 1 will be described in detail. The diagnostic assistance device 14 includes a model storage unit 30, a test information receiving unit 32, a preprocessing unit 34, a DLCO acquisition unit 36, and a diagnostic assistance information output unit 38. A computer program (e.g., a diagnostic assistance program) in which the functions of at least some of the functional blocks among these functional blocks are implemented may be stored in a non-temporary predetermined recording medium and installed in the storage of the diagnostic assistance device 14 via the recording medium. Alternatively, the diagnostic assistance program may be downloaded via a network and installed in the storage of the diagnostic assistance device 14. The CPU of the diagnostic assistance device 14 may fulfill the functions of the functional blocks shown in FIG. 1 by reading the diagnostic assistance program into a main memory and executing it.

[0028] The model storage unit 30 stores data of the pathology index derivation model provided by the model generating device 12. As a modified example, the model generating device 12 may include a storage unit that stores the pathology index derivation model, and the diagnosis support device 14 may refer to the pathology index derivation model stored in the model generating device 12 via a communication network. That is, the diagnosis support device 14 only needs to be able to access the storage unit that stores the pathology index derivation model, in other words, it only needs to be configured to be able to refer to the pathology index derivation model stored in a local or remote storage unit.

[0029] The test information receiving unit 32 receives basic data (physical information and 23 items of data that are the results of a basic respiratory function test) of a subject (hereinafter also referred to as a "target subject") whose gas diffusion capacity (DLCO%) is to be estimated, transmitted from the user device 16. The target subject is typically different from the subject who is the provider of the basic data and gas diffusion capacity used to generate the pathological index derivation model. In addition, the target subject is typically a subject for whom an estimate of gas diffusion capacity is to be obtained from the results of a basic respiratory function test without performing a precise respiratory function test.

[0030] The preprocessing unit 34 corresponds to the preprocessing unit 24 of the model generating device 12. As a first processing unit, the preprocessing unit 34 standardizes the average value and standard deviation of the data of each item of the 23 items of basic data accepted by the test information accepting unit 32 as a preprocessing of data input to the pathological index derivation model. For example, the preprocessing unit 34 converts the value of each item of the basic data of the target subject, using the basic data of the target subject and the basic data of other subjects as a population, so that the average value of the data of each item of the basic data becomes 0 and the standard deviation becomes 1. The basic data of other subjects may be, for example, basic data of subjects other than the target subject input to the diagnosis support device 14, or may be basic data of subjects used when creating the pathological index derivation model.

[0031] The DLCO acquisition unit 36 ​​inputs the basic data standardized by the preprocessing unit 34 to the pathology index derivation model stored in the model storage unit 30. The DLCO acquisition unit 36 ​​acquires an index value of gas diffusibility corresponding to the input basic data output from the pathology index derivation model. The estimated value of gas diffusibility output from the pathology index derivation model is the natural logarithm of gas diffusibility (DLCO%). Therefore, based on the index value of gas diffusibility output from the pathology index derivation model, the DLCO acquisition unit 36 ​​returns it to the original scale by raising the index value to the exponent with e (Napier's constant) as the base. The DLCO acquisition unit 36 ​​passes the result of raising the index value to the exponent with e as the base to the diagnosis support information output unit 38 as the estimated value of the gas diffusibility of the subject.

[0032] The diagnostic assistance information output unit 38 outputs diagnostic assistance information related to the estimated value of gas diffusivity (DLCO%) output from the DLCO acquisition unit 36 ​​to an external device (in this embodiment, the user device 16 which is the basic data input source). The diagnostic assistance information may include the estimated value of gas diffusivity (DLCO%) itself. The diagnostic assistance information may also include a reference value of gas diffusivity (DLCO%), or may include information indicating a comparison result between the estimated value of gas diffusivity (DLCO%) and the reference value.

[0033] As a modified example, the diagnostic assistance information output unit 38 may store the diagnostic assistance information in a storage device (not shown) local to the diagnostic assistance device 14. Alternatively, the diagnostic assistance information output unit 38 may store the diagnostic assistance information in a remote storage device (not shown) via a communication network. The diagnostic assistance information stored in these storage devices may be provided to the user device 16 in response to a request from the user device 16. In addition, if the diagnostic assistance device 14 includes a display device (not shown) or is connected to a display device, the diagnostic assistance information output unit 38 may cause the display device to display the diagnostic assistance information.

[0034] The operation of the diagnosis support system 10 having the above configuration will be described below. First, the operation of generating a pathology index derivation model will be described. The reading unit 22 of the model generating device 12 reads the basic data and the results of the precise respiratory function test (DLCO%) of multiple subjects stored in the test information storage unit 20. The preprocessing unit 24 of the model generating device 12 standardizes the values ​​of each item of the basic data of the multiple subjects, and obtains the natural logarithm of the DLCO% of the multiple subjects.

[0035] The model generating unit 26 of the model generating device 12 identifies a plurality of pairs of standardized basic data and natural logarithms of DLCO%, which are corresponding to a plurality of subjects. The model generating unit 26 executes machine learning based on the plurality of pairs of data, and generates a pathology index derivation model (neural network) by determining weights and biases between neurons so as to output an index value of DLCO% corresponding (consistent) with the basic data of each subject. The model output unit 28 of the model generating device 12 transmits data of the pathology index derivation model to the diagnosis support device 14, and stores the data of the pathology index derivation model in the diagnosis support device 14.

[0036] The test information (basic data and DLCO%) of the subject stored in the test information storage unit 20 of the model generating device 12 may be updated as needed. For example, the user device 16 may store new test information in the test information storage unit 20 of the model generating device 12 as needed. The model generating unit 26 of the model generating device 12 may generate a new pathology index derivation model having a different weight or bias between neurons from the previous pathology index derivation model based on the updated test information. The model output unit 28 may transmit the new pathology index derivation model to the diagnosis support device 14 to update the pathology index derivation model stored in the model storage unit 30 of the diagnosis support device 14. In this manner, the update of the pathology index derivation model may be repeatedly performed.

[0037] Next, an operation for estimating the gas diffusivity of a subject using the pathology index derivation model will be described. The user device 16 transmits basic data of a subject whose gas diffusion capacity is to be estimated to the diagnostic support device 14 in response to an operation by a medical professional or the like. The basic data of the subject is data on 23 items including the results of a basic respiratory function test measured by a predetermined testing device (e.g., a spirometer) and physical information. The diagnostic support device 14 may expose an API (such as a Web service) of a diagnostic support service to an external device. The user device 16 may cause the diagnostic support device 14 to execute the diagnostic support service (specifically, the process described below) by calling the API using the subject's basic data as an argument.

[0038] The test information receiving unit 32 of the diagnostic support device 14 receives the basic data (23 items) of the target subject transmitted from the user device 16. The preprocessing unit 34 of the diagnostic support device 14 standardizes the value of each item of the basic data of the target subject. The DLCO acquisition unit 36 ​​of the diagnostic support device 14 inputs the standardized value of the basic data (23 items) of the target subject to the pathological index derivation model, and acquires the index value of DLCO% corresponding to the input value from the pathological index derivation model. The DLCO acquisition unit 36 ​​obtains an estimated value of DLCO% of the target subject by exponentiating the index value of DLCO% acquired from the pathological index derivation model with e as the base. The diagnostic support information output unit 38 of the diagnostic support device 14 provides diagnostic support information related to the estimated value of DLCO% of the target subject to the user device 16 for display.

[0039] According to the diagnosis support system 10 of the first embodiment, the gas diffusion capacity of a subject can be estimated from the results of a basic respiratory function test, without requiring a detailed respiratory function test, which is difficult to perform. This reduces the burden on subjects suspected of having a lung disease, and supports rapid identification of the disease and evaluation of the condition.

[0040] <Second Example> The second embodiment of the present disclosure will be described with a focus on the differences from the first embodiment, and a description of the commonalities will be omitted. The features of the second embodiment can be arbitrarily combined with the features of the first embodiment and the features of the modified examples. The components of the second embodiment that are the same as or correspond to the components of the first embodiment will be described with the same reference numerals as appropriate.

[0041] The configuration of the diagnosis support system 10 of the second embodiment is similar to the configuration of the diagnosis support system 10 of the first embodiment. Moreover, the functional blocks of the model generating device 12 of the second embodiment are similar to the functional blocks of the model generating device 12 of the first embodiment, and the functional blocks of the diagnosis support device 14 of the second embodiment are similar to the functional blocks of the diagnosis support device 14 of the first embodiment.

[0042] The pathological condition index derivation model in the second embodiment is different from the pathological condition index derivation model in the first embodiment, and is a model constructed by machine learning based on the results of basic respiratory function tests of multiple subjects, image data obtained by image tests related to the respiratory organs of the multiple subjects, and gas diffusion capacity, which is the result of a detailed respiratory function test of the multiple subjects. Moreover, the pathological condition index derivation model in the second embodiment outputs a value of gas diffusion capacity corresponding to the inputted results of basic respiratory function tests and image data of the respiratory organs.

[0043] The configuration of a diagnosis support system 10 of the second embodiment (mainly the differences from the first embodiment) will be described with reference to Fig. 1. The test information storage unit 20 of the model generating device 12 stores basic data of a plurality of subjects and the results (DLCO%) of a precise respiratory function test in association with each other. The basic data of the second embodiment includes four items of data indicating physical information, 19 items of data obtained as a result of the basic respiratory function test, and image data (hereinafter also referred to as "X-ray images") obtained by an imaging test (X-ray examination in the second embodiment) related to the respiratory system of the subject.

[0044] The pre-processing unit 24 of the model generating device 12 performs pre-processing on the X-ray image in addition to the pre-processing of the first embodiment. The pre-processing unit 24 standardizes a plurality of X-ray images of a plurality of subjects. FIG. 5 shows an example of pre-processing on the X-ray image of a subject. An original image 50 in FIG. 5 is an X-ray image before pre-processing. The original image 50 is 2208×2208 pixels, and each pixel value is a value in the range of 0 to 255. The pre-processing unit 24 converts the original image 50 into a converted image 52 of 221×221 pixels using the resize function of the library OpenCV. The pre-processing unit 24 also divides each pixel value of the converted image 52 by 255 to convert each pixel value into a value in the range of 0 to 1.

[0045] FIG. 6 also shows an example of preprocessing of an X-ray image of a subject. An original image 54 in FIG. 6 is also an X-ray image before preprocessing. The original image 54 is 2688×2208 pixels, and each pixel value is a value in the range of 0 to 255. The preprocessing unit 24 deletes 240 pixel regions at the left and right ends of the original image 54, and converts the remaining region of the original image 54 into a converted image 52 of 221×221 pixels. The preprocessing unit 24 also divides each pixel value of the converted image 52 by 255, thereby converting each pixel value into a value in the range of 0 to 1.

[0046] The original image 56 in Fig. 6 is also an X-ray image before preprocessing. The original image 56 is 2208 x 2688 pixels, and each pixel value is a value in the range from 0 to 255. The preprocessing unit 24 deletes a region of 480 pixels at the bottom end of the original image 54, and converts the remaining region of the original image 54 into a converted image 52 of 221 x 221 pixels. The preprocessing unit 24 also divides each pixel value of the converted image 52 by 255, thereby converting each pixel value into a value in the range from 0 to 1.

[0047] Returning to Fig. 1, the model generating unit 26 of the model generating device 12 generates a pathological index derivation model (neural network) by machine learning based on the preprocessed basic data of each of the multiple subjects and the preprocessed gas diffusivity value of each of the multiple subjects. The preprocessed basic data includes preprocessed X-ray image data.

[0048] 7 shows the network configuration of the pathology index derivation model of the second embodiment. The pathology index derivation model of the second embodiment is a 54-layer neural network including an input layer, a 2D convolution layer, a fully connected layer, a batch normalization layer, a max pooling layer, an activation layer, a concatenation layer, a dropout layer, and a flattening layer. The pathology index derivation model of the second embodiment can also be said to be a multi-input model equipped with multiple input layers (input layer 60, input layer 64).

[0049] The first input layer, input layer 60, receives input data of the 221 x 221 pixel converted image 52. The activation layer 62 outputs 64 one-dimensional data. Between the input layer 60 and the activation layer 62, multiple 2D convolution layers that perform convolution processing are provided, and a flattening layer that converts a two-dimensional array into a one-dimensional array is also provided.

[0050] The second input layer, input layer 64, accepts input of 23 items of basic data (physical information and results of basic respiratory function tests) after preprocessing. In the concatenation layer 66, the output data from the activation layer 62 and the output data from the input layer 64 are concatenated. The final activation layer 68 can also be called the output layer, and outputs one DLCO% index value. The activation function used in the activation layers other than the last one among the multiple activation layers arranged in the pathology index derivation model is "relu", and the activation function used in the final activation layer 68 is "linear". The dropout rate of the dropout layer is 0.5.

[0051] Returning to Fig. 1, the examination information receiving unit 32 of the diagnosis support device 14 receives basic data (physical information, results of a basic respiratory function test, and an X-ray image) of the target subject transmitted from the user device 16. The pre-processing unit 34 of the diagnosis support device 14 executes the pre-processing of the first embodiment as the first processing unit, and in addition, executes pre-processing of the X-ray image of the target subject as the second processing unit. Specifically, the pre-processing unit 34, as the second processing unit, standardizes the data (size and pixel value range) of the X-ray image of the target subject, similar to the pre-processing unit 24 of the model generating device 12.

[0052] The DLCO acquisition unit 36 ​​of the diagnosis support device 14 inputs the basic data (physical information, results of basic respiratory function tests, and X-ray images) standardized by the preprocessing unit 34 to the pathological condition index derivation model (multi-input model in FIG. 7) stored in the model storage unit 30. The DLCO acquisition unit 36 ​​acquires the index value of DLCO% corresponding to the input basic data, which is output from the pathological condition index derivation model.

[0053] The operation related to the generation of the pathology index derivation model of the second embodiment is similar to the operation related to the generation of the pathology index derivation model of the first embodiment. Hereinafter, the operation related to the estimation of the gas diffusivity of a subject using the pathology index derivation model of the second embodiment will be described.

[0054] In response to an operation by a medical professional or the like, the user device 16 transmits basic data of the subject whose gas diffusion capacity is to be estimated to the diagnosis support device 14. The basic data of the subject includes the results of a basic respiratory function test measured by a predetermined test device (e.g., a spirometer), physical information, and an X-ray image showing the subject's respiratory system. The test information receiving unit 32 of the diagnosis support device 14 receives the basic data of the subject transmitted from the user device 16.

[0055] The preprocessing unit 34 of the diagnostic support device 14 standardizes the values ​​of each item of the basic data of the subject, for example, standardizing the size and value range of the X-ray image of the subject. The DLCO acquisition unit 36 ​​of the diagnostic support device 14 inputs the standardized basic data of the subject into a pathological index derivation model (multi-input model) and acquires an index value of DLCO% corresponding to the input value from the pathological index derivation model. The DLCO acquisition unit 36 ​​obtains an estimated value of DLCO% of the subject by exponentiating the index value of DLCO% acquired from the pathological index derivation model with e as the base. The diagnostic support information output unit 38 of the diagnostic support device 14 provides diagnostic support information related to the estimated value of DLCO% of the subject to the user device 16 for display.

[0056] The diagnosis support system 10 of the second embodiment also has the same effect as the diagnosis support system 10 of the first embodiment. Moreover, in the diagnosis support system 10 of the second embodiment, by adding data of an X-ray image of the subject to an input signal to the pathological index derivation model, it becomes easier to improve the estimation accuracy of the gas diffusivity of the subject. Furthermore, although the aspect (size, brightness, contrast ratio, etc.) of the X-ray image may differ depending on the medical institution or technician, by standardizing the X-ray image in the pre-processing, it is possible to suppress a decrease in the estimation accuracy of the gas diffusivity of the subject even when X-ray images of various aspects are input.

[0057] The estimation results using the pathological index derivation models of the first and second embodiments will be described. The inventor used data of 124 cases out of 218 patients from Hospital A as training data (in other words, learning data) for constructing a pathology index derivation model, data of 62 cases as verification data for verifying the pathology index derivation model, and data of 32 cases as test data for the pathology index derivation model. The data of each patient includes basic data of each patient (physical information, results of basic respiratory function test, and X-ray images) and results of detailed respiratory function test (DLCO%).

[0058] FIG. 8(a) shows the estimation result of the pathological index derivation model when the verification data is used, and FIG. 8(b) shows the estimation result of the pathological index derivation model when the test data is used. As shown in FIG. 8(a) and FIG. 8(b), in both the case of using the verification data and the case of using the test data, the pathological index derivation model of the second embodiment (multi-input model of FIG. 7) showed better results in both mean absolute error and correlation coefficient. In addition, it was confirmed that the pathological index derivation model of the first embodiment (value input model of FIG. 4) has a useful accuracy for estimating the result of the precise respiratory function test (DLCO%). In the future, it is expected that the estimation accuracy of the pathological index derivation model of the second embodiment (multi-input model) will be further improved by expanding the number and form of X-ray images used as learning data.

[0059] The present disclosure has been described above based on the first and second embodiments. These embodiments are merely examples, and it will be understood by those skilled in the art that various modifications are possible in the combinations of the components and processing steps described in the embodiments, and that such modifications are also within the scope of the present disclosure. Modifications are shown below.

[0060] A first modified example will be described below. In the first and second embodiments, the data output by the pathological condition index derivation model is DLCO%, but it may be a DLCO value.

[0061] A second modified example will be described. The functions of the diagnosis support device 14 of the first or second embodiment may be incorporated into a dedicated device (e.g., a spirometer) that performs a basic respiratory function test. The respiratory function test device of this modified example may (1) store a pathological index derivation model. Also, (2) accept input of the subject's physical information (or physical information and an X-ray image). Furthermore, (3) perform a basic respiratory function test and estimate the subject's gas diffusion capacity based on the test result and the subject's physical information (or further an X-ray image). Furthermore, (4) output the gas diffusion capacity (DLCO value or DLCO%) together with the result of the basic respiratory function test. According to this modified example, a respiratory function test device (e.g., a spirometer) having the functions of the diagnosis support device 14 of the first or second embodiment can be realized.

[0062] A third modified example will be described. In the first and second embodiments, the pathological index derivation model is realized by a neural network, but as a modified example, a mathematical model or function may be constructed as the pathological index derivation model by other machine learning techniques. For example, the model generation unit 26 of the model generation device 12 may generate a pathological index derivation model of a support vector machine that estimates the value (or value range) of gas diffusivity from the input basic data of the subject. As another example, the model generation unit 26 may execute a multiple regression analysis with the basic data of the subject as explanatory variables and the value (or value range) of gas diffusivity as the objective variable, and generate a regression equation that estimates gas diffusivity based on the input basic data as the pathological index derivation model.

[0063] Any combination of the above-mentioned examples and modifications is also useful as an embodiment of the present invention. A new embodiment resulting from the combination has the effects of each of the combined examples and modifications. It will also be understood by those skilled in the art that the functions to be performed by each component recited in the claims can be realized by each component shown in the examples and modifications alone or in cooperation with each other. [Explanation of symbols]

[0064] 10 diagnosis support system, 12 model generation device, 14 diagnosis support device, 16 user device, 30 model storage unit, 32 examination information receiving unit, 34 preprocessing unit, 36 DLCO acquisition unit, 38 diagnosis support information output unit.

Claims

1. A diagnostic support device capable of accessing a memory unit that stores a model constructed by machine learning based on the results of a first respiratory function test, which is a spirogram, and a second respiratory function test, which is a gas diffusion capacity test, of a plurality of subjects who have undergone the first respiratory function test, and the gas diffusion capacity values ​​that are the results of the second respiratory function test of the plurality of subjects, the model outputting a value of the gas diffusion capacity that corresponds to the input result of the first respiratory function test, A reception unit that receives a result of the first respiratory function test of a certain subject; an acquisition unit that inputs the result of the first respiratory function test into the model and acquires the value of the gas diffusion capacity output from the model; An output unit that outputs information regarding the acquired value of the gas diffusion capacity; Equipped with the model is constructed further based on image data obtained in an imaging test related to the respiratory system of the plurality of subjects, and outputs a value of the gas diffusion capacity corresponding to the input result of the first respiratory function test and the image data obtained in the imaging test; The reception unit further receives image data obtained by an image test related to the respiratory system of the certain subject, The acquisition unit further inputs data of the image into the model and acquires the value of the gas diffusion capacity output from the model. Diagnostic aids.

2. The result of the first respiratory function test includes data of a plurality of items, A first processing unit that standardizes the average value and standard deviation of data for each item in the result of the first respiratory function test, The diagnosis support device according to claim 1 .

3. Further comprising a second processing unit that standardizes the image data. The diagnosis support device according to claim 1 .

4. The model is constructed based on the natural logarithm of the gas diffusion capacity value obtained as a result of the second respiratory function test. The diagnosis support device according to claim 1 .

5. The model is a neural network. The diagnosis support device according to claim 1 .

6. A model constructed by machine learning based on the results of the first respiratory function test of a plurality of subjects who have undergone a first respiratory function test which is a spirogram and a second respiratory function test which is a gas diffusion capacity test, and the gas diffusion capacity values ​​which are the results of the second respiratory function test of the plurality of subjects, wherein a computer that can access a memory unit that stores the model and outputs the gas diffusion capacity values ​​corresponding to the input results of the first respiratory function test, receiving a result of the first respiratory function test of a subject; A function of inputting the result of the first respiratory function test into the model and obtaining the value of the gas diffusion capacity output from the model; A function of outputting information regarding the acquired value of the gas diffusion capacity; Realize this, the model is constructed further based on image data obtained in an imaging test related to the respiratory system of the plurality of subjects, and outputs a value of the gas diffusion capacity corresponding to the input result of the first respiratory function test and the image data obtained in the imaging test; The receiving function further receives image data obtained by an imaging test related to the respiratory system of the subject; The function of obtaining further inputs data of the image into the model to obtain the value of the gas diffusivity output from the model. Computer program.

7. A model constructed by machine learning based on the results of the first respiratory function test of a plurality of subjects who have undergone a first respiratory function test which is a spirogram and a second respiratory function test which is a gas diffusion capacity test, and the gas diffusion capacity values ​​which are the results of the second respiratory function test of the plurality of subjects, said model being constructed by machine learning based on the results of the first respiratory function test of the plurality of subjects and the gas diffusion capacity values ​​which are the results of the second respiratory function test of the plurality of subjects, said model being configured to output the gas diffusion capacity values ​​which correspond to the input results of the first respiratory function test ... receiving a result of the first respiratory function test of a subject; inputting the results of the first respiratory function test into the model to obtain the gas diffusion capacity value output from the model; Outputting information regarding the acquired gas diffusion capacity value; Run the model is constructed further based on image data obtained in an imaging test related to the respiratory system of the plurality of subjects, and outputs a value of the gas diffusion capacity corresponding to the input result of the first respiratory function test and the image data obtained in the imaging test; The receiving step further includes receiving image data obtained in an imaging test related to the respiratory system of the subject; The obtaining step further inputs data of the image into the model to obtain the value of the gas diffusivity output from the model. Gas diffusivity estimation method.

Images