Medical information processing apparatus, medical information processing method, and program

Abstract

To reduce a risk posed by medical treatment.SOLUTION: A medical information processing apparatus of an embodiment has an acquisition unit, a calculation unit, and a generation unit. The acquisition unit acquires a standard prediction model that predicts a specific risk posed by a standard medical treatment. The calculation unit calculates the similarity between the standard medical treatment and target medical treatment for which the specific risk is predicted. The generation unit generates a target prediction model that predicts the specific risk in the target medical treatment by using the standard prediction model and the similarity.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The embodiments disclosed in this specification and the drawings relate to a medical information processing device, a medical information processing method, and a program. [Background technology]

[0002] Some anticancer drugs used in cancer treatment are cardiotoxic, and if cardiovascular disease develops, cancer treatment must be interrupted. Therefore, cardiovascular disease may occur as a complication of anticancer drug treatment, and it is important to understand the risk of developing cardiovascular disease.

[0003] Conventionally, machine learning models have been developed to predict cardiovascular disease events in anticancer drug patients in order to understand the risk of cardiovascular disease before anticancer drug treatment. This machine learning model predicts cardiovascular disease events using electronic medical record data. However, this machine learning model cannot be applied to patients who are treated with anticancer drugs that are not included in the learning data, because the mechanism and risk of cardiotoxicity differ depending on the type and ingredients of the anticancer drug. Therefore, this learning model cannot be applied to patients who are administered anticancer drugs for rare diseases with a small number of patients or new anticancer drugs.

[0004] Furthermore, there are technologies that use anticancer drugs with fewer side effects, assuming that similar anticancer drugs have similar properties, to extract side effects specific to the target drug that are not found in similar drugs from the side effect descriptions. However, while these technologies can evaluate the similarity of anticancer drugs, they are limited to extracting drugs with similar properties and custom-made side effects, so it is not easy to avoid risks. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2012-158609 A [Patent Document 2] JP 2011-159078 A Summary of the Invention [Problem to be solved by the invention]

[0006] The problem to be solved by the embodiments disclosed in this specification and the drawings is to suppress the risks associated with treatment. However, the problem to be solved by the embodiments disclosed in this specification and the drawings is not limited to the above problem. Problems corresponding to the effects of each configuration shown in the embodiments described below can also be positioned as other problems. [Means for solving the problem]

[0007] The medical image processing device of the embodiment has an acquisition unit, a calculation unit, and a generation unit. The acquisition unit acquires a reference prediction model that predicts a specific risk due to a reference treatment. The calculation unit calculates a similarity between a target treatment that is to be predicted for the specific risk and the reference treatment. The generation unit generates a target prediction model that predicts the specific risk in the target treatment by using the reference prediction model and the similarity. [Brief description of the drawings]

[0008] [Figure 1] 1 is a block diagram showing an example of the configuration of a hospital system 1 according to a first embodiment. [Diagram 2] FIG. 1 is a block diagram showing an example of the configuration of a medical information processing apparatus 100 according to a first embodiment. [Diagram 3] FIG. 4 is a diagram showing an example of the contents of a medicine DB 151. [Figure 4] FIG. 13 is a diagram showing an example of the contents of a prediction model DB 152. [Diagram 5] 4 is a flowchart showing an example of processing in the medical information processing apparatus 100 of the first embodiment. [Figure 6] 4 is a flowchart showing an example of processing in the medical information processing apparatus 100 of the first embodiment. [Figure 7] A diagram showing an example of a visualization of the procedure for calculating the similarity between a known drug and a target drug using the “description of side effects.” [Figure 8] A diagram showing an example of a visualization of the procedure for extracting similar drugs to a target. [Figure 9] FIG. 4 is a diagram showing an example of the content displayed on a display 130 according to the first embodiment. [Figure 10] FIG. 11 is a diagram showing an example of the content displayed on a display 130 according to a second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, a medical image processing apparatus, a medical image processing method, and a program according to an embodiment will be described with reference to the drawings.

[0010] The medical information processing device generates a prediction model that predicts the risk of a patient undergoing treatment. The medical information processing device updates a ready-made prediction model to generate a prediction model, or generates a new prediction model. In the following description, the ready-made prediction model is referred to as a reference prediction model, and the newly generated or updated prediction model is referred to as a target prediction model.

[0011] (First embodiment) The prediction model in the medical information processing device of the first embodiment is a model that predicts the incidence of complications, such as cardiovascular disease, when a drug therapy (anticancer drug therapy) is performed in which a drug (anticancer drug) is administered as a cancer treatment. The medical information processing device further predicts the incidence of complications when a patient undergoes drug therapy using the prediction model. The prediction model is created, for example, for each drug administered during drug therapy.

[0012] 1 is a block diagram showing an example of the configuration of an in-hospital system 1 according to the first embodiment. The in-hospital system 1 according to the first embodiment includes, for example, a hospital information system (hereinafter, HIS) 10, a radiology information system (hereinafter, RIS) 20, a medical image diagnostic apparatus (modality) 30, a picture archiving and communication system (PACS) 40, and a medical information processing apparatus 100.

[0013] The HIS 10 is a computer system that supports operations within a hospital. Specifically, the HIS 10 has various subsystems, such as an electronic medical record system, a medical accounting system, a medical appointment system, a hospital reception system, and an admission and discharge management system.

[0014] The HIS 10 is a computer system that supports operations within a hospital. Specifically, the HIS 10 has various subsystems, such as an electronic medical record system, a medical accounting system, a medical appointment system, a hospital reception system, and an admission and discharge management system.

[0015] The HIS 10 includes a computer such as a server device or a client terminal that includes a processor such as a CPU (Central Processing Unit), memories such as a ROM (Read Only Memory) and a RAM (Random Access Memory), a display, an input interface, and a communication interface.

[0016] A user inputs and references information about a patient using an electronic medical record system included in the HIS 10. A user issues an order for an imaging examination to the HIS 10. The HIS 10 transfers order information corresponding to the imaging examination order to another system such as the RIS 20.

[0017] The RIS 20 is a computer system that supports operations in the imaging diagnostic department. The RIS 20 manages reservations for imaging examination orders in cooperation with the HIS 10, links reservation information to examination equipment, manages examination information, etc. The RIS 20 includes a computer such as a server device or a client terminal that includes, for example, a processor such as a CPU, memories such as ROM and RAM, a display, an input interface, and a communication interface.

[0018] The modality 30 performs imaging (photographing) according to imaging conditions (photographing protocol) determined based on, for example, an image examination instruction. Examples of the modality 30 include an X-ray computed tomography apparatus, an X-ray diagnostic apparatus, a magnetic resonance imaging apparatus, an ultrasonic diagnostic apparatus, and a nuclear medicine diagnostic apparatus. The modality 30 is operated by an operator such as a doctor (radiologist) or a clinical radiologist. Medical images (image data) generated by imaging by the modality 30 are transmitted to the PACS 40.

[0019] The PACS 40 is a computer system that receives medical images transmitted by the modality 30 and stores them in a database. The PACS 40 transmits (transfers) the medical images stored in the database in response to a request from a client. The PACS 40 includes a server computer that includes a processor such as a CPU, memories such as ROM and RAM, a display, an input interface, and a communication interface.

[0020] The configuration of the hospital system 1 is not limited to the above. The hospital system 1 may include, for example, an image interpretation report creation device. In addition, some elements of the hospital system 1 may be integrated. For example, the HIS 10 and the RIS 20 may be integrated into one system.

[0021] The medical information processing device 100 generates a prediction model for predicting the risk of developing complications when administering a drug therapy, and predicts the risk of developing complications in a patient administering a drug therapy using the prediction model. When generating a prediction model, the medical information processing device 100 updates a reference prediction model corresponding to a known drug, and generates a prediction model corresponding to a new drug. The medical information processing device 100 predicts risks using a ready-made prediction model, newly generates a target prediction model corresponding to a new drug, and predicts risks when administering a drug therapy in which a new drug is administered using the generated target prediction model.

[0022] 2 is a block diagram showing an example of the configuration of the medical information processing device 100 of the first embodiment. The medical information processing device 100 includes, for example, a communication interface 110, an input interface 120, a display 130, a processing circuit 140, and a memory 150. The communication interface 110, the input interface 120, and the display 130 in the medical information processing device 100 are provided separately from the communication interface, the input interface, and the display provided in the HIS 10, but these may be common.

[0023] The communication interface 110 communicates with external devices such as the RIS 20, the modality 30, and the PACS 40 via a network NW such as a LAN (Local Area Network). The communication interface 110 includes a communication interface such as a NIC (Network Interface Card). The network NW may include the Internet, a cellular network, a Wi-Fi network, a WAN (Wide Area Network), etc. instead of or in addition to the LAN.

[0024] The input interface 120 accepts various input operations from a user such as a doctor, converts the accepted input operations into electrical signals, and outputs the electrical signals to the processing circuitry 140. For example, when an input operation is performed by a user, the input interface 120 generates information corresponding to the input operation. The input interface 120 outputs the generated information corresponding to the input operation to the processing circuitry 140.

[0025] The input interface 120 includes, for example, a mouse, a keyboard, a trackball, a switch, a button, a joystick, a touch panel, etc. The input interface 120 may be, for example, a user interface that accepts voice input from a microphone, etc. When the input interface 120 is a touch panel, the input interface 120 may also have the display function of the display 130.

[0026] In this specification, the input interface is not limited to an interface having physical operation parts such as a mouse, a keyboard, etc. For example, an example of the input interface also 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 device and outputs the electrical signal to a control circuit.

[0027] The display 130 displays various types of information. For example, the display 130 displays an image generated by the processing circuit 140, a GUI (Graphical User Interface) for receiving various input operations from an operator, and the like. For example, the display 130 is an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube) display, an organic EL (Electro Luminescence) display, or the like. The display 130 is an example of a display unit.

[0028] The processing circuitry 140 includes, for example, an acquisition function 141, a calculation function 142, a generation function 143, a prediction function 144, and a display control function 145. The processing circuitry 140 realizes these functions by, for example, a hardware processor (computer) executing a program stored in a memory (storage circuitry) 150.

[0029] The hardware processor refers to circuitry such as a CPU, a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a programmable logic device (e.g., a simple programmable logic device (SPLD) or a complex programmable logic device (CPLD)), and a field programmable gate array (FPGA).

[0030] Instead of storing the program in the memory 150, the program may be directly embedded in the circuit of the hardware processor. In this case, the hardware processor realizes the function by reading and executing the program embedded in the circuit. The above program may be stored in the memory 150 in advance, or may be stored in a non-transitory storage medium such as a DVD or CD-ROM, and installed in the memory 150 from the non-transitory storage medium by mounting the non-transitory storage medium in a drive device (not shown) of the medical information processing device 100.

[0031] The hardware processor is not limited to being configured as a single circuit, but may be configured as a single hardware processor by combining multiple independent circuits to realize each function. Also, multiple components may be integrated into a single hardware processor to realize each function. The hardware processor, memory, etc. in the medical information processing device 100 are provided separately from the hardware processor, memory, etc. of the HIS 10, but these may be common.

[0032] The memory 150 stores, for example, a drug database (hereinafter, referred to as DB) 151 and a prediction model DB 152. FIG. 3 is a diagram showing an example of the contents of the drug DB 151. The drug DB 151 is a known DB of drugs that are administered in cancer treatment. For each known drug, the drug DB 151 includes an explanation of side effects, a composition formula and a structural formula (at least one of the composition formula and the structural formula), and information on vectors calculated from these. When a new drug (new drug) for cancer treatment is developed, the new drug is sequentially added to the drug DB 151 as a known drug.

[0033] 4 is a diagram showing an example of the contents of the prediction model DB 152. The prediction model DB 152 is a DB of prediction models corresponding to drugs. The prediction model is a model that predicts a specific risk, such as the onset of a complication, due to a treatment, for example, a treatment in which a drug is administered. The prediction model is, for example, a trained model that has been subjected to machine learning using training data in which the input data is a drug and the output data is the presence or absence of the onset of a complication.

[0034] The prediction model includes, for example, an input layer that inputs input data, an output layer that outputs an output unit, and an intermediate layer between the input layer and the output layer. The intermediate layer has, for example, a multi-layered neural network that connects the input layer and the output layer. As the trained model, for example, a transformer, BERT, TF-IDF (Term Frequency - Inverse Document Frequency), etc. may be used. The prediction model may be a model that predicts the onset of complications based on a rule base.

[0035] The prediction model is, for example, a trained model created using training data in which test data, medication data, and patient data are input data, and the presence or absence of complications is output data. The trained model is, for example, a trained model in which test data, medication data, and patient data are input data, and the rate of complications developing is output data. The test data is, for example, data from tests performed on a patient before drug treatment, and tests measuring height, weight, blood pressure, heart rate, respiratory rate, and the like. The medication data is data such as the drugs administered to the patient as drug treatment and the drug administration method. The patient data is data such as the patient's age and sex. The output data is, for example, displayed on the display 130 and visualized.

[0036] The acquisition function 141 acquires information for identifying a drug (hereinafter, a target drug) for which a target prediction model is to be generated. The target drug may be a known drug or a new drug. The acquisition function 141 reads out and acquires a prediction model (hereinafter, a reference prediction model) corresponding to a drug similar to the target drug (hereinafter, a similar drug) from the prediction model DB 152. A similar drug is, for example, a known drug whose similarity calculated by the calculation function 142 described below exceeds a predetermined threshold. The acquisition function 141 is an example of an acquisition unit.

[0037] The acquisition function 141 acquires medical data of a patient when predicting the risk of complications when the patient undergoes drug treatment. The medical data of a patient includes, for example, test data, drug data, and patient data when the patient has undergone drug treatment in the past. The acquisition function 141 acquires various data included in the medical data.

[0038] The calculation function 142 calculates the similarity between a target treatment to be predicted for a specific risk and a reference treatment. For example, the calculation function 142 calculates the similarity between a known drug administered to a patient in a drug treatment of a prediction model included in the prediction model DB 152 and a target drug administered to a patient in a drug treatment of a target prediction model.

[0039] In the first embodiment, a drug therapy in which a similar drug is administered is the reference therapy, and the similarity between the target drug and the similar drug means the similarity between the target therapy and the reference therapy. The calculation function 142 calculates the similarity between the known drug and the target drug, for example, based on the description of the side effects of the drug and the chemical formula, for example, the composition formula or the structural formula. The calculation function 142 may calculate the similarity based on either the description of the side effects of the drug or the chemical formula.

[0040] The calculation function 142 calculates the similarity between the descriptions of the side effects of the known drug and the target drug, for example, using a natural language model. The calculation function 142 converts, for example, the chemical formula of the drug into a vector and calculates the similarity between the known drug and the target drug. The calculation function 142 calculates the similarity between the known drug and the target drug, for example, using a similarity calculation method using Euclidean distance or cosine similarity. The similarity calculation method may be a method other than the method using Euclidean distance or cosine similarity. The calculation function 142 is an example of a calculation unit.

[0041] The generation function 143 generates a target prediction model using the reference prediction model acquired by the acquisition function 141 and the similarity between the similar drug and the target drug calculated by the calculation function 142. When the acquisition function 141 acquires multiple reference prediction models, the generation function 143 generates a target prediction model by weighting each of the multiple reference prediction models according to the similarity. The generation function 143 weights the reference prediction model more heavily as the similarity between the similar drug and the target drug increases. The generation function 143 is an example of a generation unit.

[0042] The prediction function 144 predicts a specific risk using the target prediction model generated by the generation function 143. For example, the prediction function 144 predicts the risk of developing a complication when a drug treatment is performed by administering a target drug to a patient using the target prediction model. The prediction function 144 is an example of a prediction unit.

[0043] The display control function 145 displays the risk of developing a complication predicted by the target prediction model on the display 130. When displaying the risk of developing a complication, the display control function 145 displays the progress of treatment, images obtained by examinations, names of drugs administered to the patient in drug therapy, and the like on the display 130. The display control function 145 is an example of a display control unit.

[0044] Next, the processing in the medical information processing device 100 will be described. The medical information processing device generates a target prediction model when drug therapy is performed, and executes risk prediction using the prediction model when drug therapy is performed. Each process will be described below. Figures 5 and 6 are flowcharts showing an example of the processing in the medical information processing device 100 of the first embodiment.

[0045] First, a procedure for generating a target prediction model will be described with reference to Fig. 5. For example, when a drug therapy is performed in which a known drug or a new drug is administered, the medical information processing device 100 generates a prediction model with the administered drug as a target drug. The generation of the prediction model is executed after the drug therapy is performed in which a known drug or a new drug is administered, and after the presence or absence of complications due to the drug therapy is determined.

[0046] First, the medical information processing device 100 acquires information for identifying a target drug administered in a drug treatment in the acquisition function 141, and identifies the target drug (step S101). Next, the calculation function 142 calculates the similarity between the target drug acquired by the acquisition function 141 and a drug stored in the drug DB 151 using the document of the "description of side effects" and the "composition formula / structural formula" (step S103).

[0047] When the calculation function 142 calculates the similarity between a known drug and a target drug from a document of "description of side effects", for example, first, the calculation function 142 converts the "description of side effects" of each of the target drug and the drug into vectors (latent variables) using a natural language model such as BERT or GPT. Then, the calculation function 142 calculates the similarity between the known drug and the target drug by comparing the vectors of the documents of "description of side effects" of each of the known drug and the target drug obtained by the conversion.

[0048] When calculating the degree of similarity between a known drug and a target drug from the "compositional formula / structural formula", the calculation function 142 first converts the "compositional formula / structural formula" of the target drug and the drug into vectors using a model for predicting protein structure such as Alpha Fold or a model for predicting efficacy and side effects of a drug. Next, the calculation function 142 calculates the degree of similarity between the known drug and the target drug by comparing the vectors of the documents of the "compositional formula / structural formula" of the known drug and the target drug obtained by the conversion.

[0049] FIG. 7 is a diagram showing an example of visualization of the procedure for calculating the similarity between a known drug and a target drug using "description of side effects." For example, assume that the target drug is a "new drug A" that is not used in the reference prediction model, and known drugs include "Adriamycin," "Cisplatin," and "Carboplatin." In this case, the calculation function 142 converts the document "documentation of side effects" of "new drug A" "myocardial damage, heart failure (frequency unknown for both)" into a vector x∈R D Convert to.

[0050] In addition, the calculation function 142 converts the documents of "Adriamycin's side effects" such as "myocardial damage, heart failure (frequency unknown)" into a vector x1∈R D In addition, the calculation function 142 converts the documents of the "side effects documents" of "cisplatin" such as "myocardial infarction, angina pectoris, congestive heart failure, and arrhythmia" into a vector x2∈R D In addition, the calculation function 142 converts the document of the "side effect document" of "carboplatin" "myocardial infarction, congestive heart failure" into a vector x3∈R D Convert to.

[0051] The calculation function 142 calculates a vector x ∈ R of “new drug A”. D and the vector x1∈R of "Adriasin" D The calculation function 142 compares the vector x ∈ R D and the vector of "cisplatin" x2∈R D The calculation function 142 compares the vector x ∈ R D and the vector x3∈R for "carboplatin" D The similarity S3 between "new drug A" and "carboplatin" is calculated by comparing the above.

[0052] The calculation function 142 calculates the similarity between the known drug and the target drug using Euclidean distance or cosine similarity. The calculation function 142 may calculate the similarity between the known drug and the target drug using a method other than the calculation method using Euclidean distance or cosine similarity. The Euclidean distance is calculated, for example, by the following formula (1), and the cosine similarity is calculated by the following formula (2).

number

[0053] When the target drug identified by the acquisition function 141 is a new drug, the calculation function 142 always calculates the similarity between the new drug and a known drug stored in the DB 151. The calculation function 142 may store the calculated similarity in the memory 150. In this case, when the target drug identified by the acquisition function 141 is a known drug, the calculation function 142 may read out the similarity stored in the memory 150 to obtain the similarity between the known drug and the target drug.

[0054] Next, the acquisition function 141 reads out from the prediction model DB 152 a prediction model corresponding to a similar drug whose similarity to the target drug calculated by the calculation function 142 exceeds a predetermined threshold, and acquires it as a reference prediction model (step S105). Next, the generation function 143 determines whether the target drug identified by the acquisition function 141 is a new drug (step S107).

[0055] When it is determined that the target drug identified by the acquisition function 141 is a new drug, the generation function 143 generates a target prediction model (step S109). In generating the target prediction model, the generation function 143 extracts similar drugs whose similarity to the target drug calculated by the calculation function exceeds a threshold, selects a prediction model corresponding to the similar drug from the prediction model DB 152, and reads it from the memory 150 as a reference prediction model.

[0056] FIG. 8 is a diagram showing an example of visualization of the procedure for extracting similar drugs to a target. In the example shown in FIG. 8, the target drug is new drug A. Known drugs include "Adriamycin," "Cisplatin," and "Carboplatin." The similarities to new drug A are similarity S1 for "Adriamycin," S2 for "Cisplatin," and S3 for "Carboplatin." The similarities are real numbers. Here, the generation function 143 reads out the prediction models for "Adriamycin" and "Carboplatin" as reference prediction models.

[0057] The generation function 143 generates a target prediction model by using the reference prediction model read from the memory 150 (step S109). Since the target drug is a new drug and there is no corresponding prediction model, the generation function 143 generates a new target prediction model instead of updating an existing prediction model.

[0058] When the number of extracted similar drugs is one (1), the generation function 143 uses the extracted reference prediction model as a target prediction model. When the number of extracted similar drugs is multiple (2 or more), the generation function 143 generates a target prediction model by weighting according to the similarity between the target drug and the similar drugs.

[0059] For example, when the similarity S1 of "Adriamycin" and the similarity S3 of "Carboplatin" exceed a threshold, the first prediction model W1 of "Adriamycin" and the third prediction model W3 of "Carboplatin" become the reference prediction models. The generation function 143 generates a target prediction model based on the first prediction model W1, the third prediction model W3, and the treatment data of the drug therapy in which the new drug A was administered.

[0060] Target prediction model W t In generating the similarity S1 of “Adriamycin” and the similarity S3 of “Carboplatin”, the generation function 143 uses the similarity S1 of “Adriamycin” and the similarity S3 of “Carboplatin” to generate the initial value W of the target prediction model weighted by the similarity according to the following formula (3): t=0 Note that the denominator in equation (3) represents a normalization term.

number

[0061] The medical information processing device 100 may generate a target prediction model with the new drug as the target drug in the generation function 143 before the drug therapy in which the new drug is administered is performed. In this case, the medical information processing device 100 performs the processes of steps S101 to S109 described above, and in the process, the generation function 143 may generate an initial value W t=0 may be the target prediction model.

[0062] After the initial value of the target prediction model for the new drug is generated, the target prediction model is generated based on the treatment data of the drug treatment in which the new drug A is administered, using the following formula (4). In the following formula (4), the loss coefficient L is, for example, cross entropy.

number

[0063] In step S107, when it is determined that the identified target drug is not a new drug (is a known drug), the generation function 143 reads out a reference prediction model corresponding to the known drug that is the target drug as a target prediction model. Next, the generation function 143 updates the reference prediction model to a target prediction model based on the treatment data of the drug therapy in which the identified known drug was administered (step S111). After that, the generation function 143 stores the generated target prediction model in the memory 150 (step S113). In this way, the medical information processing device 100 ends the process shown in FIG. 5.

[0064] Next, the processing in the medical information processing device 100 will be described. The medical information processing device generates a target prediction model when drug therapy is performed, and executes risk prediction using the prediction model when drug therapy is performed. Each process will be described below. Figures 5 and 6 are flowcharts showing an example of the processing in the medical information processing device 100 of the first embodiment.

[0065] Next, a procedure for predicting risks using a prediction model when administering a drug treatment will be described with reference to Fig. 6. The medical information processing device 100 first acquires, in the acquisition function 141, medical data of a patient who will start administering a drug treatment and information for identifying the drug to be administered (step S201). Next, the prediction function 144 identifies the drug to be administered to the patient, extracts a prediction model corresponding to the identified drug from among the prediction models included in the prediction model DB 152, and reads it out from the memory 150 as a target prediction model (step S203).

[0066] Next, the prediction function 144 inputs the acquired medical data as input data into the read target prediction model. A prediction result predicting the onset of a complication, which is a specific risk, is output from the target prediction model to which the input data has been input (step S205). The prediction function 144 acquires the prediction result output from the target prediction model. Next, the display control function 145 causes the display 130 to display the prediction result acquired by the prediction function 144 (step S207). In this way, the medical information processing device 100 ends the process shown in FIG. 6.

[0067] Next, information displayed on the display 130 by the display control function 145 will be described. Fig. 9 is a diagram showing an example of the contents displayed on the display 130 of the first embodiment. The display 130 is divided into, for example, a patient information display area GA11, a first test content display area GA21, a second test content display area GA22, a third test content display area GA23, and a test result display area GA31. Here, the display contents when a drug therapy is performed by administering a new drug (new medicine) A are shown.

[0068] The patient information display area GA11 displays the patient ID. The patient ID is a unique ID given to each patient. The first examination content display area GA21 displays the patient's blood pressure over time, the second examination content display area GA22 displays the patient's X-ray photograph, and the third examination content display area GA23 displays the doctor's findings.

[0069] The test result display area GA31 displays the progress of the prediction result of the risk of developing cardiovascular disease (cardiovascular disease risk) predicted using the target prediction model, the name of the drug used in the anticancer drug treatment, and the similarity to new drug A. The display of the similarity of new drug A to new drug A is omitted. A user such as a doctor can, for example, formulate or modify a treatment plan while looking at this information displayed on the display 130.

[0070] In the first embodiment, the medical information processing device 100 uses a prediction model corresponding to a similar drug similar to the target drug as a reference prediction model to generate a target prediction model corresponding to the target drug. Therefore, since the target prediction model can be generated or updated using past data, the accuracy of prediction of the rate of onset of complications can be improved. As a result, the risk of treatment can be suppressed. In addition, since the prediction model corresponding to a similar drug is used as the reference prediction model, the onset rate of complications can be predicted even if the target drug is a new drug that does not have a corresponding prediction model.

[0071] Second embodiment Next, a second embodiment will be described. In the first embodiment, the medical information processing device 100 predicts the incidence rate of complications. In contrast, in the second embodiment, the medical information processing device 100 has the same overall configuration as the first embodiment, but differs from the first embodiment in that the prediction function 144 predicts the incidence rate of perioperative complications. Hereinafter, the medical information processing device 100 of the second embodiment will be described, focusing on the differences from the first embodiment.

[0072] The prediction model in the medical information processing device of the second embodiment (hereinafter, surgical operation prediction model) is a model that predicts the incidence rate of perioperative complications in surgical operations performed as cancer treatment, for example, cardio-cerebrovascular complications (cardiovascular complications) such as myocardial infarction and cerebral infarction. The surgical operation prediction model is, for example, a trained model having an input layer, an output layer, and an intermediate layer, which is machine-learned by using medical data from a surgical operation performed on a patient as input data and the presence or absence of the onset of a perioperative complication as output data.

[0073] The surgical procedure prediction model is a model that is provided in correspondence with a surgical procedure, and outputs a prediction result of the incidence rate of perioperative complications as output data by inputting medical data. The output data is visualized, for example, by being displayed on the display 130. The surgical procedure prediction model may be a model that predicts the incidence of perioperative complications based on a rule base.

[0074] The medical data serving as input data includes, for example, surgical operation data in addition to the patient data similar to that of the first embodiment. The surgical operation data includes information related to the surgical operation, such as the surgical procedure, the surgeon, the operation time, and the facility where the operation was performed. The medical data may include test data and medication data similar to those of the first embodiment. Similarly, the medical data of the first embodiment may include surgical operation data.

[0075] In the medical information processing device of the second embodiment, the drug in the first embodiment is replaced with a surgical operation, and the generation function 143 updates a known surgical operation prediction model in a similar procedure to that of the first embodiment. In addition, when a new surgical operation is developed and a corresponding surgical operation prediction model is not included in the prediction model DB 152, the generation function 143 replaces the new drug in the first embodiment with the new surgical operation and generates a surgical operation prediction model in a similar procedure.

[0076] The medical information processing device of the second embodiment further predicts the incidence of perioperative complications when a patient undergoes a surgical treatment by using a prediction model. The prediction model is created for each surgical treatment, for example. In the second embodiment, a surgical treatment for a similar surgical treatment is a reference treatment, and a surgical treatment for a target surgical treatment is a target treatment.

[0077] In the medical information processing device of the second embodiment, the prediction function 144 predicts the incidence rate of perioperative complications by inputting medical data into a surgical operation prediction model. The display control function 145, for example, causes the display 130 to display output data including the incidence rate of perioperative complications predicted by the prediction function 144.

[0078] The prediction function 144 can predict the incidence of perioperative complications at multiple time points during the perioperative period. For example, the prediction function 144 may be capable of calculating the incidence of perioperative complications within a specific period after a surgical operation before the surgical operation, or may be capable of calculating the incidence of perioperative complications within a specific period after the surgical operation.

[0079] 10 is a diagram showing an example of contents displayed on the display 130 of the second embodiment. The display 130 of the second embodiment is partitioned into a patient information display area GA11, a first examination content display area GA21, a second examination content display area GA22, a third examination content display area GA23, and an examination result display area GA31, similar to the first embodiment. The patient information display area GA11, the first examination content display area GA21, the second examination content display area GA22, and the third examination content display area GA23 each display information similar to that of the first embodiment.

[0080] The examination result display area GA31 displays the progress of the prediction result of the cardiovascular disease risk predicted using the target prediction model, the name of the similar surgical operation, and the similarity to the latest surgical operation. The display of the similarity of the latest surgical operation to the latest surgical operation is omitted. A user such as a doctor can, for example, set or modify a treatment plan while looking at this information displayed on the display 130.

[0081] The predicted results of cardiovascular disease risk displayed in the test result display area GA31 are, for example, displayed separately for pre- and post-surgery based on the time of surgery. For example, the predicted results of pre-surgery cardiovascular disease risk show the incidence rate of cardiovascular disease within a specific period after the completion of the surgical operation, predicted at each stage before the surgical operation, in the pre-surgery column, and show the incidence rate of cardiovascular disease within a specific period after the surgical operation, in the post-surgery column. As the cardiovascular disease risk, the change in the incidence rate of cardiovascular disease over time is displayed.

[0082] The medical information processing device of the second embodiment uses a prediction model corresponding to a similar surgical procedure as a reference prediction model to generate a target prediction model corresponding to a target surgical procedure such as the latest surgical procedure. Therefore, since the target prediction model can be generated or updated using past data, the accuracy of predicting the rate of onset of complications can be improved. As a result, the same effect as that of the first embodiment can be obtained, and the risk of treatment can be reduced. In addition, since a prediction model corresponding to a known surgical procedure is used as a reference prediction model, the rate of onset of complications can be predicted even for the latest surgical procedure for which there is no corresponding prediction model.

[0083] According to at least one of the embodiments described above, the medical information processing device can reduce risks due to treatment by having an acquisition unit that acquires a reference prediction model that predicts a specific risk due to a reference treatment, a calculation unit that calculates the similarity between a target treatment to be predicted for the specific risk and the reference treatment, and a generation unit that generates a target prediction model that predicts the specific risk in the target treatment using the reference prediction model and the similarity.

[0084] Although some embodiments 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, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included in the scope of the invention and its equivalents as described in the claims, as well as in the scope and spirit of the invention. [Explanation of symbols]

[0085] 1. In-hospital system 10 Hospital Information System (HIS) 20 Radiology Information System (RIS) 30 Medical imaging diagnostic equipment (modality) 40 Picture Archiving and Communication System (PACS) 100 Medical information processing device 110 Communication Interface 120 Input Interface 130 Display 140 Processing circuit 141 Acquisition Function 142 Calculation Function 143 Generation function 144 Predictive Function 145 Display control function 150 Memory 151 Drug DB 152 Prediction Model DB GA11 Patient information display area GA21 1st inspection content display area GA22 Second inspection content display area GA23 3rd inspection content display area GA31 Test result display area NW Network

Claims

1. an acquisition unit for acquiring a reference prediction model for predicting a specific risk due to a reference treatment; A calculation unit that calculates a similarity between a target treatment that is a prediction target of the specific risk and the reference treatment; A generation unit that generates a target prediction model for predicting the specific risk in the target treatment by using the reference prediction model and the similarity. Medical information processing equipment.

2. The acquisition unit acquires a plurality of the reference prediction models, The generation unit generates the target prediction model by weighting each of the plurality of reference prediction models according to the similarity. The medical information processing device according to claim 1 .

3. The generation unit weights the reference prediction model more heavily as the similarity increases. The medical information processing device according to claim 2 .

4. The target treatment and the reference treatment include medication; The medical information processing device according to claim 1 .

5. The drug administered by the drug therapy includes a new drug not used in the reference prediction model; The medical information processing device according to claim 4 .

6. The calculation unit calculates the similarity based on at least one of an explanation of a side effect of the drug or a chemical formula. The medical information processing device according to claim 5 .

7. The calculation unit calculates a similarity of an explanation of a side effect of the drug by using a natural language model. The medical information processing device according to claim 6 .

8. The calculation unit converts the chemical formula of the drug into a vector and calculates a similarity of the drug administered by the drug therapy. The medical information processing device according to claim 6 .

9. The target treatment and the reference treatment include surgical treatment; The medical information processing device according to claim 1 .

10. The reference prediction model is a model for predicting at least one of the risk of an incidence rate of perioperative complications within a specific period after a surgical operation in a stage before the surgical operation included in the surgical treatment, or an incidence rate of perioperative complications within a specific period after the surgical operation is completed. The medical information processing device according to claim 9 .

11. The calculation unit calculates the similarity using a similarity calculation method. The medical information processing device according to claim 1 .

12. Further comprising a prediction unit for predicting the specific risk using the target prediction model; The medical information processing device according to claim 1 .

13. Further comprising a display control unit that displays the specific risk predicted by the target prediction model on a display unit. The medical information processing device according to claim 1 .

14. The computer Obtain a baseline prediction model that predicts a specific risk with a baseline treatment; Calculating a similarity between the target treatment for which the specific risk is predicted and the reference treatment; generating a subject prediction model utilizing the reference prediction model and the similarity measure to predict the particular risk of the subject treatment; Medical information processing method.

15. On the computer, Obtain a baseline prediction model that predicts a specific risk with a baseline treatment; Calculating a similarity between the target treatment for which the specific risk is predicted and the reference treatment; generating a subject prediction model utilizing the reference prediction model and the similarity measure to predict the particular risk of the subject treatment; program.

Images