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

Abstract

To allow a user to further understand output results of a machine learning model.SOLUTION: A medical information processing device is configured to: extract, when assisting in interpreting output results of a model that outputs medical treatment assistance information as information for assisting in medical treatment of an examinee in response to the fact that a plurality of pieces of medical treatment information obtained when medically treating the examinee are inputted, first medical treatment information associated with medical treatment assistance information from among the plurality of pieces of medical treatment information, on the basis of correspondence in which the medical treatment assistance information is correlated to the plurality of pieces of medical treatment information and the degree of contribution of each of the plurality of pieces of medical treatment information to the medical treatment assistance information; extract second medical treatment information associated with the first medical treatment information from among the plurality of pieces of medical treatment information, on the basis of the mutual association degree of the medical treatment information and the contribution degree of each of the plurality of pieces of medical treatment information; and cause a display unit to display the medical treatment assistance information, the first medical treatment information and the second medical treatment information.SELECTED DRAWING: Figure 1

Description

Technical Field

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

Background Art

[0002] Cancer treatment is cardiotoxic, so it is important to grasp the patient's risk of heart disease before treatment. Therefore, a clinical decision support system (CDS) has been developed. In CDS, it is important to show the basis of the estimation result. By grasping the basis, the doctor can consider the treatment policy. In RAIM (Recurrent Attentive and Intensive Model), by using a machine learning model or the like, a disease risk (support information) is calculated from a number of test values (medical information), and test items that contributed to the calculation of the disease risk can be extracted from among the number of test values. In some cases, medically unknown medical information that is not related to the disease is used as input, and the model may extract it based on the medically unknown medical information. When medically unknown medical information is presented, the user may not be able to understand why it was extracted as a basis.

[0003] In relation to this, for example, a technique for generating a graph representing the relationship between a risk and its contributing topics (words) based on medical documents or academic papers is known. In the conventional technique, since the graph is generated based on data that is already medically known, such as medical documents or academic papers, medically unknown medical information is not targeted, and the relevance cannot be presented to the user.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] The problem that the embodiments disclosed in this specification and drawings aim to solve is to make the output results of a machine learning model, which outputs diagnostic support information that assists in the diagnosis of a subject in response to the input of multiple medical information obtained when the subject is examined, more easily understood by the user. However, the problem that the embodiments disclosed in this specification and drawings aim to solve is not limited to the above problem. Problems corresponding to each effect of each configuration shown in the embodiments described later can also be positioned as other problems. [Means for solving the problem]

[0006] The medical information processing device of the embodiment is a medical information processing device for assisting the interpretation of the output results of a model that outputs diagnostic support information, which is information to support the diagnosis of a subject, in response to the input of multiple medical information obtained when a subject is examined. The medical information processing device has a first extraction unit, a second extraction unit, and a display control unit. The first extraction unit extracts first medical information related to the diagnostic support information from among the multiple medical information based on the correspondence relationship between the diagnostic support information and the multiple medical information, and the degree of contribution of each of the multiple medical information input to the model when the model is made to output the diagnostic support information. The second extraction unit extracts second medical information related to the first medical information from among the multiple medical information based on the degree of relevance between the medical information input to the model when the model is made to output the diagnostic support information, and the degree of contribution of each of the multiple medical information. The display control unit displays the diagnostic support information, the first medical information, and the second medical information on the display unit. [Brief explanation of the drawing]

[0007] [Figure 1] A diagram showing an example configuration of the medical information processing device 100 in an embodiment. [Figure 2] A flowchart illustrating the sequence of processes in the processing circuit 120 according to the embodiment. [Figure 3]This figure shows an example of the contribution of each piece of medical information output by machine learning model 200. [Figure 4] A diagram illustrating an example of a medical relevance database. [Figure 5] This figure shows an example of the weights (relevance) between multiple medical information pieces output by machine learning model 200. [Figure 6] A diagram showing an example of content to be displayed on display 113a. [Figure 7] A diagram illustrating another example of content to be displayed on display 113a. [Figure 8] A diagram illustrating another example of content to be displayed on display 113a. [Figure 9] A diagram illustrating another example of content to be displayed on display 113a. [Modes for carrying out the invention]

[0008] The medical information processing device, medical information processing method, and program of the embodiment will be described below with reference to the drawings.

[0009] [Configuration of medical information processing device] Figure 1 is a diagram showing an example configuration of the medical information processing device 100 in an 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.

[0010] The communication interface 111 communicates with external devices via the communication network NW. The communication network NW may refer to any information and communication network that utilizes telecommunications technology. For example, the communication network NW includes wireless / wired LANs such as hospital backbone LANs (Local Area Networks), the Internet network, as well as telephone communication lines, optical fiber communication networks, cable communication networks, and satellite communication networks. The communication interface 111 may include, for example, a NIC (Network Interface Card) or an antenna for wireless communication.

[0011] The input interface 112 receives various input operations from the operator, converts the received input operations into electrical signals, and outputs them to the processing circuit 120. For example, the input interface 112 includes a mouse, keyboard, trackball, switch, button, joystick, touch panel, etc. The input interface 112 may also be a user interface that accepts audio input, such as from a microphone. If the input interface 112 is a touch panel, the input interface 112 may also incorporate the display function of the display 113a included in the output interface 113, which will be described later.

[0012] In this specification, the input interface 112 is not limited to those equipped with physical operating components such as a mouse or keyboard. For example, 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 this electrical signal to a control circuit is also included as an example of the input interface 112.

[0013] The output interface 113 includes, for example, a display 113a and a speaker 113b. The display 113a displays various types of information. For example, the display 113a displays images generated by the processing circuit 120, or a GUI (Graphical User Interface) for receiving various input operations from the operator. For example, the display 113a may be an LCD (Liquid Crystal Display), a CRT (Cathode Ray Tube) display, or an organic EL (Electro Luminescence) display. The speaker 113b outputs the information input from the processing circuit 120 as sound.

[0014] Memory 114 can be implemented by semiconductor memory elements such as RAM (Random Access Memory), flash memory, hard disks, or optical discs. These non-transient storage media may also be implemented by other storage devices connected via a communication network NW, such as NAS (Network Attached Storage) or external storage server devices. Memory 114 may also include non-transient storage media such as ROM (Read Only Memory) or registers.

[0015] Memory 114 stores model information in addition to programs executed by the hardware processor. Model information refers to information (programs or algorithms) that define a machine learning model 200, which has been trained to output at least diagnostic support information, which is information to support the diagnosis of a subject (e.g., a human patient), in response to input of multiple medical information obtained when the subject is examined.

[0016] The machine learning model 200 may be implemented, for example, by a neural network. More specifically, the machine learning model 200 may be implemented by a convolutional neural network with an attention mechanism. Furthermore, the machine learning model 200 is not limited to neural networks and may be implemented by other models such as support vector machines, decision trees, Naive Bayesian classifiers, and random forests.

[0017] When the machine learning model 200 is implemented by a neural network, the model information includes, for example, connection information on how the units included in each of the input layer, one or more hidden layers (intermediate layers), and output layer that constitute the neural network are connected to each other, and weight information on how many connection coefficients are assigned to the data input and output between the connected units. The connection information includes, for example, the number of units included in each layer, information specifying the type of the unit to which each unit is connected, the activation function that realizes each unit, and information such as a gate provided between the units of the hidden layer. The activation function that realizes a unit may be, for example, a ReLU (Rectified Linear Unit) function, an ELU (Exponential Linear Units) function, a clipping function, a sigmoid function, a step function, a hyperbolic tangent function, an identity function, or the like. The gate selectively passes or weights the data transmitted between units according to the value (for example, 1 or 0) returned by the activation function. The connection coefficient includes, for example, the weight assigned to the output data when data is output from a unit in a certain layer to a unit in a deeper layer in the hidden layer of the neural network. Further, the connection coefficient may include a unique bias component of each layer and the like.

[0018] Furthermore, in addition to programs and model information, the memory 114 stores a medical relevance database. The medical relevance database is, for example, a database in which the presence or absence of the medical relevance of parameters such as blood pressure and heart rate is associated with diseases such as heart disease and cancer. The medical relevance database is an example of a "correspondence relationship". [[ID=*]]

[0019] The processing circuit 120 includes, for example, an acquisition function 121, a calculation function 122, an extraction function 123, and an output control function 124. The processing circuit 120 realizes these functions by, for example, a hardware processor (computer) executing a program stored in the memory 114 (storage circuit). The extraction function 123 is an example of a "first extraction unit", a "second extraction unit", and a "third extraction unit". The output control function 124 is an example of a "display control unit".

[0020] The hardware processor in the processing circuit 120 means, for example, a circuit such as a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), an application specific integrated circuit (ASIC), a programmable logic device (for example, a simple programmable logic device (SPLD) or a complex programmable logic device (CPLD), a field programmable gate array (FPGA)). Instead of storing the program in the memory 114, it may be configured to directly incorporate the program into the circuit of the hardware processor. In this case, the hardware processor realizes its functions by reading and executing the program incorporated into the circuit. The above program may be stored in the memory 114 in advance, or may be stored in a non-temporary storage medium such as a DVD or a CD-ROM, and may be installed from the non-temporary storage medium to the memory 114 when the non-temporary storage medium is mounted on a drive device (not shown) of the user interface 10. The hardware processor is not limited to being configured as a single circuit, and may be configured as one hardware processor by combining a plurality of independent circuits to realize each function. Also, a plurality of components may be integrated into one hardware processor to realize each function.

[0021] [Processing flow of medical information processing equipment] The following describes a series of processes performed by the processing circuit 120 of the medical information processing device 100, following a flowchart. Figure 2 is a flowchart showing the flow of a series of processes performed by the processing circuit 120 according to this embodiment.

[0022] First, the acquisition function 121 acquires multiple pieces of medical information obtained when examining the patient who is the subject of the diagnosis (step S100). Medical information includes parameters (test items) such as blood pressure, heart rate, SpO2 (transcutaneous arterial oxygen saturation), weight, blood glucose level, NT-proBNP, and cardiac troponin I.

[0023] For example, suppose a medical professional such as a doctor or nurse inputs multiple parameters, such as a patient's blood pressure and heart rate, into the user interface 10. In this case, the acquisition function 121 acquires each parameter input into the user interface 10 as medical information. Alternatively, if a medical professional inputs multiple parameters of a patient into a dedicated terminal within the hospital instead of the user interface 10, the acquisition function 121 may communicate with the dedicated terminal via the communication interface 111 and acquire each parameter input into the dedicated terminal as medical information.

[0024] Next, when the acquisition function 121 acquires multiple medical information (multiple parameters such as blood pressure and heart rate), the calculation function 122 uses the machine learning model 200 defined by the model information stored in memory 1141 to calculate the contribution of each medical information and the weights (also called correlation) between the multiple medical information from the multiple medical information acquired by the acquisition function 121 (step S102).

[0025] As described above, when multiple pieces of medical information are input into the machine learning model 200, the machine learning model 200 outputs diagnostic support information.

[0026] Diagnostic support information is, for example, information that expresses the probability, as a quantitative risk, that a patient was already suffering from a particular disease (such as heart disease or cancer) at the time of consultation (the time the patient was examined to obtain the medical information). Alternatively, diagnostic support information may also express, for example, the probability, that a patient will suffer from a particular disease at some point in the future beyond the time of consultation, as a quantitative risk. In other words, the machine learning model 200 outputs the probability that a patient is already suffering from or will suffer from a particular disease as diagnostic support information.

[0027] The machine learning model 200 is typically trained using a dataset as training data, where multiple medical records from a given subject are labeled with diseases the subject has already had or is highly likely to have in the future. In other words, the machine learning model 200 is a model that, when given multiple medical records from a given subject, outputs diseases the subject has already had or is highly likely to have in the future. The subject may be a patient who has been diagnosed in the past. That is, the subject may be the same person as the person who was diagnosed, or a different person.

[0028] The machine learning model 200, having been trained in this manner, will output diagnostic support information (disease risk) as an estimation result when it receives multiple diagnostic pieces of information about a person being diagnosed. The estimation result of the machine learning model 200 can be represented, for example, as a multidimensional vector or tensor. The vector or tensor includes the probability of having a disease as an element value. For example, suppose there are three types of diseases that the person being diagnosed may have: disease A, disease B, and disease C. In this case, the vector or tensor can be represented as (e1, e2, e3), where e1 is the probability of disease A, e2 is the probability of disease B, and e3 is the probability of disease C.

[0029] Furthermore, in addition to diagnostic support information (disease risk), the machine learning model 200 also outputs the contribution of each clinical information and the weights (relevance) between multiple clinical information sources.

[0030] The contribution of clinical information is an index that represents the extent to which each piece of clinical information input to the machine learning model 200 contributes to the diagnostic support information (disease risk) output by the machine learning model 200 when the model 200 is used to output diagnostic support information (disease risk).

[0031] The weight (relevance) between medical information is an indicator that represents the degree to which a particular medical information input into the machine learning model 200 is related to all other medical information input into the machine learning model 200 when the model outputs diagnostic support information (disease risk).

[0032] For example, if machine learning model 200 is implemented using a convolutional neural network with an attention mechanism, the attention mechanism calculates the contribution of each clinical information to the diagnostic support information, as well as the weights (relevance) between multiple clinical information pieces.

[0033] Furthermore, the calculation function 122 may use a visualization method called Class Activation Mapping (CAM) to calculate the contribution of each clinical information to the diagnostic support information and the weights (relevance) between multiple clinical information pieces. For example, Gradient-Weighted Class Activation Mapping (Grad-CAM) can be used as the Class Activation Mapping method.

[0034] Figure 3 shows an example of the contribution of each clinical information output by the machine learning model 200. As shown in the figure, the machine learning model 200 calculates the contribution of each of several clinical information (parameters), such as blood pressure, heart rate, SpO2, weight, blood glucose level, NT-proBNP, and cardiac troponin I, to diagnostic support information (disease risk).

[0035] Returning to the explanation of the flowchart in Figure 2, the extraction function 123 extracts clinical information (hereinafter referred to as "main explanatory items") related to diagnostic support information (disease risk) from among the multiple clinical information input to the machine learning model 200, based on the medical relevance database stored in memory 114 and the contribution of each clinical information calculated by the calculation function 122 (step S104). The main explanatory items are an example of "first clinical information".

[0036] Figure 4 shows an example of a medical relevance database. As shown in the figure, a medical relevance database is a table data in which the presence or absence of medical relevance of each clinical information (parameter) to a target disease, such as heart disease, is associated with a numerical value such as "1" or "0". In the example medical relevance database shown, out of seven types of clinical information (parameters)—blood pressure, heart rate, SpO2, weight, blood glucose level, NT-proBNP, and cardiac troponin I—five types of clinical information, blood pressure, heart rate, SpO2, NT-proBNP, and cardiac troponin I, are medically relevant to heart disease (the table field is "1"), while the other two types of clinical information, weight and blood glucose level, are not medically relevant to heart disease (the table field is "0").

[0037] For example, suppose seven types of medical information—blood pressure, heart rate, SpO2, weight, blood glucose level, NT-proBNP, and cardiac troponin I—are input to machine learning model 200, and machine learning model 200 outputs the risk of heart disease as diagnostic support information. In this case, the extraction function 123 first refers to the medical relevance database and extracts medical information that is medically relevant to heart disease (those with a field of "1") from the seven types of medical information input to machine learning model 200. In the example in Figure 4, five types of medical information are extracted.

[0038] Furthermore, the extraction function 123, while referring to a table of contributions to each clinical information item as illustrated in Figure 3, extracts a predetermined number of clinical information items with the highest contribution to the diagnostic support information of heart disease from among the five types of clinical information items that are medically relevant to heart disease, as primary explanatory items. The predetermined number is a natural number (a positive integer that does not include 0). For example, if the predetermined number is 1, blood pressure (contribution 0.9), which has the highest contribution among the five types of clinical information items, is extracted as the primary explanatory item.

[0039] Furthermore, the extraction function 123 may extract all clinical information from among the five types of clinical information that are medically relevant to heart disease, whose contribution to the diagnostic support information of heart disease is above a threshold, as primary explanatory items. For example, if the threshold is 0.4, then blood pressure, with a contribution of 0.9, and cardiac troponin I, with a contribution of 0.4, will be extracted as primary explanatory items from among the five types of clinical information.

[0040] Returning to the explanation of the flowchart in Figure 2, the extraction function 123 extracts the primary explanatory item and related medical information (hereinafter referred to as the secondary explanatory item) from among the multiple medical information input to the machine learning model 200, based on the contribution of each medical information and the weights (relevance) between multiple medical information calculated by the calculation function 122 (step S106). The secondary explanatory item is an example of the "second medical information".

[0041] Figure 5 shows an example of the weights (relevance) between multiple medical information outputs generated by the machine learning model 200. As shown in the figure, weights (relevance) to all other medical information are calculated for each piece of medical information. For example, suppose that from the seven types of medical information input into the machine learning model 200, blood pressure is extracted as the primary explanatory item for the diagnostic support information of heart disease. In this case, the extraction function 123 refers to the top record in the table in Figure 5 where the weights of other medical information for blood pressure are stored, and extracts medical information that is not medically relevant to heart disease as candidate supplementary explanatory items. In the example shown, weight and blood glucose level are extracted as candidate supplementary explanatory items.

[0042] The extraction function 123 extracts one or more candidate supplementary explanation items from the seven types of medical information input into the machine learning model 200. From these candidate supplementary explanation items, it extracts a predetermined number of top candidates with the highest weight (relevance) to the primary explanation item as the supplementary explanation item. For example, if the predetermined number is 1, among the candidate supplementary explanation items, weight and blood glucose level, blood glucose level, which has a higher weight to the primary explanation item, blood pressure, is extracted as the official supplementary explanation item.

[0043] Furthermore, the extraction function 123 may extract one or more candidate supplementary explanatory items from the seven types of medical information input to the machine learning model 200, and then extract all of those candidate supplementary explanatory items whose weight (relevance) to the primary explanatory item is above a threshold as supplementary explanatory items. For example, if the threshold is 0.5, the weight of one of the candidate supplementary explanatory items, weight, to blood pressure is 0.6, and the weight of the other, blood glucose level, to blood pressure is 0.9, both of which are above the threshold of 0.5, so weight and blood glucose level are extracted as supplementary explanatory items.

[0044] Furthermore, the extraction function 123, after extracting one or more supplementary explanatory items, may extract medical information related to the supplementary explanatory items (hereinafter referred to as "second supplementary explanatory items") from among the multiple medical information input to the machine learning model 200, based on the contribution of each medical information and the weights (relevance) between the multiple medical information. The second supplementary explanatory items are an example of "third medical information."

[0045] For example, suppose blood glucose level is extracted as a supplementary explanatory item. In this case, the extraction function 123 refers to the fifth record from the top in the table in Figure 5, which stores the weights of other medical information relative to blood glucose level, and extracts medical information that is not medically relevant to heart disease as a candidate for a second supplementary explanatory item. In the illustrated example, body weight is extracted as a candidate for the second supplementary explanatory item (blood glucose level is not extracted again as it has already been extracted).

[0046] The extraction function 123 extracts one or more candidates for second auxiliary explanatory items, and then extracts a predetermined number of the top candidates from among those candidates that have a large weight (degree of relevance) to the auxiliary explanatory items as the second auxiliary explanatory items. In the illustrated example, since there is only one type of candidate for the second auxiliary explanatory item, weight is automatically extracted as the second auxiliary explanatory item.

[0047] Furthermore, the extraction function 123 may, after extracting one or more candidates for second auxiliary explanatory items, extract all of those candidates for second auxiliary explanatory items whose weight (relevance) to the auxiliary explanatory item is above a threshold, as second auxiliary explanatory items.

[0048] Returning to the explanation of the flowchart in Figure 2, the output control function 124 then outputs the extraction result of the extraction function 123 (step S108).

[0049] For example, the output control function 124 may display content on the display 113a of the output interface 113 that includes the main explanatory items and auxiliary explanatory items extracted by the extraction function 123, and the diagnostic support information (disease risk) calculated by the calculation function 122.

[0050] Furthermore, the output control function 124 may, in addition to or instead of displaying content on the display 113a, transmit content to an external device (for example, a dedicated terminal within the hospital used for inputting medical information) via the communication interface 111.

[0051] Figure 6 shows an example of content to be displayed on display 113a. In the figure, the node labeled "cardiac disease risk" represents the probability calculated by the machine learning model 200 as diagnostic support information, the node labeled "blood pressure" represents the medical information extracted as the primary explanatory item, and the node labeled "blood glucose level" represents the medical information extracted as a secondary explanatory item.

[0052] For example, the output control function 124 determines the display mode of the content based on the contribution of the main explanatory item to the diagnostic support information, the contribution of the auxiliary explanatory item to the diagnostic support information, and the relative weights (relevance) of the main explanatory item and the auxiliary explanatory item. Specifically, the output control function 124 displays a directed graph as content, in which one-way solid arrows are provided at the edges between the diagnostic support information node and the main explanatory item node, one-way dashed arrows are provided at the edges between the diagnostic support information node and the auxiliary explanatory item node, and one-way solid arrows are provided at the edges between the main explanatory item node and the auxiliary explanatory item node, while distinguishing the color and pattern of each node. On the directed graph, each edge may be associated with a contribution or weight (relevance).

[0053] Furthermore, if the extraction function 123 extracts a second auxiliary explanation item in addition to the auxiliary explanation item, the output control function 124 may display content including the diagnostic support information (disease risk), the main explanation item, the auxiliary explanation item, and the second auxiliary explanation item on the display 113a.

[0054] Figure 7 shows another example of content to be displayed on the display 113a. In the figure, the node labeled "cardiac disease risk" represents the probability calculated by the machine learning model 200 as diagnostic support information, the node labeled "blood pressure" represents medical information extracted as the primary explanatory item, the node labeled "blood glucose level" represents medical information extracted as a secondary explanatory item, and the node labeled "weight" represents medical information extracted as a second secondary explanatory item. Even when such a second secondary explanatory item is extracted, the output control function 124 may display a directed graph as content, distinguishing the colors and patterns of each node.

[0055] Furthermore, while the above explanation primarily described the machine learning model 200 as outputting only the risk of heart disease as diagnostic support information, it may also output the risk of other diseases besides heart disease. In this case, for each of the multiple diseases, a primary explanatory item and a secondary explanatory item will be extracted. Therefore, the output control function 124 may display a directed graph as content, which associates the primary explanatory item and the secondary explanatory item for each disease.

[0056] Figure 8 shows another example of content to be displayed on display 113a. As shown in the figure, when multiple disease risks are calculated, such as disease A, disease B, and disease C, and primary and secondary explanatory items are extracted for each disease, the output control function 124 displays three directed graphs as content, each associating the nodes of the primary and secondary explanatory items with the nodes of disease A, disease B, and disease C.

[0057] Figure 9 shows another example of content to be displayed on the display 113a. For example, disease risk may change over time, or the medical information extracted as primary and secondary explanatory items may change. In such cases, the output control function 124 may display the disease risk and explanatory items that change over time as a graph.

[0058] According to the embodiment described above, the processing circuit 120 of the medical information processing device 100 extracts medical information related to the diagnostic support information from among the multiple medical information input to the machine learning model 200 as a primary explanatory item (an example of a "first medical information"), based on a medical relevance database (an example of a "correspondence relationship") in which the presence or absence of medical relevance of multiple medical information such as blood pressure and heart rate is associated with the diagnostic support information, and the contribution of each of the multiple medical information output by the machine learning model 200 to the diagnostic support information. Furthermore, the processing circuit 120 extracts medical information related to the primary explanatory item as a secondary explanatory item (an example of a "second medical information") from among the multiple medical information input to the machine learning model 200, based on the contribution of each of the multiple medical information output by the machine learning model 200 to the diagnostic support information and the weights (relevance) between the multiple medical information. The processing circuit 120 then displays the content including the diagnostic support information, primary explanatory item, and secondary explanatory item on the display 113a or transmits it to an external device via the communication interface 111.

[0059] Thus, when medical information with unknown medical relevance to the diagnostic support information (disease risk) output by the machine learning model 200 is extracted as supplementary explanatory items, the relationship between the "diagnostic support information" and the "medical information with unknown medical relevance (i.e., supplementary explanatory items)" can be visualized and presented to the user. As a result, the user can visually understand the relationship between the "diagnostic support information" and the "medical information with unknown medical relevance," enabling them to investigate the causes of diseases or acquire new medical insights.

[0060] (Other embodiments) Other embodiments (modifications of the embodiments described above) will be described below. In the embodiments described above, the extraction function 123 was described as extracting medical information that does not have a medical relevance to the diagnostic support information (medical information assigned "0" in the medical relevance database in Figure 4) from among other medical information for which a weight has been calculated with respect to the main explanatory item, as candidates for auxiliary explanatory item, but it is not limited to this.

[0061] For example, the extraction function 123 may extract, as candidates for auxiliary explanation items, medical information from other medical information for which a weight has been calculated with respect to the primary explanation item, that is not associated with the presence or absence of medical relevance to diagnostic support information in the medical relevance database and does not exist in the database. In other words, the extraction function 123 may extract, as candidates for auxiliary explanation items, medical information that is not assigned "1" or "0" in the medical relevance database and does not even have a field.

[0062] The machine learning model 200 described above is a general-purpose model. Therefore, even if medical information whose relevance is unknown and is not registered in the medical relevance database is input to machine learning model 200, machine learning model 200 can calculate diagnostic support information, the contribution of medical information, and the weights (relevance) between multiple pieces of medical information. In this way, even medical information whose medical relevance is unknown and does not exist in the medical relevance database (medical information whose medical relevance is unknown and could not have been anticipated when the medical relevance database was created) can be extracted as candidate supplementary explanation items, allowing users to further understand the output results of the machine learning model.

[0063] Furthermore, in the embodiments described above, diagnostic support information was explained as information that expresses the probability that the patient was already suffering from the target disease (such as heart disease or cancer) at the time of treatment, or the probability that the patient will suffer from it at some point in the future, as a quantitative risk, but it is not limited to this. For example, diagnostic support information may also quantitatively express risks related to other medical factors or medical events different from the disease, such as survival rate, mortality rate, readmission rate, health status, severity, and risk of adverse events.

[0064] While several embodiments have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These embodiments can be carried out in a variety of other forms, 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, as well as in the claims and their equivalents. [Explanation of symbols]

[0065] 100...Medical information processing device, 111...Communication interface, 112...Input interface, 113...Output interface, 113a...Display, 114...Memory, 120...Processing circuit, 121...Acquisition function, 122...Calculation function, 123...Extraction function, 124...Output control function

Claims

1. A medical information processing device for assisting the interpretation of the output results of a machine learning model that outputs diagnostic support information, which is information to support the diagnosis of a subject, in response to input of multiple medical information obtained when a subject is examined, A database to which the diagnostic support information and the multiple medical records are associated, and a first extraction unit that extracts first medical records related to the diagnostic support information from the multiple medical records based on the contribution of each of the multiple medical records input to the machine learning model to the diagnostic support information when the machine learning model is made to output the diagnostic support information, and the contribution output by the machine learning model in response to the input of the multiple medical records to the machine learning model, A second extraction unit extracts second medical information related to the first medical information from among the multiple medical information, based on the degree of relevance between the medical information input to the machine learning model when the machine learning model is made to output the diagnostic support information, the degree of relevance output by the machine learning model in response to the input of the multiple medical information to the machine learning model, and the degree of contribution output by the machine learning model in response to the input of the multiple medical information to the machine learning model. A display control unit that displays the diagnostic support information, the first medical information, and the second medical information on a display unit, A medical information processing device equipped with [a specific feature].

2. The aforementioned database contains information that associates the presence or absence of medical relevance between each of the multiple pieces of medical information with the diagnostic support information. The medical information processing device according to claim 1.

3. The first extraction unit extracts a predetermined number of medical information from the database that have a medical relevance to the diagnostic support information, with the highest contribution, as the first medical information. The medical information processing device according to claim 2.

4. The first extraction unit extracts, from among the multiple clinical information records in the database that have a medical relevance to the diagnostic support information, clinical information records whose contribution is equal to or greater than a threshold, as the first clinical information records. The medical information processing device according to claim 2 or 3.

5. The second extraction unit extracts, as second medical information, medical information that does not have the medical relevance to the diagnostic support information in the database, or medical information that does not exist in the database and for which the presence or absence of the medical relevance to the diagnostic support information is not associated, but whose degree of relevance to the first medical information is equal to or greater than a threshold. A medical information processing device according to any one of claims 2 to 4.

6. The medical information processing device according to claim 5, wherein the second extraction unit extracts a predetermined number of medical information that have a high degree of relevance to the first medical information from among medical information that has no medical relevance to the diagnostic support information or for which the presence or absence of medical relevance to the diagnostic support information has not been associated, as the second medical information.

7. The second extraction unit extracts, as the second medical information, medical information that has no medical relevance to the diagnostic support information or for which the presence or absence of medical relevance is not associated with the diagnostic support information, and whose degree of relevance to the first medical information is equal to or greater than the threshold. The medical information processing device according to claim 5 or 6.

8. The system further includes a third extraction unit that extracts a third medical information related to the second medical information from among the multiple medical information based on the degree of relevance of the medical information to each other and the respective degree of contribution of the multiple medical information. A medical information processing device according to any one of claims 1 to 7.

9. The display control unit determines the display configuration for displaying the diagnostic support information, the first medical information, and the second medical information on the display unit, based on the degree of contribution of the first medical information to the diagnostic support information, the degree of contribution of the second medical information to the diagnostic support information, and the degree of relevance between the first medical information and the second medical information. A medical information processing device according to any one of claims 1 to 8.

10. A medical information processing device that supports the interpretation of the output results of a machine learning model that outputs diagnostic support information, which is information to support the diagnosis of a subject, in response to input of multiple medical information obtained when a subject is examined, Based on the database to which the diagnostic support information and the multiple medical records are associated, and the contribution of each of the multiple medical records to the diagnostic support information when the machine learning model is input to the machine learning model, and the contribution output by the machine learning model in response to the input of the multiple medical records, a first medical record related to the diagnostic support information is extracted from the multiple medical records. When the machine learning model outputs the diagnostic support information, the degree of relevance between the medical information input to the machine learning model is determined by the degree of relevance output by the machine learning model in response to the input of the multiple medical information, and the degree of contribution output by the machine learning model in response to the input of the multiple medical information, and a second medical information related to the first medical information is extracted from the multiple medical information. The diagnostic support information, the first medical information, and the second medical information are displayed on the display unit. Medical information processing method.

11. A computer that assists in interpreting the output results of a machine learning model that outputs diagnostic support information, which is information to support the diagnosis of a subject, in response to inputting multiple medical information obtained when a subject is examined, Based on a database to which the diagnostic support information and the multiple medical records are associated, and the contribution of each of the multiple medical records to the diagnostic support information when the machine learning model is input to the machine learning model, and the contribution output by the machine learning model in response to the input of the multiple medical records, a first medical record related to the diagnostic support information is extracted from the multiple medical records. When the machine learning model outputs the diagnostic support information, the degree of relevance between the medical information input to the machine learning model is determined by the degree of relevance output by the machine learning model in response to the input of the multiple medical information, and the degree of contribution output by the machine learning model in response to the input of the multiple medical information, and the second medical information related to the first medical information is extracted from the multiple medical information. The diagnostic support information, the first medical information, and the second medical information are displayed on the display unit. A program to execute.

Images