Medical information processing system, medical information processing method and storage medium

Abstract

This invention relates to a medical information processing system, a medical information processing method, and a storage medium, enabling patient diagnosis and treatment regardless of the instability of responses to medical inquiries. The medical information processing system of one embodiment includes an acquisition unit, an inference unit, and an output control unit. The acquisition unit acquires examination data representing the results of medical examinations of a patient and response data representing the results of responses to medical inquiries. The inference unit infers information related to the patient's diagnosis and treatment by inputting the examination data of the patient into a first model. Furthermore, the inference unit infers information related to the patient's diagnosis and treatment by inputting the examination data and the response data of the patient into a second model. The output control unit outputs a first inference result of the first model and a second inference result of the second model via an output unit.

Description

Technical Field

[0001] The embodiments disclosed in this specification and accompanying drawings relate to medical information processing systems, medical information processing methods, and storage media. This application claims priority to and incorporates the contents of Japanese Patent Application No. 2021-131303 filed on August 11, 2021, and Japanese Patent Application No. 2022-122133 filed on July 29, 2022. Background Technology

[0002] Patient responses to medical history checks are crucial for understanding their condition, but they are also highly variable, often fluctuating due to the patient's own feelings and emotions. Currently, healthcare professionals utilize and understand these fluctuations to aid in diagnosis and treatment. On the other hand, research is underway to automate these processes using artificial intelligence (AI). However, the instability of patient responses can sometimes negatively impact AI-based diagnostic and treatment methods.

[0003] Patent Document 1: Japanese Patent Publication No. 2006-511880 Summary of the Invention

[0004] The problem to be solved by the embodiments disclosed in this specification and accompanying drawings is to enable patient diagnosis and treatment without being affected by the instability of responses to medical history taking. However, the problem to be solved by the embodiments disclosed in this specification and accompanying drawings is not limited to the above-described problem. Problems corresponding to the effects of the various configurations shown in the embodiments described below can also be identified as other problems.

[0005] The medical information processing system of this embodiment includes an acquisition unit, an inference unit, and an output control unit. The acquisition unit acquires examination data representing the results of medical examinations conducted on a patient, and response data representing the results of interviews conducted on the patient. The inference unit infers treatment-related information by inputting the patient's examination data into a first model. The first model is a model learned based on a first training dataset, which uses treatment-related information of the patient as labels for correct answers and establishes a correspondence between the patient's examination data and the first training dataset. Furthermore, the inference unit infers treatment-related information by inputting the patient's examination data and the response data into a second model. The second model is a model learned based on a second training dataset, which uses treatment-related information of the patient as labels for correct answers and establishes a correspondence between the patient's examination data and the response data. The output control unit outputs a first inference result representing information related to the diagnosis and treatment inferred using the first model, and a second inference result representing information related to the diagnosis and treatment inferred using the second model. Attached Figure Description

[0006] Figure 1 This is a diagram illustrating a configuration example of the medical information processing system according to the first embodiment.

[0007] Figure 2 This is a diagram showing an example of the configuration of the user interface in the first embodiment.

[0008] Figure 3 This is a diagram illustrating an example of the configuration of the medical information processing device according to the first embodiment.

[0009] Figure 4 This is a flowchart illustrating a series of processing steps of the processing circuit in the first embodiment.

[0010] Figure 5 It is a graph used to illustrate the data in the response.

[0011] Figure 6 This is a diagram representing an example of the first model.

[0012] Figure 7 This is a diagram representing an example of the second model.

[0013] Figure 8 This is an example of a picture showing the screen on a monitor.

[0014] Figure 9 This is a diagram showing the comparison of inference results.

[0015] Figure 10 These are diagrams showing other examples of what a monitor displays.

[0016] Figure 11 It is a schematic diagram representing the weighting of the response data.

[0017] Figure 12 These are diagrams showing other examples of what a monitor displays.

[0018] Explanation of symbols

[0019] 1: Medical information processing system; 10: User interface; 11: Communication interface; 12: Input interface; 13: Output interface; 14: Memory; 20: Processing circuit; 21: Acquisition function; 22: Output control function; 23: Communication control function; 100: Medical information processing device; 111: Communication interface; 112: Input interface; 113: Output interface; 114: Memory; 120: Processing circuit; 121: Acquisition function; 122: Inference function; 123: Judgment function; 124: Output control function; 125: Communication control function. Detailed Implementation

[0020] Hereinafter, the medical information processing system, medical information processing method, and storage medium according to the embodiments will be described with reference to the accompanying drawings.

[0021] (First Embodiment)

[0022] [Composition of a Medical Information Processing System]

[0023] Figure 1 This diagram illustrates a configuration example of the medical information processing system 1 according to the first embodiment. The medical information processing system 1 includes, for example, a user interface 10 and a medical information processing device 100. The user interface 10 and the medical information processing device 100 are communicatively connected via a communication network NW.

[0024] A communication network (NW) can be an entire information communication network that utilizes electrical communication technology. For example, in addition to wireless / wired LANs such as hospital backbone LANs (Local Area Networks) and the Internet, a communication network (NW) also includes telephone communication line networks, fiber optic communication networks, cable communication networks, and satellite communication networks.

[0025] User interface 10 is for use by patients and healthcare professionals. For example, user interface 10 is a touch interface, a voice user interface, or more specifically, a terminal device such as a personal computer, tablet, or mobile phone. Healthcare professionals are typically doctors, but can also be nurses or other personnel involved in the diagnosis and treatment. For instance, patients may input their answers to a medical consultation into user interface 10 via touch or voice. Healthcare professionals may also conduct a verbal consultation with a patient, listen to their answers, and input the results into user interface 10.

[0026] In this embodiment, "diagnosis and treatment" includes not only treatments such as surgery and medication, but also examinations up to or after treatment, and all other medical procedures.

[0027] User interface 10 sends information entered by patients or medical staff to medical information processing device 100 via communication network NW, or receives information from medical information processing device 100.

[0028] The medical information processing device 100 receives information from the user interface 10 via a communication network NW and processes the received information. Then, the medical information processing device 100 sends the processed information to the user interface 10 via the communication network NW. Alternatively, the medical information processing device 100 may send the processed information to the user interface 10 or, instead, to a dedicated terminal for medical personnel located within the hospital.

[0029] The medical information processing device 100 can be a single device or a system of multiple devices connected via a communication network NW that operate collaboratively. That is, the medical information processing device 100 can also be implemented using multiple computers (processors) included in a distributed computing system or a cloud computing system. The medical information processing device 100 does not necessarily need to be a separate device from the user interface 10; it can also be an integrated device with the user interface 10.

[0030] [Composition of the terminal device]

[0031] Figure 2 This is a diagram illustrating an example configuration of the user interface 10 according to the first embodiment. The user interface 10 includes, for example, a communication interface 11, an input interface 12, an output interface 13, a memory 14, and a processing circuit 20.

[0032] The communication interface 11 communicates with the medical information processing device 100 and the like via the communication network NW. The communication interface 11 includes, for example, a NIC (Network Interface Card) and an antenna for wireless communication.

[0033] Input interface 12 accepts various input operations from the operator (e.g., a patient) and converts the received input operations into electrical signals, which are then output to processing circuit 20. For example, input interface 12 may include a mouse, keyboard, trackball, switch, button, joystick, touch panel, etc. Input interface 12 may also be a user interface that accepts sound input, such as a microphone. When input interface 12 is a touch panel, it may also also have the display function of the display 13a included in output interface 13, which will be described later.

[0034] In this specification, the input interface 12 is not limited to having physical operating components such as a mouse and keyboard. For example, a signal processing circuit that receives electrical signals corresponding to input operations from an external input device that is separate from the device and outputs such electrical signals to the control circuit is also included in the example of the input interface 12.

[0035] The output interface 13 may include, for example, a display 13a and a speaker 13b.

[0036] Display 13a displays various information. For example, display 13a displays images generated by processing circuit 20, a GUI (Graphical User Interface) for accepting various input operations from the operator, etc. For example, display 13a is an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube) display, an organic EL (ElectroLuminescence) display, etc.

[0037] The speaker 13b outputs sound as the information input from the processing circuit 20.

[0038] The memory 14 may be implemented using semiconductor storage elements such as RAM (Random Access Memory), flash memory, hard disks, or optical disks. These non-transitory storage media may also be implemented using other storage devices connected via a communication network (NW), such as NAS (Network Attached Storage) or external storage server devices. The memory 14 may also include non-transitory storage media such as ROM (Read Only Memory) and registers.

[0039] The processing circuit 20 may include, for example, an acquisition function 21, an output control function 22, and a communication control function 23. The processing circuit 20 may implement these functions, for example, by executing a program stored in the memory 14 (memory circuit) through a hardware processor (computer).

[0040] The hardware processor in processing circuit 20 refers to circuits such as CPU (Central Processing Unit), GPU (Graphics Processing Unit), Application Specific Integrated Circuit (ASIC), and programmable logic device (e.g., Simple Programmable Logic Device (SPLD), Complex Programmable Logic Device (CPLD), Field Programmable Gate Array (FPGA)). Alternatively, instead of storing the program in memory 14, the program can be directly programmed into the hardware processor's circuitry. In this case, the hardware processor implements its functions by reading and executing the program programmed into the circuitry. The program can be pre-stored in memory 14 or stored on a non-transitory storage medium such as a DVD or CD-ROM, and then installed from the non-transitory storage medium into memory 14 via a drive device (not shown) that mounts the non-transitory storage medium onto the user interface 10. The hardware processor is not limited to being a single circuit; multiple independent circuits can be combined to form a single hardware processor and implement various functions. In addition, multiple components can be integrated into a single hardware processor to achieve various functions.

[0041] The acquisition function 21 acquires input information via the input interface 12, or acquires information from the medical information processing device 100 via the communication interface 11.

[0042] Output control function 22 causes display 13a to display information obtained by acquisition function 21, or outputs information obtained by acquisition function 21 from speaker 13b.

[0043] The communication control function 23 sends the information input to the input interface 12 to the medical information processing device 100 via the communication interface 11.

[0044] [Composition of Medical Information Processing Devices]

[0045] Figure 3 This is a diagram illustrating a configuration example of the medical information processing device 100 according to the first embodiment. The medical information processing device 100 includes, for example, a communication interface 111, an input interface 112, an output interface 113, a memory 114, and a processing circuit 120.

[0046] The communication interface 111 communicates with the user interface 10, etc., via the communication network NW. The communication interface 111 includes, for example, a NIC. The communication interface 111 is an example of an "output unit".

[0047] Input interface 112 accepts various input operations from the operator and converts the received input operations into electrical signals, which are then output to processing circuit 120. For example, input interface 112 includes a mouse, keyboard, trackball, switch, button, joystick, touch panel, etc. Input interface 112 can also be a user interface that accepts sound input, such as a microphone. When input interface 112 is a touch panel, it can also also function as the display of the monitor 113a included in output interface 113 (described later).

[0048] In this specification, the input interface 112 is not limited to having physical operating components such as a mouse and keyboard. For example, a signal processing circuit that receives electrical signals corresponding to input operations from an external input device that is separate from the device and outputs such electrical signals to the control circuit is also included in the example of the input interface 112.

[0049] The output interface 113 may include, for example, a display 113a and a speaker 113b. The output interface 113 is another example of an "output unit".

[0050] Display 113a displays various information. For example, display 113a displays images generated by processing circuit 120, GUIs for accepting various input operations from the operator, etc. For example, display 113a is an LCD, CRT display, organic EL display, etc.

[0051] The speaker 113b outputs sound as the information input from the processing circuit 120.

[0052] The memory 114 may be implemented using semiconductor storage elements such as RAM and flash memory, hard disks, and optical disks. These non-transitory storage media may also be implemented using other storage devices connected via a communication network (NW), such as NAS or external storage server devices. The memory 114 may also include non-transitory storage media such as ROM and registers.

[0053] In addition to storing the program executed by the hardware processor, memory 114 also stores model information. This model information (program or data structure) defines the first model MDL1 and the second model MDL2. The first model MDL1 and the second model MDL2 can be implemented, for example, using DNNs (Deep Neural Networks) such as CNNs (Convolutional Neural Networks). The first model MDL1 and the second model MDL2 are not limited to DNNs; they can also be implemented using other models such as support vector machines, decision trees, Naive Bayes classifiers, and random forests. Details regarding the first model MDL1 and the second model MDL2 will be described later.

[0054] When both Model 1 (MDL1) and Model 2 (MDL2) are implemented using a DNN, the model information includes, for example, information about the combination of units within the input layer, one or more hidden layers (intermediate layers), and the output layer, as well as weighting information such as the number of combination coefficients assigned to the input and output data between the combined units. The combination information may include, for example, the number of units in each layer, the type of unit to which each unit is combined, the activation function implementing each unit, and the gates set between units in the hidden layers. The activation function implementing the unit may be, for example, ReLU (Rectified Linear Unit), ELU (Exponential Linear Units), a pruning function, a sigmoid function, a step function, a higher-order tangent function, or an identity function. The gates, for example, selectively pass through or weight the data passed between units based on the value returned by the activation function (e.g., 1 or 0). Associative coefficients can include, for example, the weights assigned to output data in the hidden layers of a neural network when data is output from a unit in one layer to a unit in a deeper layer. Associative coefficients can also include inherent bias components of each layer.

[0055] The processing circuit 120 includes, for example, an acquisition function 121, an inference function 122, a determination function 123, an output control function 124, and a communication control function 125. The acquisition function 121 is an example of an "acquisition unit," the inference function 122 is an example of an "inference unit," and the determination function 123 is an example of a "determination unit." The output control function 124 is an example of an "output control unit," and the communication control function 125 is another example of an "output control unit."

[0056] The processing circuit 120 performs these functions, for example, by executing a program stored in the memory 114 (storage circuit) through a hardware processor (computer).

[0057] The hardware processor in processing circuit 120 refers to a circuit such as a CPU, GPU, application-specific integrated circuit (ASIC), or programmable logic device (e.g., simple or complex programmable logic device, field-programmable gate array). Alternatively, instead of storing the program in memory 114, the program can be directly programmed into the hardware processor's circuitry. In this case, the hardware processor performs its function by reading and executing the program programmed into the circuitry. The program can be pre-stored in memory 114 or stored on a non-transitory storage medium such as a DVD or CD-ROM, and then installed from the non-transitory storage medium into memory 114 by installing the non-transitory storage medium into the drive device (not shown) of the medical information processing device 100. The hardware processor is not limited to being a single circuit; multiple independent circuits can be combined to form a single hardware processor and implement various functions. Alternatively, multiple components can be integrated into a single hardware processor to implement various functions.

[0058] [Processing flow of medical information processing devices]

[0059] The following describes a series of processes performed by the processing circuit 120 of the medical information processing device 100 according to the flowchart. Figure 4 This is a flowchart illustrating a series of processing steps performed by the processing circuit 120 in the first embodiment.

[0060] First, the acquisition function 121 acquires the patient's (hereinafter referred to as the patient) response data to the consultation from the user interface 10 via the communication interface 111, and acquires the patient's examination data from the examination device (step S100).

[0061] Examination equipment refers to devices used for medical examinations of patients, such as X-ray CT (Computed Tomography) devices, MRI (Magnetic Resonance Imaging) devices, mammography devices, ultrasound imaging diagnostic devices, nuclear medicine diagnostic devices, body fluid analysis devices, and devices for measuring vital signs. Examination data is quantitative digital data obtained by measuring the patient's physical information using these various examination devices. On the other hand, response data is qualitative digital data that includes the patient's subjective information. In other words, function 121 acquires both quantitative and qualitative data related to the patient.

[0062] Figure 5This is a diagram used to illustrate the data provided in the answers. For example, when a patient visits a medical institution, they are asked to fill out a questionnaire like the one shown in the diagram, including necessary information such as their current physical and mental state, medical history, and any allergies (R1 in the diagram). The questionnaire can be displayed on the monitor 13a of the user interface 10, or it can be printed on paper and distributed to the patient. When the patient fills out an answer on the paper questionnaire, the medical staff at the medical institution can input the information into the user interface 10. At this time, OCR (Optical Character Recognition / Reader) can also be used. The questions on the questionnaire can be output as sound from the speaker 113b of the user interface 10, or they can be read aloud by the medical staff at the medical institution. When the patient answers aloud towards the user interface 10, the user interface 10 can also acquire the patient's aloud answer via a microphone. Alternatively, or based on this, the medical staff can also listen to the patient's aloud answer. When the medical staff hears the answer from the patient, they can also input the listening result into the user interface 10. Unlike a medical history form, questions don't need to be predetermined; instead, medical staff can freely decide the content of the consultation during the examination. The responses to these questions are medically termed the chief complaint. Therefore, the data from these responses can also be referred to as chief complaint data.

[0063] Return to Figure 4 The flowchart is explained. When the acquisition function 121 acquires the response data and examination data for the consultation, the inference function 122 inputs the examination data of the patient acquired by the acquisition function 121 into the first model MDL1 defined by the model information stored in the memory 141 (step S102).

[0064] Figure 6 This diagram illustrates an example of Model 1, MDL1. Model 1 is a machine learning model that learns by using a dataset as training data. This dataset maps information related to the patient's medical history as labels (also called targets) to the patient's examination data. In other words, Model 1 learns by outputting information related to the patient's medical history when given examination data. The patient can be a past patient. That is, the patient can be the same person as the patient or a different person.

[0065] Treatment-related information includes, for example, information relating to the time point at which the learner undergoes examination, and the treatment that should be administered after that point. Treatment-related information typically includes information inferring the patient's existing illnesses or illnesses they are likely to develop at the starting point, but it is not limited to this. For example, treatment-related information may also include information inferring or determining the name or type of examination, the name of the prescribed medication, the treatment plan, whether the patient has gone home, the type of patient's room, and whether other staff are available to assist. In other words, treatment-related information can include all future events determined through medical history taking. In the following description, as an example, treatment-related information is "inference of disease."

[0066] When the information related to diagnosis and treatment is "disease inference", the training data used to learn the first model MDL1 is a dataset that uses the diseases that the learner already has or the diseases that are likely to be in the future as labels for the correct answers and establishes a corresponding dataset with the examiner's examination data.

[0067] As shown in the figure, the first model MDL1, trained using such data, outputs the patient's disease (an example of "treatment-related information") as an inference result when given the patient's examination data. The inference result of the first model MDL1 is represented, for example, by multi-dimensional vectors or tensors. The vectors or tensors, as element values, contain the probability of the disease. For example, suppose there are three possible diseases that a patient may have: disease A, disease B, and disease C. In this case, when the probability of disease A is set as e1, the probability of disease B as e2, and the probability of disease C as e3, the vector or tensor can be represented as (e1, e2, e3).

[0068] Return to Figure 4 The flowchart is explained below. Next, the inference function 122 obtains the inference result of the disease from the first model MDL1, which has been input with the examination data of the patient (step S104). The inference result includes diseases that are inferred to be already present in the patient or diseases that are inferred to be present in the patient in the future.

[0069] On the other hand, when the acquisition function 121 acquires the response data and examination data for the consultation, the inference function 122 inputs the response data and examination data of the patient acquired by the acquisition function 121 into the second model MDL2 defined by the model information stored in the memory 141 (step S106).

[0070] Figure 7This diagram illustrates an example of Model 2, MDL2. Model 2 is a machine learning model that uses a dataset as training data. This dataset maps information related to the patient's diagnosis and treatment as labels for correct answers to the patient's response and examination data. In other words, Model 2 is a machine learning model that outputs information related to the patient's diagnosis and treatment when given the patient's response and examination data.

[0071] The treatment-related information here includes, for example, information related to the treatment that should be taken after the learning subject's examination or the time when the learning subject answered the medical question, using the time as the starting point. As mentioned above, treatment-related information typically includes information inferring the diseases the patient already has at the starting point or inferring the diseases that are likely to be present in the future, but it is not limited to this and can also include all future matters determined through medical questioning. In the following explanation, as an example, treatment-related information also refers to "inferences about diseases".

[0072] In cases where the information related to diagnosis and treatment is "disease inference", the training data used to learn the second model MDL2 becomes a dataset that establishes a correspondence between the learner's existing diseases or diseases that are likely to be acquired in the future as labels for the correct answers and the learner's response data and examination data.

[0073] As shown in the figure, the second model, MDL2, trained using such data, outputs the patient's disease as an inference result when given the patient's response data and examination data. Similar to the inference result of the first model, MDL1, the inference result of the second model, MDL2, can also be represented by multidimensional vectors and tensors.

[0074] return Figure 4 The flowchart is explained below. Next, the inference function 122 obtains the inference result of the disease from the second model MDL2, which has been input with the response data and examination data of the patient (step S108). The inference result here also includes diseases that are inferred to be present in the patient or diseases that are inferred to be present in the patient in the future.

[0075] Next, the output control function 124 outputs via the output interface 113 a first inference result representing the patient's disease inferred using the first model MDL1, and a second inference result representing the patient's disease inferred using the second model MDL2 (step S110). Thus, the processing of this flowchart ends.

[0076] Figure 8This diagram illustrates an example of the screen displayed on monitor 113a. As shown, for example, output control function 124 can also cause monitor 113a to display the first inference result and the second inference result side by side. In the illustrated example, a case is shown where the disease inferred solely from examination data without using consultation response data (i.e., the first inference result) is "Disease A," and a case is shown where the disease inferred using both consultation response data and examination data (i.e., the second inference result) is "Disease B." By displaying the data in this way, when using a patient's response data, healthcare professionals can be alerted to the possibility that instability in the response data might affect the inference result. As a result, healthcare professionals can have a suspicion that there is instability in the patient's complaint, and therefore can make a more careful diagnosis of the patient compared to cases where the first and second inference results are the same.

[0077] The communication control function 125 can also send the first inference result and the second inference result to the user interface 10 via the communication interface 111. The output control function 22 of the user interface 10 can also be configured to, when the communication interface 11 receives the first inference result and the second inference result from the medical information processing device 100, cause the display 13a of the output interface 13 to display these inference results as images, or cause the speaker 13b to output these inference results as sound.

[0078] According to the first embodiment described above, the processing circuit 120 of the medical information processing device 100 acquires examination data and consultation response data of the patient. The processing circuit 120 infers treatment-related information by inputting the patient's examination data into a pre-learned first model MDL1. For example, the processing circuit 120 can infer the patient's disease based on this treatment-related information.

[0079] The processing circuit 120 infers treatment-related information by inputting the examination data and response data of the patient into the pre-learned second model MDL2. For example, the processing circuit 120 can infer the patient's disease based on the treatment-related information.

[0080] Then, the processing circuit 120 outputs via the output interface 113 a first inference result representing treatment-related information (e.g., the patient's disease) inferred using the first model MDL1, and a second inference result representing treatment-related information (e.g., the patient's disease) inferred using the second model MDL2. Thus, healthcare professionals can treat patients without being affected by the instability of responses to medical inquiries.

[0081] (Second Implementation)

[0082] The second embodiment will now be described. In the second embodiment, the determination of whether the first and second inference results are consistent differs from that in the first embodiment. The description will focus on the differences from the first embodiment, omitting explanations of points common to the first embodiment. In the description of the second embodiment, the same symbols will be used to denote the parts that are the same as in the first embodiment.

[0083] The determination function 123 in the second embodiment is to compare a first inference result representing treatment-related information (e.g., the disease of the treatment subject) inferred using the first model MDL1 with a second inference result representing treatment-related information (e.g., the disease of the treatment subject) inferred using the second model MDL2, and determine whether these inference results are consistent.

[0084] Figure 9 This is a diagram illustrating the comparison of inference results. For example, decision function 123 compares a first inference result with a second inference result and calculates the similarity between these inference results. For instance, decision function 123 can calculate the cosine similarity between the vector / tensor representing the first inference result and the vector / tensor representing the second inference result. Then, if the calculated similarity is above a threshold, decision function 123 determines that the first inference result and the second inference result are consistent; if the similarity is below the threshold, it determines that the first inference result and the second inference result are inconsistent.

[0085] The output control function 124 of the second embodiment is to output an alarm AR via the output interface 113 when the determination function 123 determines that the first inference result and the second inference result are inconsistent, so as to notify medical staff of the situation where these inference results are inconsistent.

[0086] Figure 10 This diagram illustrates other examples of the screen displayed on display 113a. For instance, output control function 124 allows alarm AR to be displayed simultaneously on display 113a if the first inference result and the second inference result are inconsistent. Output control function 124 can also output an alarm sound via speaker 113b.

[0087] According to the second embodiment described above, the processing circuit 120 compares the first inference result with the second inference result and determines whether the first inference result and the second inference result are consistent. When the processing circuit 120 determines that the first inference result and the second inference result are inconsistent, it outputs an alarm AR via the output interface 113. This more strongly reminds medical personnel that instability in the response data can affect the inference results of the disease.

[0088] (Third Implementation)

[0089] The third embodiment will now be described. In this third embodiment, the difference from the embodiments described above is that, when inferring treatment-related information by inputting the patient's examination data and response data into the pre-learned second model MDL2, weighting the response data is applied. The description will focus on the differences from the first and second embodiments; points common to the first and second embodiments will be omitted. In the description of the third embodiment, the same symbols will be used to indicate parts identical to those in the first or second embodiments.

[0090] Figure 11 This is a schematic diagram illustrating the weighting of the response data. The inference function 122 in the third embodiment weights the response data of the patient when inputting it into the pre-learned second model MDL2. For example, the inference function 122 determines the weight coefficient within the range of 0.0 to 1.0, multiplies this weight coefficient by the response data, and then inputs it into the second model MDL2. Thus, the degree to which the response data contributes to the second inference result can be adjusted according to the weight coefficient. The smaller the weight coefficient (closer to 0.0), the closer the second inference result is expected to be to the first inference result. That is, the higher the similarity between the first and second inference results is expected.

[0091] Inference function 122 repeatedly performs the process of changing the weight coefficients while inputting the response data into the second model MDL2. Thus, a second inference result is obtained for each weight coefficient.

[0092] The determination function 123 in the third embodiment is to compare the multiple second inference results obtained for each weight coefficient with each other and determine whether the second inference results are consistent with each other.

[0093] The output control function 124 in the third embodiment is, for example, to display on the display 113a multiple second inference results obtained for each weight coefficient side by side. The output control function 124 can also display an alarm AR on the display 113a if the determination function 123 determines that the second inference results are inconsistent with each other.

[0094] Figure 12This diagram illustrates another example of the screen displayed on monitor 113a. As shown, output control function 124 can cause monitor 113a to display diseases inferred by the second model MDL2 side-by-side for each weighting coefficient. In the illustrated example, a result of "Disease A" is shown when the weighting coefficients are 0.2 and 0.4, but a result of "Disease B" is shown when the weighting coefficients are 0.6 and 0.8. In this case, "Disease A" and "Disease B" are listed as candidates, representing diseases that the patient may have. Therefore, output control function 124 can also cause monitor 113a to display both "Disease A" and "Disease B" as candidates.

[0095] According to the third embodiment described above, when the processing circuit 120 infers treatment-related information about a patient by inputting examination data and response data from a pre-learned second model MDL2, it weights the response data. The processing circuit 120 repeatedly infers treatment-related information while changing the weighting coefficients. The processing circuit 120 compares each of the second inference results representing treatment-related information repeatedly inferred using the second model MDL2 with each other and determines whether the multiple second inference results are consistent. Then, the processing circuit 120 outputs a second inference result for each weighting coefficient. Thus, for example, if the second inference result does not change even when the weighting coefficients are changed, it can be determined that the patient's response data does not affect the inference result of treatment-related information. That is, treatment-related information can be inferred with high accuracy using the patient's response data.

[0096] (Fourth implementation)

[0097] The fourth embodiment will now be described. In this fourth embodiment, the difference from the embodiments described above is that the weighting coefficients for the response data are determined based on the patient's age, the period elapsed since the onset of the disease (onset period), and the patient's experience with answering questions. The description will focus on the differences from the first to third embodiments; points common to the first to third embodiments will be omitted. In the description of the fourth embodiment, the same symbols will be used to indicate the parts that are the same as in the first to third embodiments.

[0098] Similar to the third embodiment, the inference function 122 of the fourth embodiment weights the response data of the patient when inputting the pre-learned second model MDL2. In this case, the inference function 122 of the fourth embodiment determines the weighting coefficients of the response data based on the patient's age, the period elapsed since the patient developed the disease (onset period), and the patient's experience with answering questions.

[0099] For example, children and the elderly are more likely to show fluctuations in their answers during a consultation compared to adults. Therefore, the inference function 122 in the fourth embodiment can be such that the younger the patient is compared to a certain baseline age (e.g., 18 years old), the lower the weighting coefficient of the answer data. Similarly, the inference function 122 can be such that the older the patient is compared to a certain baseline age (e.g., 65 years old), the lower the weighting coefficient of the answer data.

[0100] Patients who have recently developed the disease are more likely to exhibit fluctuations in their responses during a consultation compared to those who have not. Therefore, the inference function 122 in the fourth embodiment can be such that the shorter the period elapsed since the patient developed the disease (the shorter the period after developing the disease), the lower the weighting coefficient of the response data.

[0101] Patients answering questions for the first time during their initial consultation are more likely to exhibit fluctuating responses compared to experienced patients who have undergone multiple consultations and answered questions before. Therefore, the inference function 122 in the fourth embodiment can be to reduce the weighting coefficient of the response data if the patient has less experience answering questions.

[0102] Therefore, the influence of fluctuations in responses caused by factors such as the patient's age, disease duration, and prior experience can be eliminated from the output of the learned MDL model. As a result, healthcare professionals can treat patients while further considering the uncertainties inherent in the learned MDL model.

[0103] (Fifth Embodiment)

[0104] The fifth embodiment will now be described. In the fifth embodiment, the difference from the embodiments described above is that multiple machine learning models are prepared according to the age of the patient, or according to the duration of the patient's illness, or according to the patient's experience in responding to medical history. The description will focus on the differences from the first to fourth embodiments; points common to the first to fourth embodiments will be omitted. In the description of the fifth embodiment, the same symbols will be used to indicate the parts that are the same as in the first to fourth embodiments.

[0105] For example, the processing circuit 120 can generate a first model MDL1 for children by using training data corresponding to the examination data of children under 18 years old, using information related to the diagnosis and treatment of children (e.g., diseases) as labels for correct answers, or generate a second model MDL2 for children by using training data corresponding to the response data and examination data of children under 18 years old, using information related to the diagnosis and treatment of children (e.g., diseases) as labels for correct answers.

[0106] Similarly, the processing circuit 120 can generate a first model MDL1 for adults by using training data corresponding to the examination data of adults aged 18 to 65 years old, using information related to the diagnosis and treatment of adults (e.g., diseases) as labels for the correct answers, or generate a second model MDL2 for adults by using training data corresponding to the response data and examination data of adults aged 18 to 65 years old, using information related to the diagnosis and treatment of adults (e.g., diseases) as labels for the correct answers.

[0107] The processing circuit 120 can generate a first model MDL1 for the elderly by using training data corresponding to the examination data of the elderly over 65 years old, using information related to the elderly's diagnosis and treatment (e.g., illness) as labels for the correct answers; or it can generate a second model MDL2 for the elderly by using information related to the elderly's diagnosis and treatment (e.g., illness) as labels for the correct answers, using training data corresponding to the response data and examination data of the elderly over 65 years old. This allows the influence of age-related fluctuations in responses to be eliminated from the uncertainty of the learned model MDL.

[0108] The processing circuit 120 in the fifth embodiment may, for example, generate a first model MDL1 specifically for patients whose illness period is less than one year by using training data corresponding to the examination data of the patient whose illness period is less than one year by using information related to the patient's diagnosis and treatment (e.g., disease) as the label of the correct answer. Alternatively, it may generate a second model MDL2 specifically for patients whose illness period is less than one year by using training data corresponding to the response data and examination data of the patient whose illness period is less than one year by using information related to the patient's diagnosis and treatment (e.g., disease) as the label of the correct answer.

[0109] The processing circuit 120 can, for example, generate a first model MDL1 specifically for patients with an illness duration of more than one year by using training data corresponding to examination data of the patient with an illness duration of more than one year, using information related to the patient's diagnosis and treatment (e.g., illness) as labels for positive answers; or generate a second model MDL2 specifically for patients with an illness duration of more than one year by using training data corresponding to the patient's response data and examination data of the patient with an illness duration of more than one year, using information related to the patient's diagnosis and treatment (e.g., illness) as labels for positive answers. This allows the influence of fluctuations in responses corresponding to the illness duration to be eliminated from the uncertainty of the learned model MDL.

[0110] The processing circuit 120 in the fifth embodiment may, for example, generate a first model MDL1 specifically for a first-visit patient by using information related to the diagnosis and treatment of the first-visit patient (e.g., disease) as the label of the correct answer and establishing corresponding training data with the examination data of the first-visit patient whose experience in answering questions is 0, or generate a second model MDL2 specifically for a first-visit patient by using information related to the diagnosis and treatment of the first-visit patient (e.g., disease) as the label of the correct answer and establishing corresponding training data with the answer data and examination data of the first-visit patient.

[0111] The processing circuit 120 in the fifth embodiment can, for example, generate a first model MDL1 specifically for patients undergoing subsequent examinations by using training data corresponding to the examination data of the patient whose experience in answering questions during subsequent examinations is 1 or more, using information related to the patient's diagnosis and treatment (e.g., illness) as labels for correct answers. Alternatively, it can generate a second model MDL2 specifically for patients undergoing subsequent examinations by using training data corresponding to the response data and examination data of the patient whose experience in answering questions during subsequent examinations is 1 or more, using information related to the patient's diagnosis and treatment (e.g., illness) as labels for correct answers. This eliminates the influence of fluctuations in responses corresponding to the patient's experience in answering questions during subsequent examinations from the uncertainty of the learned model MDL.

[0112] (Other implementation methods)

[0113] The following describes other implementation methods. In the above implementation method, the case where the first model MDL1 and the second model MDL2 are different models has been described, but it is not limited to this, and they can also be the same model. The first model MDL1 and the second model MDL2 can also be models with some parameters (weights, bias components, etc. of the DNN) that are different, and the remaining parameters that are the same.

[0114] In the above embodiments, the user interface 10 and the medical information processing device 100 have been described as separate devices, but this is not a limitation. For example, the user interface 10 and the medical information processing device 100 may also be integrated into one device. For example, the processing circuit 20 of the user interface 10 may, in addition to having the acquisition function 21, the output control function 22, and the communication control function 23, also have the inference function 122 and the determination function 123 of the processing circuit 120 of the medical information processing device 100. In this case, the user interface 10 can independently (offline) perform the processing of the various flowcharts described above.

[0115] As described in the above embodiments, the diagnosis-related information output by the first model MDL1 and the second model MDL2 is typically information that infers the patient's disease, but is not limited to this. It may also include all future matters determined through consultation, such as the name and type of examination, the name of the prescribed drug, the treatment plan, whether the patient has gone home, the type of the patient's ward, and whether other staff members are available to help.

[0116] According to at least one embodiment described above, the processing circuit 120 of the medical information processing device 100 acquires examination data and consultation response data of the patient. The processing circuit 120 infers treatment-related information by inputting the patient's examination data into a pre-learned first model MDL1. For example, the processing circuit 120 infers the patient's disease. The processing circuit 120 also infers treatment-related information by inputting the patient's examination data and consultation response data into a pre-learned second model MDL2. For example, the processing circuit 120 infers the patient's disease. Then, the processing circuit 120 outputs a first inference result representing the treatment-related information (e.g., the patient's disease) inferred using the first model MDL1 and a second inference result representing the treatment-related information (e.g., the patient's disease) inferred using the second model MDL2 via an output interface 113. Thus, medical personnel can treat patients without being affected by the instability of consultation responses.

[0117] Several embodiments of the present invention have been described, but these embodiments are provided by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other ways, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, and are included in the scope of the invention as described in the claims and its equivalents.

[0118] Regarding the above-described embodiments, the following notes are disclosed as an aspect of the invention and as optional features.

[0119] (Postscript 1)

[0120] A medical information processing system, comprising:

[0121] The acquisition department acquires examination data representing the results of medical examinations conducted on patients, and response data representing the results of interviews conducted on the aforementioned patients.

[0122] The inference unit infers treatment-related information by inputting the examination data of the patient into the first model, and infers treatment-related information by inputting the examination data and response data of the patient into the second model. The first model learns by establishing a first training dataset corresponding to the examination data of the patient, using the treatment-related information of the patient as a label for the correct answer. The second model learns by establishing a second training dataset corresponding to the examination data and response data of the patient, using the treatment-related information of the patient as a label for the correct answer.

[0123] The output control unit outputs a first inference result representing information related to the diagnosis and treatment inferred using the first model, and a second inference result representing information related to the diagnosis and treatment inferred using the second model.

[0124] (Postscript 2)

[0125] The output unit may include a display unit. The output control unit may enable the display unit to display the first inference result and the second inference result side by side.

[0126] (Note 3)

[0127] The aforementioned medical information processing system may further include a determination unit that compares the first inference result with the second inference result and determines whether the first inference result and the second inference result are consistent.

[0128] (Note 4)

[0129] The inference unit can calculate the similarity between the first inference result and the second inference result. Furthermore, the determination unit can determine that the first inference result and the second inference result are consistent if the similarity is above a threshold, and determine that the first inference result and the second inference result are inconsistent if the similarity is below the threshold.

[0130] (Note 5)

[0131] If the first inference result is determined to be inconsistent with the second inference result, the output control unit may output an alarm via the output unit.

[0132] (Note 6)

[0133] The inference unit can weight the patient's response data when inputting it into the second model. Furthermore, by inputting the weighted response data and examination data into the second model, the inference unit can infer information related to the patient's treatment.

[0134] (Note 7)

[0135] The inference unit can repeatedly infer information related to the diagnosis and treatment while changing the weighting coefficients. The determination unit can determine whether the multiple inference results are consistent by comparing each of the multiple pieces of information related to the diagnosis and treatment repeatedly inferred using the second model.

[0136] (Note 8)

[0137] The output control unit can output the second inference result for each of the weight coefficients via the output unit.

[0138] (Note 9)

[0139] Information related to the above diagnosis and treatment may include information that infers the patient's disease.

[0140] (Postscript 10)

[0141] A medical information processing method, using a computer, includes:

[0142] Obtain examination data representing the results of medical examinations conducted on the patients, and response data representing the results of interviews conducted on the aforementioned patients;

[0143] By inputting the examination data of the aforementioned patients into the first model, information related to the diagnosis and treatment of the aforementioned patients is inferred. The first model learns by establishing a first training dataset corresponding to the examination data of the aforementioned patients, based on the information related to the diagnosis and treatment of the learning subjects as the label of the positive solution.

[0144] By inputting the aforementioned examination data and response data of the patients into the second model, information related to the diagnosis and treatment of the patients is inferred. The second model learns by establishing a second training dataset corresponding to the examination data and response data of the patients, using the information related to the diagnosis and treatment of the patients as labels for the correct solutions; and

[0145] The output unit outputs a first inference result representing information related to the diagnosis and treatment derived using the first model, and a second inference result representing information related to the diagnosis and treatment derived using the second model.

[0146] (Postscript 11)

[0147] A storage medium storing a program for executing by a computer, the program comprising:

[0148] Obtain examination data representing the results of medical examinations conducted on the patients, and response data representing the results of interviews conducted on the aforementioned patients;

[0149] By inputting the examination data of the aforementioned patients into the first model, information related to the diagnosis and treatment of the aforementioned patients is inferred. The first model learns by establishing a first training dataset corresponding to the examination data of the aforementioned patients, based on the information related to the diagnosis and treatment of the learning subjects as the label of the positive solution.

[0150] By inputting the aforementioned examination data and response data of the patients into the second model, information related to the diagnosis and treatment of the patients is inferred. The second model learns by establishing a second training dataset corresponding to the examination data and response data of the patients, using the information related to the diagnosis and treatment of the patients as labels for the correct solutions; and

[0151] The output unit outputs a first inference result representing information related to the diagnosis and treatment derived using the first model, and a second inference result representing information related to the diagnosis and treatment derived using the second model.

Claims

1. A medical information processing system, comprising: The acquisition department acquires examination data representing the results of medical examinations conducted on patients, and response data representing the results of interviews conducted on the aforementioned patients. The inference unit infers treatment-related information by inputting the examination data of the patient into the first model, and infers treatment-related information by inputting the examination data and response data of the patient into the second model. The first model learns by establishing a first training dataset corresponding to the examination data of the patient, using the treatment-related information of the patient as the label of the correct answer. The second model learns by establishing a second training dataset corresponding to the examination data and response data of the patient, using the treatment-related information of the patient as the label of the correct answer. The output control unit outputs a first inference result representing information related to the diagnosis and treatment inferred using the first model and a second inference result representing information related to the diagnosis and treatment inferred using the second model. as well as The determination unit compares the first inference result with the second inference result and determines whether the first inference result and the second inference result are consistent. If the first inference result is determined to be inconsistent with the second inference result, the output control unit outputs an alarm via the output unit.

2. The medical information processing system according to claim 1, wherein, The aforementioned output section includes a display section. The output control unit causes the display unit to display the first inference result and the second inference result side by side.

3. The medical information processing system according to claim 1, wherein, The above determination part is, Calculate the similarity between the first inference result and the second inference result. If the similarity score is above the threshold, the first inference result is determined to be consistent with the second inference result. If the similarity is less than the threshold, the first inference result is determined to be inconsistent with the second inference result.

4. The medical information processing system according to claim 1, wherein, The above inference is, When inputting the aforementioned response data from the patients into the second model, the response data from the patients is weighted. By inputting the weighted response data and examination data of the aforementioned patients into the second model, information related to the diagnosis and treatment of the aforementioned patients can be inferred.

5. The medical information processing system according to claim 4, wherein, The aforementioned inference process repeatedly modifies the weighting coefficients while inferring information related to the aforementioned diagnosis and treatment. The determination unit compares the second inference results, which represent each of the multiple pieces of information related to the diagnosis and treatment repeatedly inferred using the second model, to determine whether the multiple second inference results are consistent with each other.

6. The medical information processing system according to claim 5, wherein, The output control unit outputs the second inference result for each of the weighting coefficients via the output unit.

7. A medical information processing method, using a computer, comprising: Obtain examination data representing the results of medical examinations conducted on the patients, and response data representing the results of interviews conducted on the aforementioned patients; By inputting the examination data of the aforementioned patients into the first model, information related to the diagnosis and treatment of the aforementioned patients is inferred. The first model learns by establishing a first training dataset corresponding to the examination data of the aforementioned patients, based on the information related to the diagnosis and treatment of the learning subjects as the label of the positive solution. By inputting the examination data and response data of the aforementioned patients into the second model, information related to the diagnosis and treatment of the aforementioned patients is inferred. The second model learns by establishing a second training dataset corresponding to the examination data and response data of the aforementioned patients, based on the information related to the diagnosis and treatment of the aforementioned patients as labels for the correct solutions. The output unit outputs a first inference result representing information related to the diagnosis and treatment inferred using the first model, and a second inference result representing information related to the diagnosis and treatment inferred using the second model; and Compare the first inference result with the second inference result, and determine whether the first inference result is consistent with the second inference result. If the first inference result is determined to be inconsistent with the second inference result, an alarm is output via the output unit.

8. A storage medium storing a program for causing a computer to execute, the program comprising: Obtain examination data representing the results of medical examinations conducted on the patients, and response data representing the results of interviews conducted on the aforementioned patients; By inputting the examination data of the aforementioned patients into the first model, information related to the diagnosis and treatment of the aforementioned patients is inferred. The first model learns by establishing a first training dataset corresponding to the examination data of the aforementioned patients, based on the information related to the diagnosis and treatment of the learning subjects as the label of the positive solution. By inputting the examination data and response data of the aforementioned patients into the second model, information related to the diagnosis and treatment of the aforementioned patients is inferred. The second model learns by establishing a second training dataset corresponding to the examination data and response data of the aforementioned patients, based on the information related to the diagnosis and treatment of the aforementioned patients as labels for the correct solutions. The output unit outputs a first inference result representing information related to the diagnosis and treatment inferred using the first model, and a second inference result representing information related to the diagnosis and treatment inferred using the second model; and Compare the first inference result with the second inference result, and determine whether the first inference result is consistent with the second inference result. If the first inference result is determined to be inconsistent with the second inference result, an alarm is output via the output unit.

Images