Diagnosis assistance system and diagnosis assistance method

Abstract

The present disclosure proposes a technology for establishing the relationship between a specific numerical value (inspection value) of an inspection result and a disease, and recommending highly accurate inspection contents (secondary inspection) for diagnosis of various diseases. As an example, the present disclosure proposes a diagnosis assistance system that generates a trained model for estimating the inspection worth of at least one inspection item included in the secondary inspection from a plurality of patient inspection results and patient information prepared for training. The diagnosis assistance system executes: data acquisition processing for acquiring patient information of a plurality of patients, inspection items of a plurality of inspections of the plurality of patients, time-series data of the inspection results, and time-series data of disease information of the plurality of patients; feature amount extraction processing for extracting time-series data of at least feature amounts of the inspection results by using the inspection items of the plurality of inspections, the time-series data of the inspection results, and inspection item information stored in advance; time-series training processing for performing time-series training regarding the time-series data of the feature amounts extracted by the feature amount extraction processing, and outputting time-series feature amounts; probability distribution estimation processing for estimating the probability distribution of the time-series feature amounts from the time-series training processing; inspection result restoration training processing for restoring, on the basis of the probability distribution, the result of the secondary inspection at each time point in the time series; and processing for determining, on the basis of the result of the secondary inspection at each time point in the time series restored by the inspection result restoration training processing, a parameter for estimating the inspection worth of the inspection items in the trained model.

Description

Diagnosis support system and diagnosis support method

[0001] The present disclosure relates to a diagnosis support system and a diagnosis support method.

[0002] The task of a physician involved in ordering tests is to integrate and interpret a large amount of clinical information and select appropriate tests for each patient. Specifically, the physician (i) examines the patient and collects patient information, lists suspected diseases based on the examination results, and prioritizes the diseases, (ii) selects and performs necessary tests from all possible tests, taking into account their contribution to disease identification and the patient burden (issuing a test order), (iii) interprets the test results and updates the list of suspected diseases, and (iv) repeats steps (ii) and (iii) until the disease is identified.

[0003] In order to accurately order such tests, doctors often need extensive experience. In particular, they must properly interpret a large number of test results (primary test results) to make a diagnosis and then perform subsequent tests (secondary tests) to avoid overlooking any diseases. In response to this situation, diagnostic support systems that assist in the interpretation of test results (for example, the test result analysis support system by Abbott Japan Co., Ltd.) have been proposed.

[0004] JP 2015-197761 A

[0005] However, existing diagnostic support systems require doctors and other medical professionals to build logic for interpreting test results for each patient, which still places a heavy burden on doctors and other medical professionals. Furthermore, because the logic must be built for each patient, it is difficult to apply the same logic to other patients once it has been built, which increases sales costs (making it impossible to provide a general-purpose system). Furthermore, the logic for interpreting test results used in existing diagnostic support systems is limited to simple systems, making them unsuitable for building complex logic for interpreting test results for specific diseases.

[0006] In light of this situation, the present disclosure proposes a technology that establishes a relationship between specific numerical values ​​of test results (test values) and diseases, and recommends highly accurate test contents (secondary tests) for diagnosing various diseases.

[0007] In order to solve the above-mentioned problems, the present disclosure provides, as an example, a diagnostic support system that generates a trained model that estimates the test value of at least one test item included in a secondary test from test results and patient information of a plurality of patients prepared for learning, the system comprising: a storage device that stores a program for generating the trained model; and a computer that generates the trained model by reading and executing the program from the storage device, wherein the computer performs: a data acquisition process that acquires patient information of the plurality of patients, time series data of test items of a plurality of tests on the plurality of patients and their test results, and time series data of disease information of the plurality of patients; a feature extraction process that extracts time series data of at least feature amounts of the test results using the time series data of the test items of the plurality of tests and their test results and test item information that has been stored in advance; a time series learning process that performs time series learning on the time series data of the feature amounts extracted by the feature extraction process and outputs time series feature amounts; and a probability distribution estimation process that estimates a probability distribution of the time series feature amounts obtained by the time series learning process. We propose a diagnostic support system that performs the following: a test result restoration learning process that restores the results of the secondary test at each point in time in the time series based on the probability distribution; and a process that determines parameters for estimating the test value of the test item in the trained model based on the results of the secondary test at each point in time in the time series restored by the test result restoration learning process.

[0008] Further features related to the present disclosure will become apparent from the description of this specification and the accompanying drawings. Also, aspects of the present disclosure are achieved and realized by the elements and combinations of various elements and the aspects of the following detailed description and the appended claims. The description of this specification is merely exemplary and does not limit the scope or application of the claims of the present disclosure in any way.

[0009] According to the technology disclosed herein, it is possible to establish a relationship between specific numerical values ​​(test values) of test results and diseases, and to recommend highly accurate test contents (secondary tests) for diagnosing various diseases.

[0010] 1 is a diagram showing an example of a schematic configuration of an information processing system (diagnosis support system) 10 according to this embodiment. FIG. 2 is a diagram showing an example of a functional configuration (example of a software configuration) of the information processing system (diagnosis support system) 10 according to this embodiment. FIG. 3 is a flowchart for explaining an overview of the learning process by the test order generation unit 1012. FIG. 4 is a flowchart for explaining details of the learning process by the test order generation unit 1012. FIG. 5 is a diagram showing an example of a configuration of the architecture of the learning process used in the learning processes of FIGS. 3 and 4. FIG. 6 is a diagram showing an example of restoring test value data including missing values. FIG. 7 is a flowchart for explaining an overview of a process for predicting a patient's disease and generating a test order by applying patient information and test results of a suspected patient (new patient) to a trained model. FIG. 8 is a flowchart for explaining details of a process for predicting a patient's disease and generating a test order by applying patient information and test results of a suspected patient (new patient) to a trained model. FIG. 9 is a diagram showing an example of specific test value data 900. FIG. 10 is a diagram showing an example of a configuration of a screen display (test order GUI) 1000 that outputs a patient's disease predicted by a trained model and recommended test items. FIG. 11 is a diagram showing an example of the configuration of a screen (time-series pathological condition transition analysis result display 1100) for displaying the results of time-series pathological condition transition analysis.

[0011] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. In the accompanying drawings, functionally identical elements may be designated by the same numerals. Note that the accompanying drawings illustrate specific embodiments and implementation examples according to the principles of the present disclosure, but these are for understanding the present disclosure and are not to be used to interpret the present disclosure in any way as being limiting.

[0012] Although the present embodiment has been described in sufficient detail to enable those skilled in the art to practice the present disclosure, it should be understood that other implementations and forms are possible, and that changes in configuration and structure and substitutions of various elements are possible without departing from the scope and spirit of the technical ideas of the present disclosure. Therefore, the following description should not be interpreted as being limited thereto.

[0013] <Motivation for the idea of ​​the disclosed technology> Most diseases can be diagnosed based on the results of a primary test such as a blood test, and the contents of the next test (secondary test) can often be easily predicted. However, there are some diseases for which this is difficult. For example, some thyroid diseases can be predicted, while others cannot. In other words, it is not possible to determine whether a patient has Graves' disease or simply destructive thyroiditis based on the results of the primary test alone.

[0014] Faced with this situation, the inventors realized that, in the above-mentioned example of thyroid disease, because the same thyrotoxicosis has different prescription options, it is difficult to determine which disease it is without using time-series data of test results. Therefore, the inventors thought that in order to prevent incorrect diagnoses, it is necessary to capture changes over time as a solution for disease diagnosis and predict and optimize the content of secondary tests at the appropriate time.

[0015] On the other hand, in typical cases, it may be possible to easily determine the test content using time-series data from a small number of test results. However, attempting to comprehensively determine (predict) the results of over 10 different tests by capturing trends over time for each is cumbersome and impossible in clinical practice. Ideally, the interpretation of test results would also include subtle changes over time that doctors may not notice. In other words, it would be desirable to be able to determine whether a case does not show typical time-series changes, but whether the trend indicates that a medication should be prescribed, and to provide appropriate support for such cases. Therefore, after extensive research, the inventors discovered that the above situation could be improved by using AI to model the relationship between test values ​​and disease.

[0016] <Example of Hardware Configuration of Information Processing System> FIG. 1 is a diagram showing an example of the schematic configuration of an information processing system (also called a diagnosis support system) 10 according to this embodiment.

[0017] The information processing system (diagnostic support system) 10 includes, for example, a computer 100 provided on the cloud, an information terminal 110 installed in a doctor's examination room, and an external recording device (medical information server) 111, which are connected via a network (hospital network) 109.

[0018] The computer 100 includes a processor 101 such as a CPU, a main memory device 102, a secondary memory device 103, a network adapter 104, an input device 105, and an output device 106. The main memory device 102 holds, for example, a program for determining an examination order, which will be described later. The secondary memory device 103 stores, for example, information (electronic medical record information, medical receipt information, guideline information, etc.) acquired from an external recording device 111. The processor 101 reads the program from the main memory device 102, expands it into an internal memory (not shown), and realizes each processing unit (various functions for determining an examination order), which will be described later.

[0019] The input device 105 is composed of devices for inputting information to the computer 100, such as a keyboard, a mouse, or a touch panel. The output device 106 is composed of devices for outputting information, such as a display or a printer. The network adapter 104 receives information from the information terminal 110 via the network 109, passes the information to the processor 101, and stores it in the secondary storage device 103. Furthermore, the network adapter 104 transmits determined examination order information to the information terminal 110 via the network 109 in response to, for example, a command from the processor 101.

[0020] <Example of Functional Configuration of Information Processing System> FIG. 2 is a diagram showing an example of the functional configuration (example of software configuration) of the information processing system (diagnosis support system) 10 according to this embodiment.

[0021] The information processing system 10 has, as its functions, a patient information acquisition unit 1011 and an examination order generation unit 1012 set up in a processor 101 in a computer 100, an examination order output unit 1013 provided in an information terminal 110, and an examination ordering system 1114 provided in an external recording device (medical information server) 111.

[0022] The information processing system 10 also handles the following information stored in an external storage device (medical information server) 111: electronic medical record information 1111, medical receipt information 1112, and guideline information 1113; patient information 201, time-series data 202 of test items and test result information, time-series data 203 of patient perception information, restored test results 204, future test item candidates 205, test value 206, and patient disease prediction results 207. The medical receipt information 1112 includes medical fee (point) information. The guideline information 1113 stores information on test items medically determined by symptoms (e.g., test X should be performed for symptoms A and B).

[0023] The patient information acquisition unit 1011 accesses the electronic medical record information 1111 in the external recording device (medical information server) 111, acquires the identification information of the target patient, information on past tests (test items and test results) corresponding to the target patient, and information on past diagnoses, and acquires this as patient information 201, time-series data 202 of test items and test result information, and time-series data 203 of patient disease information. The patient information 201 is information including, for example, patient attribute information (such as gender and age). The time-series data 202 of test items and test result information includes the items and test result information of tests previously taken by the target patient. The time-series data 203 of patient disease information includes time-series data of patient disease information (included in the electronic medical record information) provisionally determined from test results for the target patient.

[0024] The test order generation unit 1012 uses patient information 201, time-series data 202 of test items and test result information, and time-series data 203 of patient disease information to generate test result reconstructions (values) 204, future test item candidates 205, test values ​​206 indicating the amount of information according to disease type (the probability that a disease can be identified by each test item), and patient disease prediction results 207. Test items with high test values ​​(large amount of information: high probability value) are identified as the future test item candidates 205. The test order generation unit 1012 transmits the identified future test item candidates 205 to the information terminal 110 as recommended test information.

[0025] The examination order output unit 1013 receives the future examination item candidates 205 and displays them on the display screen (not shown) of the information terminal 110. When a doctor checks the future examination item candidates 205 displayed on the display screen and selects a necessary examination item from the candidates, the examination order output unit 1013 transmits an examination order to the examination ordering system 1114. The examination ordering system 1114 places the examination order received from the information terminal 110 to the examination room.

[0026] <Overview of Learning Process> Fig. 3 is a flowchart for explaining an overview of the learning process performed by the test order generation unit 1012. Note that in Fig. 3, rectangles with rounded corners represent information or data, and rectangles with corners represent processing. In addition, an architecture for the learning process and an example of learning processing for test value data including missing values ​​(defective values) will be described later with reference to Figs. 5 and 6, respectively.

[0027] (i) Step S301 The test order generation unit 1012 receives as input the patient information 201, the time-series data 202 of test items and test result information, and the time-series data 203 of patient disease information for learning, extracts features of the test value data, and generates time-series data of the features.

[0028] (ii) Step S302 The examination order generation unit 1012 learns the time-series data of the features generated in step S301 using a neural network (for example, repetitive processing using LSTM (Long Short Term Memory)) or the like, and generates time-series features (LSTM time-series features).

[0029] (iii) Step S303: The test order generation unit 1012 generates a trained model by restoring and learning the time-series data of the test results and learning a patient disease prediction model based on the time-series feature quantities. Note that the former restores missing values ​​(missing values ​​of the secondary test results) in the time-series data 202 of the input test result information.

[0030] 4 is a flowchart for explaining details of the learning process by the inspection order generation unit 1012. Note that steps S401 and S402 correspond to step S301 in FIG. 3, step S403 corresponds to step S302, and steps S404 to S412 correspond to step S303.

[0031] (i) Step S401 The test order generation unit 1012 receives, for learning, patient information 201, time-series data of test items and test result information 202, and time-series data of patient disease information 203. As learning data, patient information, time-series data of test items and test result information, and time-series data of patient disease information of a plurality of patients prepared in advance are used.

[0032] (ii) Step S402: The test order generation unit 1012 creates a time gap mask and a test item mask and applies the masks to the input data. Then, it learns and extracts features of the test value data to generate feature time-series data. Here, the time gap mask is information that fills the time gap between each test date and time. For example, to fill the date gap between the test on February 1, 2018 and the test on February 18, 2018, it is possible to insert a "0" between the two data sets as mask information, or to use the number of days between the two sets as mask information. Furthermore, as the test item mask, for example, a "0" is inserted into the index of an item that has not been tested (secondary test), and a "1" is inserted into the index of an item that has been tested.

[0033] (iii) Step S403: The inspection order generation unit 1012 performs time-series learning on the feature time-series data to generate time-series features. As described above, the learning is processing using a neural network (e.g., repetitive processing using LSTM (Long Short Term Memory)), and time-series features (LSTM time-series features) are generated.

[0034] (iv) Step S404: The inspection order generation unit 1012 estimates the probability distribution (random variable 9 of the distribution) of the inspection results for each inspection item. Specifically, a calculation is performed to estimate the probability distribution of each element of the time-series feature quantities generated in step S403, and as a result, the mean μ and variance ν of each feature quantity are calculated.

[0035] (v) Step S405 The examination order generation unit 1012 restores the original input data (secondary examination result data including missing values) from the mean μ and variance v (estimated probability distribution) of each feature calculated in step S404.

[0036] (vi) Step S406 The test order generation unit 1012 estimates the test value (information amount according to disease type (probability of being able to identify a disease by each test item)) 206 based on the restored test results. The test value estimation (information amount estimation) is performed for each time step (e.g., test date). Here, an example of calculation is shown when the time step = S. Using the distribution parameters (μ, ν) obtained from the learning model (the encoder shown in Figure 5 below), the information amount of candidate item C is estimated by Monte Carlo sampling. This sampling is performed M times, and the average value (expected value E) is taken as the information amount. Furthermore, the i-th estimation is as follows:

[0037] The probability distribution parameters of the test results are defined as follows:

[0038] The information amount (estimated inspection value) obtained by the i-th estimation can be calculated according to the following information amount estimation formula.

[0039]

[0040] Here, the Kullback-Leibler Divergence (D KL ) is defined by the following formula:

[0041] (vii) Step S407 The inspection order generating unit 1012 compares the estimated inspection value calculated in step S406 with the correct data of the inspection value, and calculates (estimates) the learning error.

[0042] (viii) Step S408 The test order generation unit 1012 predicts the patient's disease based on the reconstructed test results acquired in step S405.

[0043] Here, the probability distribution parameters of the disease information are defined as follows:

[0044] Similarly to the estimation of the information amount (test value), the prediction of the patient's disease can also be calculated according to the information amount estimation formula.

[0045] (ix) Step S409 The test order generation unit 1012 compares the predicted value of the patient's disease calculated in step S407 with the correct data of the patient's disease, and calculates (estimates) the learning error.

[0046] (x) Step S410 The test order generation unit 1012 compares each of the learning errors calculated in steps S407 and S409 with a preset threshold (different thresholds may be set for the test value and the disease prediction value) and determines whether the learning errors are equal to or less than the threshold (converged). If each learning error is equal to or less than the threshold (Yes in step S410), the process proceeds to step S411. On the other hand, if each learning error is greater than the threshold (No in step S410), the process proceeds to step S412.

[0047] (xi) Step S411: The test order generation unit 1012 determines the learning parameters when the learning error is equal to or less than the threshold as the parameters of the trained model. This completes the learning process for the learning model of test value and patient disease prediction, and the trained model is generated. The parameters included in the trained model include parameters for generating time-series feature quantities from feature quantity time-series data and parameters for estimating a probability distribution from the time-series feature quantities and generating restored data.

[0048] (xii) Step S412: The inspection order generation unit 1012 updates the learning parameters according to a predetermined rule or randomly, and proceeds to step S404. In this case, the processing from step S404 is repeated using the updated learning parameters.

[0049] <Example of Architecture Configuration of Learning Process> FIG. 5 is a diagram showing an example of architecture configuration of the learning process used in the learning process of FIGS. 3 and 4. In FIG.

[0050] The architecture of the learning process can be composed of, for example, a feature extraction learning unit 501 that performs the processing of step S301, a time series learning unit 502 that performs the processing of step S302, and a learning unit 503 that performs the processing of step S303 to restore time series data of test results and predict patient diseases.

[0051] (i) The feature extraction learning unit 501 includes a first connection unit 5012 that performs weighted addition of a plurality of processing target data sequences (inputs) and outputs a vector E, and a second connection unit 5013 that performs weighted addition of E[k]*M[k] (k=1, 2, ..., N) obtained by adding an input missing mask M to each component E[k] (k=1, 2, ..., N) of the vector E, and outputs time-series data (S, E ２ ) and a second connection unit 5014 that outputs the signal.

[0052] Each input data string to be processed includes a time gap mask (size (S, N)) that fills the time interval between each test result, input data X (size (S, N): for example, composed of the primary test result data for each test date and the secondary test result including missing parts as shown in FIG. 6), and learnable variables (N, E １ ) where X[;,k] * F[k] (k = 1, 2, ..., N) represents the feature extraction operation, and the learnable variables (N, E １ ) and b[k] (k=1, 2, . . . , N) which denotes the operation of concatenating bias values.

[0053] The input missing mask M[k] is a mask assigned to the test items. For example, in FIG. 6 , "0" is assigned to the index of an unperformed test item, and "1" is assigned to the index of a performed test item. The input missing mask M is assigned because the number of test items varies from patient to patient, and the number of meaningful inputs varies from patient to patient. Therefore, even if the number of input test items varies, it is necessary to generate features of the same length by time-series learning.

[0054] (ii) The time-series learning unit (size (S, R)) 502 includes a plurality of time-series learning units 5021_1 to 5021_n, and outputs time-series feature quantities.

[0055] Each time series learning unit 5021_k (k = 1, 2, ..., R) can be configured with an LSTM (Long Short Term Memory) or a GRU (Gated Recurrent Unit), which is a simplified model of an LSTM. As a result, the output from the time series learning unit 502 becomes a time series feature of the LSTM.

[0056] (iii) The restoration learning unit for time-series data of test results and the learning unit for patient disease prediction 503 includes an encoder 5031 and a decoder 5032.

[0057] The encoder 5031 is configured, for example, by a variational autoencoder (VAE), and performs a calculation to estimate the probability distribution of each element of the time-series feature from the time-series learning unit 502, and calculates the latent variable mean value μ and latent variable variance ν of each feature (estimation of probability distribution).

[0058] The decoder 5032 reconstructs the original input data from the probability distribution (μ and ν) and predicts the disease of the target patient, and outputs the reconstructed data and the disease prediction result 5033.

[0059] <Example of Test Value Restoration> FIG. 6 is a diagram showing an example of restoring test value data including missing values ​​(defective values).

[0060] As shown in FIG. 6 , when inspection value data 601 including the primary inspection results and the secondary inspection results is given, the feature extraction learning unit 501 performs feature extraction processing (using, for example, a convolutional neural network) on input data 602 corresponding to the plurality of data strings to be processed, and generates time-series data of the features.

[0061] The time series learning unit 502 performs time series learning on the time series data of the feature amounts, and generates time series feature amounts 603 .

[0062] The test result time-series data restoration learning unit and patient disease prediction learning unit 503 estimate the probability distribution of each element of the time-series feature 603 and, based on this, restore the test values ​​of the secondary test results, including missing values. As shown in Figure 6, in the test value restoration 604, the numerical values ​​of the primary test results remain unchanged, but the restored values ​​of the secondary test results are supplemented with the missing values ​​of the original secondary test results. Based on this restored test values, the test information volume (test value) and estimated test order are generated.

[0063] 7 is a flowchart outlining the process of predicting a patient's disease and generating a test order by applying the patient information and test results of a patient (new patient) to a trained model. When the patient information and time-series data of test items and test result information of the patient are applied to the trained model, the processes of steps S701 to S706 are executed.

[0064] (i) Step S701 The patient information acquisition unit 1011 acquires patient information 201 of a suspected patient (e.g., a new patient), time-series data 202 of test items and test result information of tests performed on the suspected patient, and the trained model 700. Then, the test order generation unit 1012 uses the acquired patient information 201 and time-series data 202 of test items and test result information to extract time-series data of the feature quantities of the test results.

[0065] (ii) Step S702 The inspection order generation unit 1012 applies the time-series data of the feature quantities to the trained model 700 to calculate (estimate) the time-series feature quantities.

[0066] (iii) Step S703 The examination order generation unit 1012 restores (estimates) time-series data of the examination results (secondary examination results) based on the time-series feature amounts calculated in step S702.

[0067] (iv) Step S704 The examination order generation unit 1012 estimates the examination value of each examination item at each time step (each examination date and time) from the restored data of the examination results estimated in step S703.

[0068] (v) Step S705 The test order generation unit 1012 generates a disease prediction result for the patient based on the test value calculated in step S704 and the guideline information acquired from the medical information server 111.

[0069] (vi) Step S706: The inspection order generation unit 1012 refers to the guideline information and extracts candidate inspection items corresponding to the inspection value calculated in step S704. The inspection order output unit 1013 then acquires information on the candidate inspection items from the inspection order generation unit 1012 and displays it on the display screen (not shown) of the information terminal 110.

[0070] <Details of inference process using trained model> Fig. 8 is a flowchart for explaining the details of the process of predicting a patient's disease and generating a test order by applying the patient information and test results of a patient to be estimated (new patient) to a trained model. In Fig. 8, as in Fig. 7, the processes of steps S802 to S806 are executed by applying the data acquired in step S801 to the trained model.

[0071] (i) Step S801 Step S801 corresponds to a part (first half) of the processing of step S701 in Fig. 7. The patient information acquisition unit 1011 receives patient information 201 of a patient to be estimated (e.g., a new patient), time-series data 202 of test items and test result information of tests performed on the patient to be estimated, and a trained model 700.

[0072] (ii) Step S802 Step S802 corresponds to part (the latter half) of step S701. The examination order generation unit 1012 creates the time gap mask and examination item mask described above, integrates the acquired patient information 201 and the time-series data 202 of examination items and examination result information (corresponding to input data 602 in FIG. 6 ), and extracts feature quantities of the patient information. Specifically, the examination order generation unit 1012 converts the received patient information 201 and the time-series data 202 of examination items and examination result information into text and digitizes the text so that it can be processed by a neural network (patient information integration), and assigns the time gap mask and examination item mask to the data. Then, the examination order generation unit 1012 calculates time-series data of feature quantities of the numerical information based on the integrated patient information (with mask) and the trained model.

[0073] (iii) Step S803 Step S803 corresponds to the process of step S702. The examination order generation unit 1012 calculates (estimates) time-series feature amounts from the time-series data of the feature amounts calculated in step S802.

[0074] (iv) Step S804 Step S804 corresponds to the processing of step S703. The examination order generation unit 1012 restores (estimates) time-series data of the examination results (secondary examination results) from the time-series feature amounts calculated in step S803.

[0075] (v) Step S805 Step S805 corresponds to the processing of step S704. The test order generation unit 1012 estimates the test value of each test item at each time step (each test date and time) from the restored data of the test results estimated in step S804. The test value is a probability value that indicates how much the probability of diagnosing a disease can be improved if the corresponding test item candidate is hypothetically selected. The test value (information amount) can be calculated by applying the information of the probability distribution estimated in step S804 to the above-mentioned information amount estimation formula.

[0076] (vi) Step 806 Step S806 corresponds to the processing of step S705. The test order generation unit 1012 references the guideline information 1113 acquired from the medical information server 111 to acquire information on the disease corresponding to the test value calculated in step S805. The test order generation unit 1012 outputs information on the disease corresponding to the test value showing the highest value as the disease prediction result for the predicted patient. Here, the guideline information 1113 is information including information on test items determined by medical symptoms (e.g., test X is performed in the case of symptoms A and B) and information on the diseases corresponding to each test item (e.g., test X1 is performed to identify disease D1, test X2 is performed to identify disease D2, etc.).

[0077] (vii) Step S807 Step S807 corresponds to the processing of step S705. The inspection order generation unit 1012 refers to the guideline information and extracts candidate inspection items corresponding to the inspection values ​​calculated in step S805. At this time, inspection items (secondary inspection items) corresponding to inspection values ​​with high numerical values ​​may be generated as high-priority candidate inspection items.

[0078] The test order output unit 1013 then acquires information on the candidate test items from the test order generation unit 1012 and displays it on the display screen (not shown) of the information terminal 110. When the doctor checks the candidate test items displayed on the display screen and selects the necessary test items from the candidates, the test order output unit 1013 transmits the test order to the test ordering system 1114. The test ordering system 1114 transmits the received test order to a terminal (not shown) in the examination room.

[0079] <Example of Test Values> Fig. 9 is a diagram showing a specific example of test value data 900. The test value data 900 represents raw data of actual test results.

[0080] The test value data 900 includes, as components, a patient ID 901 that uniquely identifies a patient, patient attribute information 902, date and time 903, each test item and its result 904, and disease information 905. The patient attribute information 902 includes information on the gender and age of the patient corresponding to the patient ID 901. The date and time 903 is information indicating the date and time when the test was performed.

[0081] Each test item and its result 904 is information indicating the test item and its result carried out at the corresponding date and time 903. In each test item and its result 904, the test value and the normal / abnormal judgment result are entered for the test item that has been carried out, and the symbol "-" is entered for the test item that has not been carried out. If the test value is too high or too low, it is judged to be abnormal.

[0082] Disease information 905 indicates information about a disease that has been provisionally diagnosed by a doctor after looking at the actual test results.

[0083] 10 is a diagram showing an example of the configuration of a screen display (test order GUI) 1000 that outputs patient diseases predicted by a trained model and recommended test items. The screen display (test order GUI) 1000 is displayed on the display screen of an information terminal 110 operated by a doctor or the like.

[0084] The screen display (GUI for test orders) 1000 includes, as its constituent items, a patient ID display section 1001, a current date display section 1002, a disease probability etc. display section 1003, a disease test item display section 1004, a recommended test item display section 1005, a pathological progression analysis execution button 1006, and a test order output button 1007.

[0085] The disease probability display unit 1003 displays the prediction results of the patient to be estimated using the trained model, for example, the probability and risk for each disease.

[0086] The disease probability display section 1003 provides information on the disease name, disease probability, and disease risk (the degree of risk of the disease itself) obtained through testing and disease prediction processing (prediction simulation). For example, it shows that disease A obtained as a result of the prediction processing is correct with a 70% probability, and that the risk of the disease is medium. It also shows that disease C is correct with a 2% probability, and that the risk of the disease is high.

[0087] The disease test item display section 1004 displays, for example, test items for diseases with a disease probability higher than a predetermined value or test items for high-risk diseases, patient costs indicating the costs paid by the patient for the test, and hospital costs indicating the costs borne by the hospital. In the example of Figure 10, test items for disease A, which has a probability of 70%, and disease C, which has a high risk, are displayed. The test items for disease A, which has the highest probability, are tests a, c, and f, and the test items for disease C, which has a high risk, are tests d, g, and q. In addition, the patient costs are higher for disease C than for disease A, but the hospital costs are shown to be 55 for disease A and 0 for disease C.

[0088] The recommended test item display section 1005 displays recommended test items obtained by comprehensively judging the probability and risk of a disease. For example, among the test items for disease A and disease C, test items having a test value (information amount) equal to or greater than a predetermined threshold are displayed. As of May 15, 2018, all of the test items for disease A (tests a, c, and f) have a test value equal to or greater than the threshold, while all of the test items for disease C (tests d, g, and q) have a test value less than the threshold. Therefore, only tests a, c, and f are displayed in the recommended test item display section 1005.

[0089] When the pathological condition transition analysis execution button 1006 is pressed, the pathological condition transition analysis is executed, and a time-series pathological condition transition analysis result display 1100 shown in FIG. 11 is displayed on the display screen of the information terminal 110.

[0090] When the test order output button 1007 is pressed, information on the test items listed as recommended test items (tests a, c, and f) is sent as a test order from the doctor's information terminal 110 to a terminal in the examination room (not shown).

[0091] <Display of Analysis Results of Time-Series Pathological Transition> FIG. 11 is a diagram showing an example of the configuration of a screen (display of analysis results of time-series pathological transition 1100) that displays the results of analysis of time-series pathological transition.

[0092] The time-series pathological condition transition analysis result display 1100 includes, as components, a patient ID display section 1101, a current date display section 1102, a disease-related secondary test information amount display section 1103, and a recommended test item transition display section 1104.

[0093] The disease-related secondary test information amount display section 1103 displays a graph showing the time-dependent change in the calculated test value (the amount of information obtained by the above-described inference process) for each secondary test item for determining each disease (e.g., diseases A and C) displayed in the disease test item display section 1004 of Fig. 10. According to Fig. 11, the amount of information (test value) of each test item related to disease A (tests a, c, and f) increases over time, while the amount of information (test value) of each test item related to disease C (tests d, g, and q) decreases over time. Therefore, the disease-related secondary test information amount display section 1103 shows that, currently (May 15, 2018), the test items for disease A are more valuable to perform than the test items for disease C.

[0094] The recommended test item trend display section 1104 displays the time-dependent trend in the probability of each disease (e.g., diseases A and C) displayed in the disease test item display section 1004 of FIG. 10 and the time-dependent trend in the recommended test items displayed in the recommended test item display section 1005 of FIG. 10 . This allows the user to check how appropriate test items have changed over time. For example, as of February 1, 2018, tests a, c, and f for disease A and test d for disease C have test values ​​(information content) above a predetermined threshold and are therefore listed as recommended test items. However, as of February 18, 2018, only tests a, c, and f for disease A have test values ​​above the predetermined threshold and are listed as recommended test items.

[0095] Figure 11 allows doctors and other medical professionals to check the changes over time in the test value (information volume) of each test item for each disease and the changes over time in recommended test items, allowing them to be more confident about the test items that should be performed.

[0096] <Other> The functions of the present embodiment can also be realized by software program code. In this case, a storage medium on which the program code is recorded is provided to a system or device, and the computer (or CPU or MPU) of the system or device reads the program code stored in the storage medium. In this case, the program code read from the storage medium itself realizes the functions of the above-described embodiment, and the program code itself and the storage medium on which it is stored constitute the present disclosure. Examples of storage media for providing such program code include flexible disks, CD-ROMs, DVD-ROMs, hard disks, optical disks, magneto-optical disks, CD-Rs, magnetic tape, non-volatile memory cards, and ROMs.

[0097] In addition, an operating system (OS) running on a computer may perform some or all of the actual processing based on instructions in the program code, and the functions of the above-described embodiments may be realized by this processing. Furthermore, after the program code is read from a storage medium and written to memory on a computer, a CPU of the computer may perform some or all of the actual processing based on instructions in the program code, and the functions of the above-described embodiments may be realized by this processing.

[0098] Furthermore, the program code of the software that realizes the functions of this embodiment may be distributed via a network and stored in a storage means such as a hard disk or memory of the system or device, or in a storage medium such as a CD-RW or CD-R, so that when in use, the computer (or CPU or MPU) of the system or device reads and executes the program code stored in the storage means or storage medium.

[0099] The processes and techniques described herein are not inherently related to any specific device and can be implemented by a combination of components. Various types of general-purpose devices can also be added. A dedicated device may be constructed to perform the functions of this embodiment. Various functions can also be formed by appropriately combining multiple components disclosed in this embodiment. For example, some components may be omitted from all the components shown in the embodiments and examples, or components from different examples may be appropriately combined.

[0100] Although specific embodiments are described in this disclosure, they are in all respects for the purpose of explanation (understanding the technology of the present disclosure) and not for the purpose of limitation. Those skilled in the art will recognize that there are many combinations of hardware, software, and firmware suitable for implementing the technology of the present disclosure. For example, the software described can be implemented in a wide variety of programming or scripting languages, such as assembler, C / C++, Perl, Shell, PHP, Java (registered trademark), etc.

[0101] Furthermore, in the above-described embodiment, the control lines and information lines are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. All components may be interconnected.

[0102] In addition, other implementations of the present disclosure will be apparent to those skilled in the art from consideration of the present embodiments and examples. The specification and examples are exemplary only, with the scope and spirit of the present disclosure being indicated by the following claims.

[0103] REFERENCE SIGNS LIST 10 Information processing system (diagnosis support system) 100 Computer 101 Processor 102 Main memory device 103 Sub-memory device 104 Network adapter 105 Input device 106 Output device 109 Network 110 Information terminal 111 External recording device (medical information server) 1011 Patient information acquisition unit 1012 Examination order generation unit 1013 Examination order output unit 1114 Examination ordering system

Claims

1. A diagnostic support system that generates a trained model that estimates the test value of at least one test item included in a secondary test from the test results and patient information of multiple patients prepared for learning, the system comprising: a storage device that stores a program for generating the trained model; and a computer that generates the trained model by reading and executing the program from the storage device, wherein the computer performs: a data acquisition process that acquires the patient information of the multiple patients, time series data of test items for multiple tests on the multiple patients and their test results, and time series data of disease information of the multiple patients; a feature extraction process that extracts time series data of at least the feature amounts of the test results using the time series data of the test items and their test results and pre-stored test item information; a time series learning process that performs time series learning on the time series data of the feature amounts extracted by the feature extraction process and outputs time series feature amounts; a probability distribution estimation process that estimates a probability distribution of the time series feature amounts obtained by the time series learning process; and a test result restoration learning process that restores the results of the secondary test at each point in the time series based on the probability distribution. and a process of determining parameters for estimating the test value of the test item in the trained model based on the results of the secondary test at each time point in the time series restored by the test result restoration learning process.

2. A diagnostic support system according to claim 1, wherein the computer, in the data acquisition process, further acquires information regarding the time intervals between the plurality of tests in a time series; and in the feature extraction process, creates a time gap mask that fills in the time intervals between the plurality of tests in the time series data of the test results, and, if there is a missing value in the result of the secondary test included in the test results, creates a test item mask for the test item with the missing value; and applies the time series data of the test results to which the time gap mask and the test item mask have been added to a neural network, thereby extracting time series data of the feature amounts of the test results.

3. A diagnostic support system as claimed in claim 1, wherein the computer, in the probability distribution estimation process, calculates the mean and variance of the latent variables that make up the probability distribution by performing multiple LSTM operations on the time series features.

4. A diagnostic support system as claimed in claim 1, wherein the computer further executes a process of determining parameters for predicting patient diseases in the trained model based on the results of the secondary test at each point in the time series restored by the test result restoration learning process and time series data of disease information of the multiple patients.

5. A diagnostic support system that estimates the test value of each test item included in a secondary test by applying patient information of a patient to be estimated, the test items of a primary test administered to the patient to be estimated, and time-series data of the test results to a trained model, the system comprising: a storage device that stores a program for estimating the test value of each test item included in the secondary test; and a computer that estimates the test value by reading and executing the program from the storage device, wherein the computer performs a data acquisition process that acquires the trained model, the patient information of the patient to be estimated, the test items of a plurality of primary tests of the patient to be estimated and time-series data of the test results, and time-series data of disease information of the patient to be estimated; and a process that estimates the test value of each test item included in the secondary test by applying the patient information, the test items of the plurality of primary tests, and time-series data of the test results of the plurality of primary tests to the trained model, The trained model is a diagnostic support system that (i) extracts time series data of features of the test results from the patient information and time series data of the test items of the multiple primary tests and the test results of the multiple primary tests, (ii) performs time series learning on the extracted time series data of features, the primary test results included in the multiple primary tests, and the patient information, and outputs time series features, (iii) estimates a probability distribution of the time series features, (iv) reconstructs the results of the secondary tests at each time point in the time series based on the probability distribution, and (v) estimates the test value of the test item based on the reconstructed results of the secondary tests at each time point in the time series.

6. A diagnostic support system as set forth in claim 5, wherein the computer, in the data acquisition process, further acquires information regarding the time intervals between the multiple primary examinations in a time series; and, in the process of estimating the test value, further applies the information regarding the time intervals between the multiple primary examinations to the trained model; and the trained model creates a time gap mask that fills in the time intervals between the multiple primary examinations in the time series data of the test results, and an examination item mask that fills in the test items of the results of the secondary examination that have missing values, and applies the time series data of the test results to which the time gap mask and the examination item mask have been added to a neural network, thereby extracting time series data of the feature quantities of the test results.

7. A diagnostic support system as claimed in claim 5, wherein the trained model, when estimating the probability distribution, calculates the mean and variance of the latent variables that constitute the probability distribution by performing multiple LSTM operations on the time series features.

8. A diagnostic support system as claimed in claim 5, wherein the trained model further includes parameters for predicting a patient's disease based on the results of the secondary examination at each time point in the restored time series and time series data of disease information of the patient to be estimated, and the computer further executes a process for predicting the disease of the patient to be estimated.

9. A diagnostic support system according to claim 5, further comprising the computer executing a process of generating and outputting an order for the secondary examination based on the estimated examination value.

10. A diagnostic support method for generating a trained model that estimates the test value of at least one test item included in a secondary test from the test results and patient information of multiple patients prepared for learning, the method comprising: a computer that generates the trained model by reading and executing a program from a storage device that stores the program for generating the trained model, acquiring patient information of the multiple patients, time series data of test items and test results of multiple tests for the multiple patients, and time series data of disease information of the multiple patients; the computer extracts time series data of at least the feature quantities of the test results using the time series data of the test items and test results of the multiple tests and pre-stored test item information; the computer performs time series learning on the extracted time series data of the feature quantities and outputs the time series feature quantities; the computer estimates a probability distribution of the time series feature quantities; and the computer restores the results of the secondary test at each time point in the time series based on the probability distribution. and determining, by the computer, parameters for estimating the test value of the test item in the trained model based on the results of the secondary test at each time point in the restored time series.

11. A diagnostic support method as claimed in claim 10, further comprising the computer determining parameters for predicting patient diseases in the trained model based on the results of the secondary examination at each time point in the restored time series and time series data of disease information of the plurality of patients.

12. A diagnostic support method using a computer to estimate the test value of each test item included in a secondary test by applying patient information of a patient to be estimated, the test items of a primary test administered to the patient to be estimated, and time-series data of the test results of the primary test to a trained model, the method comprising: the computer reading a program for estimating the test value of each test item included in the secondary test from a storage device that stores the program; the computer acquiring the trained model, the patient information, the test items of a plurality of primary tests administered to the patient to be estimated and time-series data of the test results, and time-series data of disease information of the patient to be estimated; and the computer applying the patient information, the test items of the plurality of primary tests, and time-series data of the test results of the plurality of primary tests to the trained model to estimate the test value of each test item included in the secondary test. The trained model is a model that (i) extracts time series data of features of the test results from the patient information and time series data of the test items of the multiple primary tests and the test results of the multiple primary tests, (ii) performs time series learning on the extracted time series data of features, the test results of the multiple primary tests, and the patient information, and outputs time series features, (iii) estimates a probability distribution of the time series features, (iv) reconstructs the results of the secondary tests at each time point in the time series based on the probability distribution, and (v) estimates the test value of the test item based on the reconstructed results of the secondary tests at each time point in the time series.

13. A diagnostic support method as claimed in claim 12, wherein the trained model further includes parameters for predicting a patient's disease based on the results of the secondary examination at each time point in the restored time series and time series data of disease information of the patient to be estimated, and the diagnostic support method further includes the computer predicting the disease of the patient to be estimated.

14. A diagnostic support method according to claim 12, further comprising the computer generating and outputting an order for the secondary examination based on the estimated examination value.

Images