Medical Support System

Abstract

To flexibly support diagnosis and treatment of a patient.SOLUTION: A diagnosis and treatment support system comprises a calculation section, a generation section, and a derivation section. The calculation section calculates a conversion function that enables statistical derivation of a possible examination value for another examination item from an examination value or values for one or a plurality of result items on the basis of correlations between examination values for examination items included in diagnosis and treatment information on a plurality of patients. The generation section generates support information for supporting diagnosis and treatment of the target patient on the basis of an inference result obtained by inputting an examination value for an examination item included in diagnosis and treatment information on a target patient into a learned model that infers information related to a state of a patient from an examination value for a predetermined examination item. The derivation section derives an examination value for a lacking examination item from an examination value for an examination item included in the diagnosis and treatment information on the target patient on the basis of the conversion function in a case where examination items included in the diagnosis and treatment information on the target patient do not satisfy a requirement for the predetermined examination item. The generation section also causes the learned model to infer information related to a state of the target patient using the examination value, which has been derived by the derivation section, for the lacking examination item.SELECTED DRAWING: Figure 3

Description

[Technical field]

[0001] The embodiments disclosed in this specification relate to a medical support system. [Background technology]

[0002] Traditionally, in medical facilities, examinations and treatments are performed by doctors (specialists) in the medical department that corresponds to the patient's illness. In recent years, support technologies such as Clinical Decision Support (CDS) Systems have been proposed that use models obtained by analyzing past patient medical information (examination results, etc.) using machine learning and other techniques to infer the disease state and risk of progression of the target patient.

[0003] Meanwhile, while medical specialists can appropriately detect diseases in their own fields, it is difficult for them to detect diseases in fields outside their specialty at an early stage. Therefore, it is conceivable that early detection of diseases in fields outside their specialty can be achieved by using the above-mentioned support technology, but since different medical departments require different types of tests, it is not easy to use support technology for other medical departments. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Special Publication No. 2016-502650 Summary of the Invention [Problem to be solved by the invention]

[0005] One of the problems that the embodiments disclosed in this specification are intended to solve is to provide flexible medical support to patients. However, the problems solved by the embodiments disclosed in this specification are not limited to the above problems. Problems corresponding to the effects of each configuration shown in the embodiments described below can also be considered as other problems solved by the embodiments disclosed in this specification. [Means for solving the problem]

[0006] The medical support system according to the embodiment includes a calculation unit, a storage unit, a generation unit, and a derivation unit. The calculation unit calculates a conversion function capable of statistically deriving a value that the test value of one or more result items can take from the test value of another test item based on a correlation between the test values ​​of each test item included in the medical information of a plurality of patients. The storage unit stores the conversion function calculated by the calculation unit in association with a condition related to the test item used in calculating the conversion function. The generation unit generates support information for supporting medical care of the target patient based on an inference result obtained by inputting the test value of the test item included in the medical information of the target patient into a trained model that infers information about a patient's condition from the test value of a predefined test item. When the test item included in the medical information of the target patient does not satisfy the requirements of the predefined test item, the derivation unit derives the test value of the missing test item using a conversion function stored in the storage unit that is capable of deriving the test value of the missing test item from the test item included in the medical information of the target patient. In addition, the storage unit stores, as conditions related to the test items, index values ​​based on any or all of the total number of the medical information used in calculating the conversion function, the number of days on which the medical information was acquired, the number of the test items included in the medical information, and the number of the plurality of patients; When a plurality of conversion functions capable of deriving test values ​​for missing test items are stored in the memory unit, the derivation unit selects a conversion function to be used based on the condition associated with the conversion function, and the generation unit causes the trained model to infer information regarding the condition of the target patient using the test values ​​for the missing test items derived by the derivation unit. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of a medical support system according to the first embodiment. [Diagram 2] FIG. 2 is a diagram illustrating an example of the configuration of the medical support device according to the first embodiment. [Diagram 3] FIG. 3 is a diagram illustrating an example of components included in a processing circuit of the medical support device according to the first embodiment. [Figure 4]FIG. 4 is a diagram illustrating an example of sample data acquired by the first acquisition function according to the first embodiment. [Diagram 5] FIG. 5 is a diagram illustrating the state of a probability distribution derived by a conversion function according to the first embodiment. [Figure 6] FIG. 6 is a diagram illustrating an example of sample data acquired by the first acquisition function according to the first embodiment. [Figure 7] FIG. 7 is a diagram showing an example of a data configuration of the conversion function table according to the first embodiment. [Figure 8] FIG. 8 is a diagram showing another example of the data configuration of the conversion function table according to the first embodiment. [Figure 9] FIG. 9 is a diagram for explaining an example of the operation of the conversion function according to the first embodiment. [Figure 10] FIG. 10 is a diagram for explaining the range of the inspection value derived by the conversion function of the first embodiment. [Figure 11] FIG. 11 is a diagram illustrating an example of a medical information table according to the first embodiment. [Figure 12] FIG. 12 is a diagram illustrating an example of a support information table according to the first embodiment. [Figure 13] FIG. 13 is a diagram showing an example of a data configuration of the display configuration information according to the first embodiment. [Figure 14] FIG. 14 is a diagram illustrating an example of a screen displayed by the display function of the first embodiment. [Figure 15] FIG. 15 is a flowchart illustrating an example of a conversion function calculation process executed by the medical support apparatus according to the first embodiment. [Figure 16] FIG. 16 is a flowchart illustrating an example of medical support processing executed by the medical support device according to the first embodiment. [Figure 17] FIG. 17 is a diagram illustrating an example of components included in a processing circuit of a medical support device according to the second embodiment. [Figure 18] FIG. 18 is a flowchart illustrating an example of a learning process executed by the medical support apparatus according to the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] Hereinafter, an embodiment of a medical support device will be described with reference to the drawings.

[0009] [First embodiment] Fig. 1 is a diagram showing an example of the configuration of a medical support system according to the first embodiment. As shown in Fig. 1, the medical support system 1 includes a medical DB 10, a medical support device 20, and an inference device 30. The medical DB 10, the medical support device 20, and the inference device 30 are installed in a medical facility such as a hospital, and are communicatively connected to each other via a network N1 such as an in-hospital network.

[0010] The medical DB 10 and the inference device 30 are provided, for example, for each medical department in a medical facility. Fig. 1 shows an example in which the medical DB 10 (10a, 10b) and the inference device 30 (30a, 30b) are provided for each of the first medical department A and the second medical department B. For example, the first medical department A is a respiratory department, and the second medical department B is a cardiology department.

[0011] The medical DB 10 includes a database that stores medical information of multiple patients. The medical DB 10 is an integrated management device that acquires various types of medical information from various medical information systems, such as an electronic medical record system, a PACS (Picture Archiving and Communication System), a radiology department system, and a specimen testing system, and manages the acquired medical information. Although FIG. 1 shows an example in which a dedicated medical DB 10 is provided for each medical department, the present invention is not limited to this, and the medical DB 10, which is configured by a single or multiple storage devices, may be shared by all medical departments.

[0012] The medical DB 10 stores medical information on each patient in association with a patient ID that identifies the patient. The medical information includes, for example, examination information, medical images, document information, etc. The examination information includes examination values ​​obtained by various examinations. Specifically, the examination information is composed of a group of information such as the date the examination was performed, examination items indicating the type of examination, and examination values.

[0013] The test information includes, for example, test items common to each medical department, as well as test items specific to the medical department. Common test items include basic test items that indicate the patient's physical condition, such as blood pressure, heart rate, and body temperature. Specific test items differ depending on the medical department, and in the case of cardiology, for example, test items such as NT-proBNP, cardiac troponin I, and CTR (cardio-thoracic ratio) are included.

[0014] The medical images are images of patients acquired by various medical modalities, such as X-ray CT images acquired by an X-ray CT device, MRI images acquired by an MRI device, X-ray images acquired by an X-ray device, mammography images acquired by a mammography device, etc. The medical images are stored in association with additional information, such as the type of medical modality and the imaging site. The medical images are also stored in association with information indicating the date on which the medical images were acquired.

[0015] The document information is document data that records information such as test values ​​related to the physical condition of a patient, such as electronic medical records and report data. The document information is stored in association with information indicating the date on which the document information was recorded. In the following, the physical condition of a patient contained in medical images and document information is also referred to as test values.

[0016] The inference device 30 is a device related to Clinical Decision Support (CDS) Systems, Computer-Aided Diagnosis (CAD) systems, and the like. The inference device 30 is realized by computer equipment such as a personal computer (PC), a workstation, or a server. The inference device 30 may also be realized by a plurality of computer equipment (clouds) using cloud computing.

[0017] The inference device 30 stores a trained model M1 (see FIG. 3) created by a machine learning algorithm such as logistic regression, neural network, and deep learning. The trained model M1 is trained in advance to infer information on the patient's condition, such as a disease name and a risk value of transformation that quantitatively indicates the possibility of morbidity, from the test difference when the test value of a specified test item is input. Here, by including a test item specific to a medical department in the test items of the input data used (input) during learning, a trained model M1 specialized for the field handled by each medical department can be created. In this embodiment, it is assumed that the trained model M1 specialized for each medical department is stored in the inference device 30. Specifically, the trained model M1 specialized for a first medical department is stored in the inference device 30a, and the trained model M1 specialized for a second medical department is stored in the inference device 30b.

[0018] When the inference device 30 receives input of test values ​​for test items serving as input data from, for example, a terminal device in a medical department or the medical support device 20, the inference device 30 inputs the received test values ​​for the test items into the trained model M1. The inference device 30 then outputs the inference results, such as the disease name inferred by the trained model M1, to the device that is the input source. Note that the inference device 30 may be configured to notify the device that has accessed the device of the test items of the input data, i.e., the test items used to derive the inference results, to the device that has accessed the device.

[0019] The medical support device 20 executes a process for calculating a conversion function (described later) using the medical information of multiple patients stored in the medical DB 10. In addition, the medical support device 20 executes a process for supporting the diagnosis and medical treatment (hereinafter, also collectively referred to as medical treatment) of a target patient (hereinafter, also referred to as a target patient) using the trained model M1 stored in the inference device 30.

[0020] The medical support device 20 is realized by, for example, computer devices such as a PC, a workstation, a server, etc. Furthermore, the medical support device 20 may be realized by a plurality of computer devices (cloud) using cloud computing.

[0021] Next, the configuration of the above-mentioned medical support device 20 will be described. Fig. 2 is a diagram showing an example of the configuration of the medical support device 20. As shown in Fig. 2, the medical support device 20 has an input interface 21, a display 22, a communication interface 23, a storage circuitry 24, and a processing circuitry 25. The input interface 21, the display 22, the communication interface 23, the storage circuitry 24, and the processing circuitry 25 are connected to each other.

[0022] The input interface 21 accepts various input operations from an operator, converts the accepted input operations into electrical signals, and outputs the electrical signals to the processing circuit 25. For example, the input interface 21 can be realized by a mouse, a keyboard, a trackball, a switch, a button, a joystick, a touchpad that performs input operations by touching the operation surface, a touchscreen in which a display screen and a touchpad are integrated, a non-contact input circuit using an optical sensor, a voice input circuit, or the like.

[0023] The input interface 21 may be configured with a tablet terminal or the like capable of wireless communication with the main body of the medical support device 20. The input interface 21 is not limited to an interface having physical operation parts such as a mouse and a keyboard. For example, an example of the input interface 21 includes an electrical signal processing circuit that receives an electrical signal corresponding to an input operation from an external input device provided separately from the medical support device 20 and outputs the electrical signal to the processing circuit 25.

[0024] The display 22 displays various information. For example, under the control of the processing circuit 25, the display 22 displays the processing results by the processing circuit 25. The display 22 also displays a GUI (Graphical User Interface) for receiving various instructions and various settings from an operator via the input interface 21. For example, the display 22 is a liquid crystal display or a CRT (Cathode Ray Tube) display. The display 22 may be a desktop type, or may be configured as a tablet terminal or the like capable of wireless communication with the medical support device 20 main body.

[0025] The communication interface 23 is an interface for communicating with an external device connected to the network N1. For example, the processing circuit 25 can transmit and receive various data between the medical DB 10, the inference device 30, and the like via the communication interface 23.

[0026] The memory circuitry 24 is realized by, for example, a semiconductor memory element such as a random access memory (RAM) or a flash memory, a hard disk, an optical disk, etc. For example, the memory circuitry 24 stores a program for enabling the circuits included in the medical support device 20 to realize their functions. The memory circuitry 24 also stores various data and information such as a conversion function table T1 (see FIG. 3) for holding a conversion function described later.

[0027] The processing circuitry 25 controls the overall processing of the medical support device 20. Moreover, the processing circuitry 25 according to this embodiment has a first acquisition function 251, a calculation function 252, a second acquisition function 253, a conversion function 254, a generation function 255, a display configuration function 256, and a display function 257, as shown in FIG.

[0028] 3 is a diagram showing an example of components of the processing circuitry 25 of the medical support device 20. For example, the processing functions executed by the first acquisition function 251, the calculation function 252, the second acquisition function 253, the conversion function 254, the generation function 255, the display configuration function 256, and the display function 257, which are components of the processing circuitry 25 shown in FIG. 3, are stored in the storage circuitry 24 in the form of a program executable by a computer. The processing circuitry 25 is a processor that reads out each program from the storage circuitry 24 and executes it to realize a function corresponding to each program. In other words, the processing circuitry 25 in a state in which each program has been read out has each function shown in FIG. 3.

[0029] 3, the first acquisition function 251, the calculation function 252, the second acquisition function 253, the conversion function 254, the generation function 255, the display configuration function 256, and the display function 257 are realized by a single processing circuit 25, but the processing circuit 25 may be configured by combining multiple independent processors, and each processor may execute a program to realize the functions. Also, each processing function of the processing circuit 25 may be realized by being appropriately distributed or integrated into a single or multiple processing circuits.

[0030] The processing circuitry 25 may also realize the functions by using a processor of an external device connected via the network N1. For example, the processing circuitry 25 reads out and executes a program corresponding to each function from the storage circuitry 24, and realizes each function shown in FIG. 3 by using a server group (e.g., a cloud) connected to the medical support device 20 via the network N1 as a computational resource.

[0031] In addition, in Fig. 3, medical information D1, a conversion function table T1, and a learned model M1 are illustrated as elements other than the functions of the processing circuit 25. Here, the medical information D1 is medical information stored in the medical DB 10 (medical DB 10a, medical DB 10b). Furthermore, the conversion function table T1 is a memory unit for storing a conversion function described later that is held in the memory circuit 24 or the like. Furthermore, the learned model M1 is a learned model stored in the inference device 30 (inference device 30a, inference device 30b).

[0032] The first acquisition function 251 acquires the medical information D1 of a plurality of patients stored in the medical DB 10 as sample data for calculating a conversion function. The first acquisition function 251 may acquire the medical information D1 stored in both the medical DB 10a and the medical DB 10b, or may acquire the medical information D1 stored in either one of them.

[0033] The first acquisition function 251 may also acquire medical information D1 that meets a predetermined condition. For example, the first acquisition function 251 may acquire medical information D1 related to a specific patient ID. For example, the first acquisition function 251 may also acquire medical information D1 of patients with the same or similar disease name.

[0034] It should be noted that the conditions for acquiring the medical information D1 can be set arbitrarily. The first acquiring function 251 may be configured to switch the conditions for acquiring the medical information D1 in cooperation with the calculating function 252.

[0035] The calculation function 252 is an example of a calculation unit. The calculation function 252 calculates a conversion function capable of statistically deriving a value that the test value of one or more test items included in the medical information D1 can take, based on the medical information D1 (sample data) of multiple patients acquired by the first acquisition function 251. The conversion function is, for example, a learned model learned by machine learning, a model formula of simple regression analysis or multiple regression analysis, or the like.

[0036] Specifically, the calculation function 252 calculates a conversion function capable of deriving a probability distribution of values ​​that the test value of an arbitrary test item can take from the test values ​​of other test items by analyzing the relationship between the test values ​​of each test item included in the medical information D1 using a method such as machine learning. Here, the calculation method of the conversion function is not particularly limited, but it is preferable to calculate the conversion function using a method (algorithm) according to the type of test item and the characteristics of the test value. For example, the calculation function 252 may calculate the conversion function using the calculation methods (first to third calculation methods) described below.

[0037] (First calculation method) In the first calculation method, the calculation function 252 collects, for each patient, medical information D1 (examination information) on the same day with the same date attached, and sets the collected medical information D1 as a data set for calculating a conversion function.

[0038] Here, for example, it is assumed that the sample data acquired by the first acquisition function 251 is the content shown in FIG. 4. FIG. 4 is a diagram showing an example of the sample data acquired by the first acquisition function 251. In this case, the calculation function 252 sets each of the test information acquired on each of the dates "01.01", "01.02", "01.10", "01.11", "01.13", "02.01", and "02.03" as a data set. Note that the calculation function 252 sets each of the test information acquired on the same date for each patient ID as independent data so that test information for multiple patients on the same date is not mixed.

[0039] Next, the calculation function 252 learns the relationship between the test values ​​of each test item included in each piece of test information set in the data set by using a machine learning method such as a linear regression model or a neural network. Then, the calculation function 252 calculates a conversion function that can probabilistically derive the test value of another test item as a conversion destination from the test value of the test item as a conversion source, based on the correlation between the test values ​​of each test item included in the test information.

[0040] For example, as shown in the following formula (1), if the probability distribution that when a test value of a test item x2 is input, a test value of another test item x1 is output is P(x1|x2), the calculation function 252 calculates a conversion function f(x2) from which this P(x1|x2) can be derived. P(x1|x2)=f(x2) …(1)

[0041] As an example, of the three test items constituting the test information, "blood pressure," "NT-proBNP," and "cardiac troponin I," "cardiac troponin I" is set as x1 and "NT-proBNP" as x2. In this case, the calculation function 252 calculates a conversion function f(NT-proBNP) capable of deriving a probability distribution P(cardiac troponin I|NT-proBNP) of values ​​that the cardiac troponin I test value can take, for the test value of NT-proBNP.

[0042] Fig. 5 is a diagram showing a schematic state of the probability distribution P(x1|x2) derived by the conversion function (x2). In Fig. 5, the horizontal axis represents the test value of the test item x2 to be input, and the vertical axis represents the test value of the test item x1 to be output.

[0043] The shading in the background of the x1-x2 space represents the probability distribution of the test value of test item x1, with higher shading indicating a higher probability of occurrence. The dashed line L1 shown in the x1-x2 space is a regression equation that approximates the distribution of the value of test item x2 that maximizes the probability for each value of test item x1.

[0044] 5, by using the conversion function f(x2), it is possible to identify the probability distribution P(x1|x2) of possible values ​​of the test value of test item x1 from the test value of test item x2. In other words, the conversion function f(x2) can output possible values ​​of the test value of test item x1 and their probability (hereinafter also referred to as reliability) from the input test value of test item x2.

[0045] Furthermore, the calculation function 252 comprehensively combines a plurality of test items that constitute the test information, and calculates a conversion function for each combination of the test items.

[0046] For example, in the example of Fig. 4, the calculation function 252 calculates a conversion function capable of deriving a probability distribution of a test value of one test item from the test values ​​of any one or two of the three test items constituting the test information, "blood pressure," "NT-proBNP," and "cardiac troponin I." In this case, the calculation function 252 calculates nine conversion functions. Note that the conversion function for deriving a probability distribution of a test value of the remaining test item x1 from the test values ​​of the two test items x2 and x3 can be expressed, for example, by the following formula (2) similar to the above formula (1). P(x1|x2, x3)=f(x2, x3) …(2)

[0047] The calculation function 252 may determine a combination of test items for calculating the conversion function based on the type of input data input to the trained model M1. For example, when the input data of the trained model M1 are the test items "blood pressure" and "NT-proBNP", the calculation function 252 may calculate a conversion function capable of deriving a probability distribution between the test items "blood pressure" and "NT-proBNP".

[0048] Then, the calculation function 252 stores the calculated conversion function in the conversion function table T1. In addition, the calculation function 252 stores the accuracy obtained when calculating the conversion function, the condition of the examination information used in calculating the conversion function, and the like, in association with the conversion function.

[0049] Here, the precision is the random error between the actual test value of the test item x1 and the test value of the test item x1 derived (estimated) by the transformation function. The precision can be calculated based on a probability model such as a normal distribution.

[0050] Furthermore, the conditions of the test information include the data characteristics of the test information used to calculate the conversion function, the number of test information (number of records) included in the data set, the number of days counted as the same day, the number of test items (number of test items), the number of patients, etc. The data characteristics of the test information include the patient ID of the test information used to calculate the conversion function and the disease name, etc. The disease name may be determined, for example, from written information associated with the patient ID, or may be determined from other information associated with the patient ID (for example, information recording the doctor's diagnosis, etc.).

[0051] In this embodiment, the above-mentioned condition of the examination information is expressed using the index value "frequency" calculated by the following formula (3). Frequency = total number of records ÷ (number of days × number of tests × number of patients) … (3)

[0052] (Second calculation method) In the second calculation method, the calculation function 252 collects chronologically continuous medical information D1 for each patient, and sets the collected time-series data of the medical information D1 as a data set for calculating a conversion function.

[0053] For example, if the sample data acquired by the first acquisition function 251 is as shown in FIG. 4, the calculation function 252 treats the test information group of patient ID: 0001 acquired on dates "01.01" and "01.02" as one time series data. Similarly, the calculation function 252 treats the test information group of patient ID: 0002 acquired on dates "02.01" and "02.03" as one time series data. Although there are three days' worth of test information for patient ID: 0003, in order to make the conditions the same as those for patient IDs: 0001 and 0002, it is preferable to treat each of the test information groups of "01.10" and "01.11", "01.11" and "01.13", and "01.10" and "01.13" as one time series data.

[0054] Next, the calculation function 252 learns the relationship between the test values ​​of each test item included in the test information by using a machine learning method such as a recurrent neural network (RNN) capable of handling time-series data. Then, the calculation function 252 calculates a conversion function capable of probabilistically deriving the test value of another test item as a conversion destination from the test value of the test item as a conversion source, based on the correlation between the test values ​​of each test item included in the test information.

[0055] In this case, the conversion function can be expressed by, for example, the following equation (4). P(x1 t |x2 t , x1 t-1 , x2 t-1 )=f(x2 t , x1 t-1 , x2 t-1 ) …(4)

[0056] Conversion function f(x2 t , x1 t-1 , x2 t-1 ) is the test value x2 at time t for test item x2 t , the test value x2 at time t-1 t-1 And the test value x1 at the time t-1 of another test item x1 t-1 From this, the test value x1 of another test item x1 at time t t The probability distribution of possible values ​​of P(x1 t |x2 t , x1 t-1 , x2 t-1 ) is derived.

[0057] Furthermore, similarly to the first method, the calculation function 252 comprehensively combines a plurality of examination items constituting the examination information, and calculates a conversion function for each combination of the examination items.

[0058] The calculation function 252 may determine a combination of test items for calculating the conversion function based on the type of input data to be input to the trained model M1. In the above example, two days' worth of test information is treated as one piece of time-series data, but the present invention is not limited to this. Three or more days' worth of test information may be treated as one piece of time-series data to calculate the conversion function.

[0059] Then, the calculation function 252 stores the calculated conversion function in the conversion function table T1. Similarly to the first method, the calculation function 252 stores the accuracy obtained when calculating the conversion function and the condition of the test information used in calculating the conversion function in association with the conversion function. In this embodiment, the above-mentioned condition of the test information is represented by the index value "frequency" calculated by the following formula (3).

[0060] (Third calculation method) Furthermore, the calculation function 252 also calculates a conversion function capable of deriving a probability distribution of a test value of a specified test item from image information, which is unstructured data. Here, "unstructured data" means data in which the test value cannot be directly handled. For example, the test value of each test item constituting the test information can be defined as "structured data" because the test value can be directly read out. On the other hand, image information and written information are "unstructured data" because processing is required to extract the test value. Below, a method of calculating a conversion function from unstructured data will be described as a third calculation method.

[0061] In the third calculation method, the calculation function 252 sets each pair of image information and examination information on the same day that are dated the same as a data set for calculating a conversion function.

[0062] For example, it is assumed that the sample data acquired by the first acquisition function 251 has the contents shown in Fig. 6. Here, Fig. 6 is a diagram showing an example of the sample data acquired by the first acquisition function 251.

[0063] In this case, the calculation function 252 sets each pair of the image information and the CTR test value acquired on each of the dates "01.01" and "02.03" as a data set for calculating the conversion function. Note that the calculation function 252 sets each pair of the image information and the CTR test value acquired on each of the dates "01.01" and "02.03" as independent data for each patient ID so that test information for the same day for multiple patients is not mixed.

[0064] Next, the calculation function 252 learns the relationship between the image information as a processing unit and the test value of the CTR by using a machine learning method such as deep learning that can handle the relationship.The calculation function 252 then calculates a conversion function that can probabilistically derive the test value of the CTR as a conversion destination from the image information as a conversion source based on the correlation between the image information and the test value of the CTR.By using such a conversion function, it is possible to derive the probability distribution of values ​​that the test value of the CTR can take from the image information.

[0065] In the above example, the examination item derived from the image information is the CTR, but it is not limited to this as long as it is an examination item included in the examination information. However, it is preferable that the examination item to be derived is an examination item related to the type of medical modality acquired in the medical image, the imaging site, etc., so as not to be an examination item unrelated to the medical image.

[0066] The conversion function calculated by the third calculation method can be expressed by, for example, the following formula (5). P(x1|x img )=f(x img ) …(5) Here, the transformation function f(x img ) is image information x img From the test value of x1, such as CTR, included in the test information, the probability distribution P(x1|x img ) is derived.

[0067] In addition, image information x img The inspection value may be the image information itself, or may be the result of preprocessing the image information. In the latter case, for example, image information x imgThe inspection value may be a feature value extracted from image information by a known technique.

[0068] Then, the calculation function 252 stores the generated conversion function in the conversion function table T1. In addition, the calculation function 252 stores the accuracy obtained when calculating the conversion function, the conditions of the image information used to calculate the conversion function, and the like, in association with the conversion function.

[0069] The calculation function 252 can also calculate a conversion function for a pair of written information and examination information on the same day with the same date, using the same method as above. However, in this case, learning is performed using a natural language processing or machine learning method that can handle the correlation (or co-occurrence relationship) between the wording of the examination items and examination values, etc., that indicate the physical condition of the patient, contained in the written information, and the examination values ​​of the examination information. In this case, the wording and terms to be extracted from the written information may be preset.

[0070] Although three calculation methods have been described above, these calculation methods can be used in combination. For example, by combining the first calculation method with the third calculation method, the test value of the test item x2 and the image information x img A conversion function capable of deriving a probability distribution of possible values ​​of the test value of the test item x1 may be calculated from the above. In addition, for example, the above-mentioned first calculation method and second calculation method may be used in combination by extracting the wording of the test value, etc. from the written information and using the extracted information as the test values ​​of the test items x2, x3, etc.

[0071] Fig. 7 is a diagram showing an example of the data configuration of the conversion function table T1. As shown in Fig. 7, the conversion function table T1 has data items such as "ID", "output", "input", "conversion function", "precision", and "frequency".

[0072] Here, the "ID" column stores an identification number for identifying each of the conversion functions registered in the conversion function table T1. The "Output" column stores the converted inspection item output by the conversion function. The "Input" column stores the inspection item and type of unstructured data (image information, written information) input to the conversion function. The "Conversion function" column stores the conversion function calculated by the calculation function 252 and the address of the storage destination where the conversion function is stored. The "Accuracy" column stores the accuracy value obtained when the conversion function is calculated. The "Frequency" column stores the frequency value calculated by the above formula (3).

[0073] 8 is a diagram showing another example of the data configuration of the conversion function table T1. The conversion function table T1 in FIG. 8 is obtained by adding a data item of "sample data" to the data configuration of the conversion function table T1 described in FIG. 7. The "data characteristics" column stores conditions related to the medical information D1 (sample data) used when calculating the conversion function. For example, when the conversion function is calculated by limiting the medical information D1 related to a specific patient or a specific disease name, information indicating the limiting conditions is stored in the "sample data" column.

[0074] In this way, in the medical support device 20, the first acquisition function 251 and the calculation function 252 calculate a conversion function that can derive a probability distribution of values ​​that can be taken by test values ​​of any type from test values ​​of other types, based on correlations of various test values ​​included in the medical information D1 of multiple patients. By using such a conversion function, for example, when test values ​​of some test items are insufficient, it is possible to derive values ​​that can be taken by test values ​​of the insufficient test items from the test values ​​of the other test items.

[0075] Incidentally, when the medical information D1 is obtained from a medical DB 10a provided in a first medical department A, this medical information D1 can be applied (input) to a learned model M1 stored in an inference device 30a of the same medical department, thereby obtaining inference results such as disease names that are in line with the medical scope of the same medical department.

[0076] In addition, by applying the medical information D1 in the medical DB 10a to the learned model M1 stored in the inference device 30b of the second medical department B, it is possible to obtain disease names and risks that are in line with the medical scope of the second medical department B, making it possible to treat the patient's condition from multiple angles.

[0077] However, when the medical information D1 obtained from the medical DB 10b is applied to the learned model M1 stored in the inference device 30b of the second medical department B, it is possible that an inference result cannot be derived because the items of input data used to estimate the disease name are different.

[0078] Therefore, in the case where some of the items of input data required for inference of the trained model M1 are missing, the medical support device 20 executes a process to supplement the missing test values ​​of the test items from the test values ​​of the existing test items included in the medical information D1 by using the above-mentioned conversion function. Then, the medical support device 20 executes a process to generate support information for supporting medical treatment of the target patient based on an inference result obtained by inputting the existing test values ​​included in the medical information D1 and the test values ​​after conversion derived using the conversion function into the trained model M1, and to provide the support information. The functions of the medical support device 20 related to these processes will be described below.

[0079] The second acquisition function 253 acquires the medical information D1 of the target patient from the medical DB 10. For example, the second acquisition function 253 acquires the medical information D1 of the target patient from the medical DB 10a of the first medical department A.

[0080] The target patient is specified by inputting the patient ID of the target patient via, for example, the input interface 21 or the communication interface 23. The medical information D1 of the target patient is not limited to one, and multiple pieces of medical information D1 (for multiple days) may be acquired. Furthermore, the target patient is not limited to one, and multiple people may be specified. The medical information D1 of the target patient acquired by the second acquisition function 253 is output to the conversion function 254 and the generation function 255.

[0081] The conversion function 254 is an example of a derivation unit. The conversion function 254 cooperates with the generation function 255 to execute a conversion process for deriving (converting) the test values ​​of other test items from the test values ​​of existing test items included in the medical information D1 of the target patient.

[0082] Specifically, when the conversion function 254 is notified of the missing test items from the generation function 255 described later, the conversion function 254 derives test values ​​for the missing test items from the test values ​​of the existing test items included in the medical information D1 based on the conversion functions stored in the conversion function table T1.

[0083] The following describes an example of the operation of the conversion function 254. Fig. 9 is a diagram for explaining an example of the operation of the conversion function 254. Note that Fig. 9 shows some of the conversion functions stored in the conversion function table T1.

[0084] For example, if the missing test item is x1, the conversion function 254 searches the conversion function table T1 for a conversion function in which the test item x1 is registered in the "output" column and the existing test items included in the medical information D1 are registered in the "input" column. Here, for example, if x2 and x3 are included in the medical information D1 as existing test items, the conversion function 254 identifies conversion functions with IDs: 1 to 3 from the conversion function table T1 in FIG. 9.

[0085] As described above, when there are multiple conversion functions that meet the search criteria, the conversion function 254 selects one conversion function based on the accuracy and frequency stored in association with those conversion functions. For example, the conversion function 254 may select the conversion function with the highest accuracy value, or the conversion function with the highest frequency value.

[0086] Furthermore, the conversion function 254 may be configured to calculate weighted accuracy from accuracy and frequency by using, for example, the following formula (5), and select one conversion function based on the calculation result. Weighted accuracy = accuracy × inverse frequency … (5)

[0087] The smaller the weighted accuracy calculated by the above formula (5), the higher the accuracy. Figure 9 shows the weighted accuracy calculated by the above formula (5). In this case, the conversion function 254 selects the conversion function with the smallest weighted accuracy value, that is, the conversion coefficient with ID: 3, and can derive the test values ​​of the missing test items using a conversion function with higher accuracy.

[0088] The weighted accuracy may be calculated based on the following formula (6). In this case, since the calculated weighted accuracy has a higher accuracy as the value is larger, it is preferable that the conversion function 254 selects the conversion function with the largest weighted accuracy value. Weighted accuracy = accuracy × frequency … (6)

[0089] Next, the conversion function 254 uses the selected conversion function to derive test values ​​for the missing test items. Specifically, the conversion function 254 reads out search values ​​for the test items specified in the "input" field of the selected conversion function from the existing test items included in the medical information D1 of the target patient, and inputs them to the conversion function.

[0090] When a test value of a test item to be converted is input, the conversion function can output (derive) a possible value of the test value of the test item to be converted as a random variable.

[0091] FIG. 10 is a diagram for explaining the range of a random variable derived by a conversion function. Here, FIG. 10 is a diagram showing a schematic diagram of a state of a probability distribution derived by a conversion function. The horizontal axis represents the test item x i The vertical axis is the test value of the test item x to be output. o In addition, in Figure 10, the test value of the test item x i is the feature value extracted from the image information, and the inspection item x o We will proceed with the explanation assuming that this is the CTR.

[0092] For example, test item x i If the test value of falls in position P1, the test item x derived by the transformation function oThe test value of x can take any value within the range of the probability distribution at the position P1. o The test values ​​are then narrowed down to one or more within the range of the probability distribution.

[0093] For example, when narrowing down the test values ​​derived by the conversion function to one, the conversion function 254 selects the test item x with the highest probability. o 10, the conversion function 254 obtains the test value of the test item x on the dashed line L1 corresponding to the position P1 as the derivation result (conversion result). o The inspection value “53” is obtained as the derived result.

[0094] In addition, for example, when outputting a plurality of test values ​​derived by a conversion function, the conversion function 254 converts the test values ​​into a plurality of test values ​​each having a probability of being greater than or equal to a threshold value Th p Set the range (the range of the arrow in the figure) where the test item x is greater than or equal to this range. o The maximum and minimum values ​​of the test values ​​and the test value with the highest probability are obtained as the derivation results. o The maximum inspection value is "63", the minimum is "43", and the inspection value with the highest probability is "53." Note that in this embodiment, the conversion function 254 will be described as deriving a plurality of inspection values ​​from a conversion function.

[0095] Then, the conversion function 254 outputs each of the test values ​​derived for the missing test items to the generation function 255. Specifically, the conversion function 254 associates the test values ​​derived for the missing test items with characteristic information indicating the characteristics of the test values ​​on the probability distribution, and outputs them to the generation function 255. Here, the characteristic information indicates, for example, the above-mentioned maximum value, minimum value, or test value with the highest probability.

[0096] The conversion function 254 may narrow down the conversion functions to be searched for depending on the number of days included in the medical information D1 of the target patient. For example, when the medical information D1 of the target patient is composed of a single date, the conversion function 254 may narrow down the conversion functions to those calculated by the first calculation method or the third calculation method described above. When the medical information D1 of the target patient is composed of two or more dates, the conversion function 254 may narrow down the conversion functions to those calculated by the second calculation method described above.

[0097] 3, the generating function 255 is an example of a generating unit. The generating function 255 generates support information for supporting medical treatment of a target patient based on an inference result obtained by inputting test values ​​of test items included in the medical treatment information D1 of a target patient to the trained model M1 to be applied.

[0098] Here, the trained model M1 to be applied may be specified via the input interface 21 or the communication interface 23, or may be set in advance. In this embodiment, the trained model M1 to be applied is described as being the trained model M1 stored in the inference device 30b.

[0099] Specifically, the generating function 255 compares the test items included in the medical information D1 input from the second acquiring function 253 with the requirements (test items) of the input data to the trained model M1 to be applied. If the comparison result shows that the test items included in the medical information D1 of the target patient do not satisfy the requirements of the trained model M1, that is, if there are missing test items in the medical information D1 of the target patient, the generating function 255 notifies the converting function 254 of the missing test items. The converting function 254 derives the test values ​​of the missing test items notified from the generating function 255 from the test values ​​of the existing test items by using the conversion function described above, and inputs the derivation results to the generating function 255.

[0100] When the generation function 255 obtains test values, etc. for the missing test items from the conversion function 254, it generates a medical information table that integrates the test values ​​of the existing test items contained in the medical information D1 of the target patient and the test values, etc. for the test items obtained from the conversion function 254.

[0101] Fig. 11 is a diagram showing an example of a medical information table. The medical information table is a memory unit for storing medical information D1 of target patients, etc., held in the memory circuitry 24, etc. Fig. 11 shows an example in which there are three target patients (patient IDs: 1001, 1002, 1003). Fig. 11 also shows an example in which "blood pressure" and "image information" are included as existing test items included in the medical information D1.

[0102] For example, when test values ​​of the test items "blood pressure" and "CTR" are specified as input data for the trained model M1, CTR becomes the missing test item in the medical information D1 shown in Fig. 11. In this case, the generation function 255 notifies the conversion function 254 of the missing test item "CTR". In this case, the conversion function 254 executes a process for deriving the test value of the missing test item "CTR" from the existing image information included in the medical information D1 by using the conversion function calculated by the above-mentioned third calculation method.

[0103] When the generation function 255 obtains the derivation result of the test value of the test item "CTR" from the conversion function 254, it integrates it with the medical information D1 of the corresponding target patient and stores it in a medical information table, as shown in Fig. 11. In Fig. 11, the existing test items "blood pressure" and "image information" included in the medical information D1 and the test item "CTR" derived by the conversion function 254 are shown with different types of hatching lines. Also, an arrow indicates that the test value of the test item "CTR" is derived from the test item "image information".

[0104] Here, when there are multiple test values ​​derived by the conversion function 254, each of the test values ​​is registered in the medical information table as shown in Fig. 11. Specifically, in Fig. 11, based on the characteristic information of each test value, the test value with the highest probability is "estimated" and the probability is calculated based on the threshold value Th pThe minimum test value in the range above is set as "minimum", and the probability is set to the threshold value Th p The maximum test value within the above range is indicated as "Max."

[0105] When the generation function 255 stores the medical information D1 of the target patient and the derivation result of the conversion function 254 in the medical information table, it inputs each test value of the test items "blood pressure" and "CTR" into the trained model M1 and acquires an inference result of the trained model M1. More specifically, the generation function 255 inputs the test value for each characteristic of the test value of the test item "CTR" together with the test value of the test item "blood pressure" into the trained model M1, thereby acquiring an inference result of the trained model M1 for each characteristic of the test value of the test item "CTR".

[0106] In the trained model M1, when test values ​​of blood pressure and CTR are input, for example, the name of the disease inferred from the relationship between the two test values ​​and a risk value that quantitatively represents the possibility of having the disease and the incidence rate are output as inference results. Then, the generation function 255 generates support information based on the inference results for each characteristic of the test values ​​inferred by the trained model M1, and registers the generated support information in a support information table.

[0107] 12 is a diagram showing an example of a support information table. The support information table is a storage unit for holding support information stored in the storage circuitry 24 or the like.

[0108] The support information table in FIG. 12 shows support information obtained by inputting each test value of the test items "blood pressure" and "CTR" shown in the medical information table in FIG. 11 into the trained model M1. For example, focusing on the target patient with patient ID: 1001, it shows that the disease name "heart failure" was inferred from the test values ​​of the test items "blood pressure" and "CTR" of this target patient. In addition, the risk values ​​inferred for each of the test values ​​of "CTR (estimated)", "CTR (minimum)", and "CTR (maximum)" are registered in association with the disease name.

[0109] Here, "risk (estimated)" means a risk value inferred from the test value of "CTR (estimated)". In addition, in "risk (minimum)" and "CTR at minimum risk", the smaller risk value of the risk values ​​inferred for "CTR (minimum)" and "CTR (maximum)" and the test value of CTR corresponding to the risk value are registered. In the example of FIG. 12, it means that the risk value is minimum when the test value of CTR is 45. In addition, in "risk (maximum)" and "CTR at maximum risk", the larger risk value of the risk values ​​inferred for "CTR (minimum)" and "CTR (maximum)" and the test value of CTR corresponding to the risk value are registered. In the example of FIG. 12, it means that the risk value is maximum when the test value of CTR is 63.

[0110] When the risk value inferred from the test value of "CTR (estimated)" is the minimum, the risk value may be registered in the "Risk (minimum)" field, and the test value of "CTR (estimated)" may be registered in the "CTR at minimum risk" field. Also, when the risk value inferred from the test value of "CTR (estimated)" is the maximum, the same registration may be performed.

[0111] The display configuration function 256 generates a display configuration table that serves as the basis of a screen displayed by the display function 257 described later, based on the medical information D1 and support information registered in the medical information table and support information table.

[0112] Specifically, the display configuration function 256 generates a display configuration table by reconfiguring the registered contents of both the medical information table and the support information table based on the registered contents of both tables.

[0113] Fig. 13 is a diagram showing an example of the data configuration of the display configuration table. As shown in Fig. 13, the display configuration table associates the registered contents of the medical information table shown in Fig. 11 with the registered contents of the support information table shown in Fig. 12 based on the patient ID and date of the target patient. In other words, the display configuration table includes the test values ​​of the existing test items "blood pressure" and "image information" acquired from the target patient, the test value of "CTR" derived by the conversion function 254, and the inference result of the target patient inferred from these test values.

[0114] Fig. 13 also shows an example of reconstructing the registered contents of the test item "CTR" registered in the medical information table. Specifically, Fig. 13 shows an example in which only the test value of "CTR (estimated)", which has the highest probability, is registered among the test items "CTR" shown in Fig. 11. Fig. 13 also shows the probability of the test value "CTR (estimated)" being associated with the test value and registered as "reliability".

[0115] FIG. 13 also shows an example in which three disease names "heart failure," "lung cancer," and "diabetes" inferred for a target patient with a patient ID of "1001" are arranged side by side. These disease names may be inferred, for example, from trained models M1 of different medical departments, or may be inferred from a single trained model M1. In either case, when there are missing test items, the medical support device 20 can derive test values ​​for the missing test items using a conversion function, and can therefore flexibly accommodate various trained models M1.

[0116] The display configuration table generated by the display configuration function 256 is not limited to the example in Fig. 13. For example, inference results (disease names) with risk values ​​below a threshold value may be excluded from the display configuration information. Also, the test values ​​"CTR (minimum)" and "CTR (maximum)" shown in Fig. 11 may be registered in the display configuration table.

[0117] The display function 257 is an example of an output unit. The display function 257 displays, on the display 22, a screen for supporting medical treatment of a target patient based on the display configuration table generated by the display configuration function 256.

[0118] Fig. 14 is a diagram showing an example of a screen displayed by the display function 257. Fig. 14 shows a screen relating to patient ID: 1001 registered in the display configuration table of Fig. 13.

[0119] 14, the screen G1 has a first area G11 to a fourth area G14. In the first area G11, the patient ID of the target patient registered in the display configuration table is displayed.

[0120] In the second area G12, existing medical information D1 (examination information, image information, etc.) registered in the display configuration table is displayed. Fig. 14 shows an example in which blood pressure examination values ​​and X-ray images, which are image information, are displayed. Note that, although Fig. 14 shows blood pressure examination values ​​as time-series data, it is also possible to display only examination values ​​on a specific date.

[0121] In the third area G13, the derivation result of the conversion function registered in the display configuration table is displayed as estimated information. Fig. 14 shows an example in which the original examination item (image information) "X-ray", the converted examination item "CTR", and the examination values ​​(estimated), (minimum), and (maximum) are displayed as estimated information.

[0122] In the fourth area G14, the inference results of the trained model registered in the display configuration table are displayed as disease risks. Fig. 14 shows an example in which each inferred disease name and the risk value of the disease name are displayed as a horizontal bar graph as the disease risk.

[0123] Here, the bar graph is based on the risk value of the risk (estimated) item, and the range of risk values ​​between risk (minimum) and risk (maximum) is represented by a gauge G14a superimposed on the bar graph. In other words, the display function 257 displays a plurality of disease names and risk values ​​inferred by the trained model M1 in a comparable state.

[0124] Furthermore, the display function 257 changes the display form of the bar graph according to the magnitude of the risk value. Specifically, when the risk value is equal to or greater than a threshold value (e.g., 70%), the display function 257 displays the bar graph representing the risk value in a more emphasized manner than the bar graph representing a risk value less than the threshold value. This allows a user, such as a doctor, who views the screen G1 to easily recognize the names of diseases with high risk values.

[0125] Furthermore, when the mouse pointer PO operated by the operator moves to the gauge G14a, the display function 257 displays the risk values ​​of the risk (minimum) and risk (maximum) defined by the gauge G14a in a pop-up window G15.

[0126] In this way, the display function 257 distinguishably displays the test values ​​of existing test items included in the medical information D1 of the target patient and the test values ​​of the test items derived by the conversion function 254. The display function 257 also distinguishably displays the existing test items from which the test values ​​of the test items derived by the conversion function 254 were derived. The display function 257 also displays the disease name of the target patient inferred from the input of the test values ​​and its risk value in a comparable state.

[0127] As a result, the medical support device 20 allows a user such as a doctor viewing the screen G1 to easily understand the disease name and risk value to which the target patient may be suffering, thereby efficiently providing medical support to the target patient. In addition, the medical support device 20 allows a user to easily understand which test item's test value is insufficient in the inference of the applied trained model M1. Therefore, a user such as a doctor can make a decision based on the screen G1, such as performing additional tests for the insufficient test items, depending on the inferred disease name, risk value, etc.

[0128] The screen displayed by the display function 257 is not limited to the example screen shown in Fig. 14, and may be displayed in other display formats. For example, the display function 257 may display the display configuration table shown in Fig. 13 on the display 22. Even when displaying the display configuration table, it is preferable that the display function 257 highlights the disease name or the like according to the risk value. In addition, it is possible to switch the target patient to be displayed via the input interface 21 or the like.

[0129] The following describes the operation of the medical support device 20. Fig. 15 is a flowchart showing an example of a conversion function calculation process executed by the medical support device 20.

[0130] First, the first acquisition function 251 acquires the past medical information D1 of a patient stored in the medical DB 10 as sample data, and inputs the acquired sample data to the calculation function 252 (step S11).

[0131] Here, the medical DB10 from which the sample data is obtained is not particularly limited, and may be, for example, the medical DB10 of a specific medical department (e.g., the first medical department A), or it may be all of the medical DBs 10 provided in each medical department.

[0132] Furthermore, the first acquisition function 251 may acquire all medical information D1 stored in the medical DB 10 as sample data, or may acquire only medical information D1 that meets a condition as sample data. Examples of sample data acquisition conditions include conditions related to characteristic information such as a specific patient (patient ID) or disease name. The sample data acquisition conditions may be instructed via the input interface 21 or the communication interface 23, or may be preset. For example, a setting may be made to acquire sample data for each disease name.

[0133] Furthermore, the timing at which the first acquisition function 251 acquires the sample data is not particularly important, and the sample data can be acquired at any timing. For example, the first acquisition function 251 may start acquiring sample data when it receives an instruction via the input interface 21. Furthermore, for example, the first acquisition function 251 may acquire sample data according to a preset schedule. In the latter case, the schedule may be set to an execution timing, such as once a day or once a week.

[0134] Next, the calculation function 252 calculates the above-mentioned conversion function for each combination of test items based on the correlation between the test values ​​of each test item included in the input sample data (step S12).Then, the calculation function 252 stores the calculated conversion function in the conversion function table T1 (step S13), and ends this process.

[0135] By the above-mentioned conversion function calculation process, various conversion functions are stored in the conversion function table T1.

[0136] 16 is a flowchart showing an example of medical support processing executed by the medical support device 20. Note that, as a premise of this processing, it is assumed that various conversion functions have already been stored in the conversion function table T1.

[0137] First, the second acquisition function 253 acquires medical information D1 of the target patient from the medical DB 10 in which the patient ID of the target patient is registered, and inputs the acquired medical information D1 to the conversion function 254 and the generation function 255 (step S21).

[0138] Here, the patient ID of the target patient is specified by an operator such as a doctor via, for example, the input interface 21 or the communication interface 23. When the patient ID of the target patient is registered in multiple medical DBs 10 (for example, medical DBs 10a and 10b), the second acquisition function 253 may acquire the medical information D1 from each of the medical DBs 10, or may have the operator select the medical DB 10 (medical department) from which the medical information is to be acquired.

[0139] When the medical information D1 is input, the generating function 255 compares the test items included in the medical information D1 with the test items of the input data required for inference in the inference device 30 (trained model M1) and determines whether or not there are any missing test items. If there are any missing test items, the generating function 255 notifies the converting function 254 of the missing test items.

[0140] If there are no missing test items, the generation function 255 inputs the test values ​​of the existing test items included in the medical information D1 into the trained model M1, and causes the trained model M1 to infer the disease name, etc. The processing in this case is similar to the system configuration of conventional CDS systems and CAD systems, so a description thereof will be omitted.

[0141] If there are missing test items, the conversion function 254 executes a conversion process to derive test values ​​for the missing test items from the test values ​​of existing test items included in the medical information D1 based on the conversion function stored in the conversion function table T1 (step S22).

[0142] Next, the generation function 255 inputs the test values ​​of the existing test items included in the medical information D1 and the test values ​​of the missing test items derived in step S22 to the trained model M1 (step S23). Next, the generation function 255 generates support information based on the inference results output by the trained model M1 (step S24).

[0143] Next, the display configuration function 256 generates a display configuration table for configuring a display screen based on the medical information D1, the converted test values ​​derived in step S22, the support information generated in step S24, etc. (step S25). Then, the display function 257 displays a display screen (for example, screen G1) according to the display configuration table (step S26), and ends this process.

[0144] In this way, the medical support device 20 calculates a conversion function capable of statistically deriving possible values ​​of the test values ​​of one or more result items from the test values ​​of other test items based on the correlation between the test values ​​of each test item included in the medical information D1 of multiple patients. In addition, the medical support device 20 generates support information for supporting the medical treatment of the target patient based on an inference result obtained by inputting the test values ​​of the test items included in the medical information D1 of the target patient to the trained model M1 that infers information about the patient's condition from the test values ​​of the preset test items. In addition, when the test items included in the medical information D1 of the target patient do not satisfy the requirements of the preset test items to be input to the trained model M1, the medical support device 20 derives the test values ​​of the missing test items from the test values ​​of the test items included in the medical information D1 of the target patient based on the conversion function, and causes the trained model M1 to infer information about the condition of the target patient using the derived test values ​​of the missing test items.

[0145] As a result, even if a test item included in the medical information D1 of a target patient does not satisfy the requirements of the trained model M1, the medical support device 20 can obtain an inference result of the trained model M1 by using the test value of the test item derived based on the conversion function. In addition, since the conversion function is calculated based on the correlation between the test values ​​of each test item, the medical support device 20 can derive a statistically significant test value.

[0146] In this way, the medical support device 20 can automatically supplement the missing test results from the medical information D1 of the target patient according to the specifications of the applied trained model M1. Therefore, the medical support device 20 can obtain the inference results of the applied trained model M1 without performing additional tests, and can flexibly provide medical support to patients. As a result, the medical support system 1 can easily obtain inference results such as the disease name and risk value of the target patient using, for example, the trained model M1 of a department other than the department visited by the target patient, thereby contributing to early detection of diseases.

[0147] The above-described embodiment can be modified as appropriate by changing a part of the configuration or function of each device. Therefore, some modified examples of the above-described embodiment will be described below as other embodiments. Note that the following mainly describes the differences from the above-described embodiment, and a detailed description of the common points with the contents already described will be omitted. The modified examples described below may be implemented individually or in appropriate combination.

[0148] (Variation 1) In the above-described embodiment, the medical support device 20 performs the process of calculating the conversion function and the process of deriving the test value using the conversion function, but each of these processes may be executed by different devices.

[0149] In this case, the device that executes the process related to calculation of the conversion function (hereinafter referred to as the medical information processing device) has the same hardware configuration as in Fig. 2, and has the above-mentioned first acquisition function 251 and calculation function 252. Moreover, the device that executes the process related to derivation of test values ​​using the conversion function (hereinafter referred to as the medical support device) has the same hardware configuration as in Fig. 2, and has the above-mentioned second acquisition function 253, conversion function 254, generation function 255, display configuration function 256, and display function 257.

[0150] That is, in the medical support system 1 according to this modification, the medical support device derives the test values ​​of the missing test items from the medical information D1 of the target patient using the conversion function calculated by the medical information processing device. As a result, the medical support system 1 according to this modification can achieve the same actions and effects as the above-mentioned embodiment, and can flexibly provide medical support to patients.

[0151] (Variation 2) In the above embodiment, the medical support device 20 is configured to accept the designated input of the target patient, the trained model M1, etc., via the input interface 21, but is not limited thereto, and may be configured to accept the designated input of the target patient, the trained model M1, etc., from an external device other than the medical support device 20. For example, the medical support device 20 may accept the designated input of the target patient, the trained model M1, etc., via the communication interface 23 from a terminal device (not shown) provided in the first medical department A or the second medical department B. Also, for example, the second acquisition function 253 of the medical support device 20 may be configured to acquire the medical information D1 of the target patient transmitted from the external device via the communication interface 23.

[0152] In addition, in the above-described embodiment, the display function 257 displays and outputs the contents of the target patient's medical information D1, the derivation result (conversion result) of the conversion function 254, and the support information (support information table) generated by the generation function 255 on the display 22, but the method of outputting this information is not limited to display.

[0153] For example, the display function 257 may be configured to output (store) a screen based on the display configuration table to the storage circuitry 24 or the like. Also, the display function 257 may be configured to output (transmit) a screen based on the display configuration table to an external device via the communication interface 23.

[0154] In addition, the data to be output is not limited to a screen based on the display configuration table. For example, the medical information table or the support information table generated by the generating function 255 may be output to the display 22 or an external device. In this case, the generating function 255 may function as an example of an output unit by cooperating with the display function 257.

[0155] (Variation 3) In the above embodiment, the medical support device 20 infers information about the condition of a target patient using the medical DB 10 and the inference device 30 provided in each of the medical departments (first medical department A, second medical department B). However, the configuration of the medical support system 1 is not limited to this. For example, the medical support device 20 may infer information about the condition of a target patient using the medical DB 10 and the inference device 30 provided in each of the medical facilities such as a hospital.

[0156] [Second embodiment] In the above embodiment, the trained model M1 has been trained in advance, but the training process may be performed during a series of processes executed by the medical support device 20. The medical support device 20 according to this embodiment will be described below. Note that the same elements as those in the first embodiment are given the same reference numerals and will not be described.

[0157] Fig. 17 is a diagram showing an example of components included in the processing circuitry 25 of the medical support device 20 according to the second embodiment. The processing circuitry 25 according to this embodiment has a learning function 258 in addition to the functions described in Fig. 3. The learning function 258 is stored in the storage circuitry 24 in the form of a program executable by a computer, for example.

[0158] The learning function 258 is an example of a learning unit. The learning function 258 executes a learning process of the learned model M1. Specifically, the learning function 258 executes additional learning (fine tuning, etc.) of the learned model M1 by using the test value of the test item included in the medical information acquired by the second acquisition function 253 and the test value of another test item derived from the test value by the conversion function 254 as learning data. In this case, the teacher data (disease name, etc.) may be, for example, data input by the operator of the medical support device 20 or data selected from the inference results displayed by the display function 257.

[0159] As shown by the dashed line in the figure, the input of medical information to the learning function 258 may be performed by the first acquisition function 251. In addition, the trained model M1 to be subjected to additional learning is not particularly limited and can be set arbitrarily.

[0160] Next, the operation of the medical support device 20 of this embodiment will be described with reference to Fig. 18. Here, Fig. 18 is a flowchart showing an example of a learning process executed by the medical support device 20 of this embodiment.

[0161] First, the second acquisition function 253 acquires medical information for learning from the medical DB 10, and inputs the medical information to the conversion function 254 and the learning function 258 (step S31). Here, the medical information for learning is not particularly limited, and may be, for example, the medical information of the target patient described above.

[0162] The conversion function 254 cooperates with the generation function 255 to derive (convert) test values ​​of missing test items among the test items that are input parameters of the trained model M1 to be trained, from the medical information input in step S31. Then, the conversion function 254 inputs the converted test values ​​of the test items to the learning function 258 (step S32).

[0163] Next, the learning function 258 executes a learning process to perform additional learning of the trained model M1 based on the test values ​​of the test items included in the medical information and the test values ​​of the test items converted by the conversion function 254 (step S33).

[0164] As a result, in the medical support system 1 according to this embodiment, the derivation result of the conversion function 254 can be used to train the trained model M1. Therefore, in the medical support system 1 according to this embodiment, it is possible to maintain the expertise of the trained model M1 specialized for each medical department, and to perform additional training of the trained model using the latest medical information.

[0165] In the above embodiment, the functional configuration of the medical support device 20 is realized by the processing circuitry 25, but the embodiment is not limited to this. For example, the functional configuration in this specification may be realized by hardware alone or a combination of hardware and software.

[0166] The term "processor" used in the above description means, for example, a circuit such as a central processing unit (CPU), a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a programmable logic device (for example, a simple programmable logic device (SPLD), a complex programmable logic device (CPLD), and a field programmable gate array (FPGA)). When the processor is, for example, a CPU, the processor realizes a function by reading and executing a program stored in the memory circuit 24. On the other hand, when the processor is, for example, an ASIC, instead of storing a program in the memory circuit 24, the function is directly incorporated as a logic circuit in the circuit of the processor. Note that each processor in this embodiment is not limited to being configured as a single circuit for each processor, but may be configured as one processor by combining multiple independent circuits to realize the function. Furthermore, the multiple components in each figure may be integrated into one processor to realize the function.

[0167] Here, the program executed by the processor is provided in advance in a ROM (Read Only Memory) or a storage circuit. The program may be provided in a format that can be installed in these devices or in a format that can be executed, recorded on a computer-readable storage medium such as a CD (Compact Disk)-ROM, a FD (Flexible Disk), a CD-R (Recordable), or a DVD (Digital Versatile Disk). The program may be provided or distributed by being stored on a computer connected to a network such as the Internet and downloaded via the network. For example, the program is composed of modules including each of the above-mentioned functional units. As actual hardware, a CPU reads out the program from a storage medium such as a ROM and executes it, so that each module is loaded onto a main storage device and generated on the main storage device.

[0168] According to at least one of the embodiments described above, medical support for patients can be performed more flexibly.

[0169] Although several embodiments (variations) have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, modifications, and combinations of embodiments can be made without departing from the gist of the invention. These embodiments and their variations are included in the scope of the invention and its equivalents as described in the claims, as well as in the scope and gist of the invention. [Explanation of symbols]

[0170] 1 Medical Support System 10, 10a, 10b Medical DB 20 Medical support equipment 30, 30a, 30b Reasoning device 251 First Acquisition Function 252 Calculation Function 253 Second Acquisition Function 254 conversion function 255 Generation function 256 Display configuration function 257 Display function 258 Learning Function A. First medical department B. Second medical department

Claims

1. a calculation unit that calculates a conversion function that can statistically derive possible values ​​of test values ​​of one or more result items from the test values ​​of other test items based on correlations between test values ​​of each test item included in the medical information of multiple patients; a storage unit that stores the conversion function calculated by the calculation unit in association with a condition related to the test item used in the calculation of the conversion function; A generation unit that generates support information for supporting medical care of a target patient based on an inference result obtained by inputting test values ​​of test items included in the medical information of a target patient into a trained model that infers information about a patient's condition from test values ​​of predetermined test items; a derivation unit that, when an examination item included in the medical information of the target patient does not satisfy the requirements of the predetermined examination item, derives an examination value of the missing examination item by using a conversion function that can derive the examination value of the missing examination item from the examination item included in the medical information of the target patient, among the conversion functions stored in the storage unit; having the storage unit stores, as conditions related to the test items, index values ​​based on any or all of the total number of the medical information used in calculating the conversion function, the number of days on which the medical information was acquired, the number of the test items included in the medical information, and the number of the plurality of patients; When a plurality of conversion functions capable of deriving the test values ​​of the missing test items are stored in the storage unit, the derivation unit selects a conversion function to be used based on the condition associated with the conversion function, The generation unit is a medical support system that causes the trained model to infer information regarding the condition of the target patient using the test values ​​of the missing test items derived by the derivation unit.

2. 2. The medical support system according to claim 1, wherein the derivation unit derives a value with a maximum probability as the test value of the missing test item from a probability distribution of values ​​that the test value of the missing test item can take, which is derived by the transformation function.

3. 3. The medical support system according to claim 2, wherein the derivation unit derives, as the test values ​​of the missing test items, minimum and maximum values ​​from a range in which the probability is equal to or greater than a threshold value, of a probability distribution of values ​​that the test values ​​of the missing test items can take, derived by the transformation function.

4. The medical support system according to claim 3 , wherein the generation unit inputs each of the test values ​​derived by the derivation unit into the trained model to infer information regarding a condition of the target patient for each characteristic of the test value.

5. The medical support system according to claim 1 , further comprising an output unit that outputs the support information generated by the generation unit.

6. The medical support system according to claim 5 , wherein the output unit displays a screen based on the support information.

7. 7. The medical support system according to claim 6, wherein the output unit displays, on a single screen, the test values ​​of the test items included in the medical information of the target patient, the test values ​​of the missing test items derived by the derivation unit, and the contents of the support information.

8. The medical support system according to claim 1 , wherein the calculation unit comprehensively combines test items included in the medical information and calculates the conversion function for each combination of test items.

9. The medical support system according to claim 1 , wherein the calculation unit collects the medical information created on the same day for each patient, and calculates the conversion function using the collected medical information.

10. The medical support system according to claim 1 , wherein the calculation unit collects the chronologically continuous medical information for each patient, and calculates the conversion function using the collected medical information.

11. 2. The medical support system according to claim 1, wherein the calculation unit calculates a conversion function capable of deriving a probability distribution of possible values ​​of the test value of the test item from the image information, based on a correlation between image information included in the medical information and the test value of the test item related to the image information.

12. 2. The medical support system according to claim 1, wherein the calculation unit calculates a conversion function capable of deriving a probability distribution of values ​​that the test value of the test item can take from the wording based on a correlation between the wording described in the description information included in the medical information and the test value of the test item related to the wording.

13. 2. The medical support system according to claim 1, further comprising a learning unit that executes a learning process of the trained model based on test values ​​of test items included in the medical information of the target patient and test values ​​of the missing test items derived by the derivation unit.

Images