JP2020135513A - Medical information processing device, medical information processing system, and medical information processing method - Google Patents

Abstract

To reduce the amount of medical data collected or stored in accordance with a subject by considering estimation accuracy about a health condition of the subject by a learned model.SOLUTION: According to an embodiment, a medical information processing device includes an acquisition part, a processing part, an evaluation part, and a reduction part. The acquisition part acquires a plurality of kinds of medical data of a subject. The processing part generates information on a health condition of the subject by inputting the plurality of kinds of medical data of the subject to a learned model. The evaluation part evaluates an influence degree to estimation accuracy of the learned model about respective kinds of the plurality of kinds of medical data of the subject. The reduction part performs at least one of (1) the reduction of a collection amount from the subject of the next and succeeding kinds of medical data and (2) the reduction of a data size of a kind of medical data of the subject in which an influence degree occupying a storage part is equal to or less than a threshold about the kind of medical data of the subject in which the influence degree is equal to or less than the threshold.SELECTED DRAWING: Figure 6

Description

Embodiments of the present invention relate to a medical information processing device, a medical information processing system, and a medical information processing method.

In recent years, with the spread of IoT technology, it has become possible to collect a large amount of various types of medical data from a subject. For this reason, in the medical and healthcare fields as well, techniques for estimating information on the health condition of a subject using a trained model based on big data have made great progress. Estimating information about a subject's health includes at least one of an estimate of the subject's current health and an estimate of future health.

The medical data that can be used for the trained model includes diagnostic imaging results and test information obtained in the hospital, biological information obtained outside the hospital and at home, and information from wearable sensors and measuring devices attached to the subject. included. As the number of wearable sensors and measuring devices increases, the accuracy of the estimation result by the trained model improves, and the information on the health condition of the subject can be estimated more accurately.

However, as the number of wearable sensors and measuring devices increases, so does the amount of medical data collected, estimated, or stored. As a result, problems such as congestion of network traffic, bloated trained model, and pressure on database resources for storing medical data are caused.

Japanese Unexamined Patent Publication No. 2004-246521 Christopher M. Bishop, "Pattern recognition and machine learning", (USA), 1st Edition, Springer, 2006, P.M. 225-290

An object to be solved by the present invention is to reduce the amount of medical data collected or accumulated according to the subject in consideration of the estimation accuracy of the information on the health condition of the subject by the trained model.

The medical information processing apparatus according to the embodiment includes an acquisition unit, a processing unit, an evaluation unit, and a reduction unit. The acquisition unit acquires a plurality of types of medical data of the subject. The processing unit inputs information on the health status of the subject by inputting multiple types of medical data to the trained model that generates information on the health status of the subject based on the medical data of multiple types of subjects. Generate. The evaluation unit evaluates the degree of influence on the estimation accuracy of the information on the health condition of the subject generated by the trained model for each type of medical data of a plurality of types of subjects. Regarding the medical data of the type of subject whose degree of influence is less than the threshold value, the reduction department will (1) reduce the amount of medical data of the type collected from the subject from the next time onward, and (2) multiple types of collected subjects. At least one of reducing the data size of the medical data of the subject whose degree of influence on the storage unit storing the medical data of the sample is less than the threshold value is performed.

A block diagram showing a configuration example of a medical information processing system including a medical information processing device according to an embodiment. An explanatory diagram showing an example of a data flow during learning of the diagnostic function of the estimation function. Explanatory diagram which shows an example of data flow at the time of operation of a diagnostic function. Explanatory drawing which shows an example of a plurality of trained models. A flowchart showing an example of the type of medical data to be collected, the collection conditions, and the procedure for dynamically setting the trained model during normal operation. A flowchart showing an example of a procedure for identifying low-impact medical data for each subject and reducing the collected data. The explanatory view which shows an example of how to specify the medical data of the low influence degree of the trained model shown in FIG. 3 for each subject. Explanatory drawing which shows an example of how the priority set for each medical data is displayed on the display of a user's information processing terminal based on the cost and merit related to the collection of medical data. A flowchart showing an example of the procedure for simplifying the trained model.

Hereinafter, embodiments of a medical information processing apparatus, a medical information processing system, and a medical information processing method will be described in detail with reference to the drawings.

(Overall system configuration)
FIG. 1 is a block diagram showing a configuration example of a medical information processing system 10 including a medical information processing device 20 according to an embodiment.

The medical information processing system 10 includes a medical information processing device 20 as an example of a server, a correct answer database (correct answer DB) 31, a subject database (subject DB) 32, a plurality of modalities 35, 36, and users such as doctors. The information processing terminal 40 used by the subject, the wearable terminal 51 attached to the subject, the IoT gateway 52, and the information processing terminal 53 used for the subject are included.

The correct answer DB 31 and the subject DB 32 have a configuration including a recording medium that can be read by a processor, such as a magnetic or optical recording medium or a semiconductor memory.

The correct answer DB 31 stores a correct answer data set including a set of verification medical data collected from a plurality of subjects and a correct answer estimation result of information on a health condition. The verification medical data is data collected in the past from a plurality of subjects. The estimation result of the correct answer of the information regarding the health condition is, for example, the definitive diagnosis result regarding the lesion confirmed by extracting the tissue from the person whose verification medical data is collected corresponding to the estimation and diagnosing it.

The subject DB 32 stores medical data collected from the subject for each subject. As the subject DB 32, a PHR (Personal Health Record) or the like can be used. The medical data collected in the subject DB 32 includes various medical data (including medical image data collected in modalities 35 and 36) collected when the subject visits the hospital, and a wearable terminal attached to the subject. It includes medical data collected in 51, or medical data collected by a medical measuring device (such as a sphygmomanometer) possessed by a subject and not shown and which can be connected to a network. The subject DB 32 is an example of a storage unit.

The plurality of modalities 35 and 36 include, for example, an ultrasonic diagnostic apparatus, an X-ray CT (Computed Tomography) apparatus, an MRI (Magnetic Resonance Imaging) apparatus, a SPECT (Single Photon Emission computed Tomography) apparatus, a PET (Positron Emission Tomography) apparatus, and the like. It is composed of nuclear medicine diagnostic equipment and the like.

The information processing terminal 40 is used by a user such as a doctor, and is composed of, for example, a general personal computer or a workstation. Further, the information processing terminal 40 may be composed of a portable information processing terminal such as a smartphone type, a tablet terminal, or a notebook type personal computer. The information processing terminal 40 has at least a display for displaying the estimation result output by the medical information processing apparatus and a processor for generating an image showing the estimation result. The information processing terminal 40 is an example of a client that displays an estimation result output by the medical information processing device 20.

The wearable terminal 51 is a medical data collection device used by the subject to be attached to a part of the body such as an arm, and collects medical data of the subject. The medical data collected by the wearable terminal 51 includes various biochemical information such as blood pressure, heart rate, blood oxygen concentration, blood glucose level, intensity of ultrasonic waves, infrared rays, and the like. When the wearable terminal 51 has a display, the wearable terminal 51 is an example of a client that displays an estimation result output by the medical information processing device 20.

The IoT gateway 52 has a processor, edge-processes medical data transmitted from the wearable terminal 51, and then transmits the medical data to the network 100. The edge process is a process performed to reduce the amount of data transmitted from the wearable terminal 51, and includes, for example, a process of reducing the amount of data transmitted to the network 100 based on the feature amount extracted from the data.

In the present embodiment, the network 100 means a general information communication network using telecommunications technology such as a cloud network, and includes a wireless / wired LAN such as a hospital backbone LAN (Local Area Network), an Internet network, and telephone communication. Includes network, optical fiber communication network, cable communication network and satellite communication network.

Although FIG. 1 shows an example in which the wearable terminal 51 is connected to the network 100 via the IoT gateway 52, it may be directly connected to the network 100 without passing through the IoT gateway 52. When the wearable terminal 51 is connected to the network 100 via the IoT gateway 52, even if the collected data is always transmitted from the wearable terminal 51 in real time, the amount of data flowing through the network 100 is reduced by the edge processing of the IoT gateway 52. be able to.

The information processing terminal 53 is used for a subject, and is composed of a portable information processing terminal such as a smartphone type, a tablet terminal, or a notebook type personal computer. The information processing terminal 53 is an example of a client that displays an estimation result output by the medical information processing device 20.

(Configuration of medical information processing device)
The medical information processing device 20 is composed of a general information processing device such as a personal computer or a workstation, and has an input interface 21, a display 22, a storage circuit 23, a communication circuit 24, and a processing circuit 25.

The input interface 21 of the medical information processing device 20 is composed of general input devices such as a trackball, a switch button, a mouse, a keyboard, and a numeric keypad, and outputs an operation input signal corresponding to the user's operation to the processing circuit 25. .. The display 22 is composed of a general display output device such as a liquid crystal display or an OLED (Organic Light Emitting Diode) display.

The storage circuit 23 has a configuration including, for example, a recording medium readable by a processor such as a RAM (Random Access Memory), a semiconductor memory element such as a flash memory, a hard disk, and an optical disk, and is a program used by the processing circuit 25. And parameter data and other data are stored. A part or all of the program and data in the recording medium of the storage circuit 23 may be downloaded by communication via the network 100, or may be given to the storage circuit 23 via a portable storage medium such as an optical disk. You may.

The communication circuit 24 implements various information communication protocols according to the form of the network 100. The communication circuit 24 connects to other electric devices via the network 100 according to the various protocols.

The processing circuit 25 realizes a function of controlling the medical information processing apparatus 20 in an integrated manner. Further, the processing circuit 25 reads and executes the medical information processing program stored in the storage circuit 23, thereby considering the estimation accuracy of the information on the health state of the subject by the learned model, and depending on the subject. A processor that performs processing to reduce the amount of medical data collected or stored.

The processor of the processing circuit 25 realizes the setting function 251 and the selection function 252, the acquisition function 253, the estimation function 254, the notification function 255, the evaluation function 256, the reduction function 257, the learning function 258, and the abnormality monitoring function 259. Each of these functions is stored in the storage circuit 23 in the form of a program.

In this embodiment, an example in which each function 251-259 is realized by the processing circuit 45 of the medical information processing apparatus 20 will be described, but a part of these functions 251-259 of the medical information processing apparatus 20 or All may be realized by an external device having at least a processor and a storage circuit, independent of the medical information processing device 20, such as a hospital server, cloud console, workstation, etc. connected to the network 100. Further, the medical information processing device 20 may be configured by a plurality of information processing devices connected to each other via a network 100, and each function 251-259 may be appropriately distributed and realized by the plurality of information processing devices.

The setting function 251 is controlled by the reduction function 257 to set a type of medical data to be collected from a subject, and a collection condition including at least one of data accuracy, collection frequency (sampling cycle), and data collection method. To do. Data accuracy includes the number of bits of data. The collection frequency includes information on whether it is evenly spaced or not. The setting of the data collection method can include information on whether it is a direct conversion method or an indirect conversion method, information on whether it is laser light or ultrasonic waves, and the like.

For example, a user such as a doctor selects the type of medical data to be collected and the collection conditions according to the health risk of the subject. Health risk is determined from the subject's age, medical history, and current health status. The setting function 251 sets the type of medical data to be collected and the collection conditions based on the selection result of the doctor.

In addition, the setting function 251 collects the types and collection of medical data according to the health risk of the subject (age, medical history, current health condition, etc.) of the subject without receiving instructions from a user such as a doctor. The conditions may be set automatically.

Further, the setting function 251 may reset the type and collection conditions of the medical data to be collected from the subject based on the information regarding the health condition of the subject generated by the estimation function 254. Information about a subject's health status includes at least one of the subject's current and future health status. For example, a user such as a doctor resets the health risk based on the information on the health condition of the subject notified by the notification function 255 from the notification function 255 based on the estimation result by the estimation function 254. be able to.

In addition, the setting function 251 may reset the type of medical data to be collected and the collection conditions according to changes in the health risk of the subject. The timing of change in health risk includes the timing of occurrence of a condition or event that is not recognized as normal by the health examination process. In addition, a user such as a doctor may predict the transition of the health condition of the subject and set conditions for changing the type of medical data and the collection conditions in advance. In this case, the setting function 251 automatically changes the setting of the medical data type and the collection condition when the condition is satisfied.

The setting function 251 controls a device for collecting medical data from a subject such as a wearable terminal 51 based on the set type of medical data and collection conditions, and collects the medical data of the subject from the device. The setting function 251 is an example of a setting unit.

The selection function 252 selects a trained model that generates information on the health condition of the subject from a plurality of trained models based on the type of medical data and the collection conditions set by the setting function 251. A plurality of trained models are prepared in advance according to the purpose of diagnosis. A plurality of trained models may be prepared according to the type of medical data and the collection conditions. Each trained model outputs information on the health condition of the subject who collected the medical data based on a plurality of types of input medical data. A plurality of trained models will be described later with reference to FIG. 2-4. The selection function 252 is an example of a selection unit.

The acquisition function 253 acquires medical data of a plurality of types of subjects from the subject DB 32, modality 35, 36, wearable terminal 51, and the like. The acquisition function 253 is an example of an acquisition unit.

The estimation function 254 uses machine learning to generate information on the health status of a subject based on medical data of a plurality of types of subjects. As machine learning, deep learning using a multi-layer neural network such as CNN (Convolutional Neural Network) or Convolutional Deep Belief Network (CDBN) can be used.

That is, the estimation function 254 inputs the plurality of types of medical data acquired by the acquisition function 253 into the trained model that generates information on the health condition of the subject based on the medical data of the plurality of types of subjects. Generates information about the health status of the subject. The estimation function 254 includes a plurality of diagnostic functions 254A, 254B, and so on in advance according to the purpose of diagnosis. The details of the operation of the estimation function 254 will be described later with reference to FIG. 2-4. The estimation function 254 is an example of a processing unit.

The notification function 255 notifies at least one of the user and the subject of the information regarding the health condition of the subject generated by the estimation function 254. For example, the notification function 255 displays an image of the health condition of the subject generated by the estimation function 254 on the display of the information processing terminal 40 used by the user or the display of the wearable terminal 51 used by the subject. , Displayed on the display of the information processing terminal 53 or the like. More specifically, the notification function 255 can display a color or shape indicating the current health condition of a subject, or text information, a numerical value, or a graph on a web page on the Internet. In addition, the notification function 255 may cause these terminals to output a voice indicating information on the health condition of the subject.

In addition, the notification function 255 is at least one of the user's information processing terminal 40 and the wearable terminal 51 when the current health condition of the subject is determined to be unhealthy or when a problem is predicted to occur in the near future. Warning information should be transmitted to. At this time, an art message may be sent to call attention or take action. If the current health condition of the subject is determined to be significantly unhealthy and ambulance transport and first aid measures are required, the notification function 255 may take the necessary actions for ambulance transport.

The evaluation function 256 determines the degree of influence on the estimation accuracy of the information on the health state of the subject generated by the trained model for each type of medical data of the plurality of types of subjects. The evaluation function 256 is an example of the evaluation unit. Details of the operation of the evaluation function 256 will be described later with reference to FIGS. 6, 7 and 9. The evaluation function 256 is an example of the evaluation unit.

The reduction function 257 identifies the medical data having a low influence on the estimation accuracy among the plurality of input medical data for the trained model for each subject, and reduces the amount of the medical data collected from the next time onward or the collected subject. The storage capacity is reduced by compressing, deleting, creating a summary, etc. of the medical data stored in the DB 32. This process is performed for each subject. This is because even in the same trained model, the medical data having a low influence on the estimation accuracy may differ depending on the health risk of the subject.

Specifically, the reduction function 257 reduces the amount of medical data collected from the subject of the type of medical data whose degree of influence is less than the threshold value from the next time onward, and (2) the subject. At least one of reducing the data size of the medical data of the type of subject whose degree of influence on the sample DB 32 is less than the threshold value is performed. When reducing the collection amount from the next time onward, the reduction function 257 sets the collection condition for collecting medical data of the type whose influence degree is less than the threshold value from the subject so as to change the collection amount so as to decrease the collection amount. To control. Details of the process of identifying low-impact medical data for each subject and reducing the collected data for each subject will be described later with reference to FIGS. 6 and 7. The reduction function 257 is an example of the reduction unit.

When the trained model generates information on the health condition of the subject, the learning function 258 further trains the trained model by using the generation result and the medical data of the subject used in the generation process, that is, self. Perform learning process.

Further, the learning function 258 reconstructs the trained model by omitting the nodes corresponding to the input data having a low influence on the estimation accuracy among the plurality of input data of the trained model. Details of the reconstruction of the trained model will be described later with reference to FIG. The learning function 258 is an example of the learning unit.

The abnormality monitoring function 259 monitors the traffic volume of medical data, and if there is a delay or loss in transmission / reception of medical data, the abnormality monitoring function 254 and the notification function 255 are notified of the abnormality. When the estimation function 254 receives the notification of abnormality, it is preferable to perform an operation such as not performing the estimation process or discarding the estimation result. Further, when the notification function 255 receives the notification of the abnormality, it is preferable to notify the user or the subject of the information that the abnormality has occurred.

(Estimation function)
FIG. 2 is an explanatory diagram showing an example of a data flow during learning of the diagnostic function 254A of the estimation function 254. As described above, the estimation function 254 includes a plurality of diagnostic functions 254A, 254B, ..., Which are prepared in advance according to the diagnostic purpose. Since the learning method of other diagnostic functions is the same as that of the diagnostic function 254A, the description thereof will be omitted.

The diagnostic function 254A sequentially updates the parameter data 62A by performing deep learning using a large number of correct answer data sets of the correct answer DB31.

The correct answer data set of the diagnostic function 254A includes, for example, the verification medical data groups 711A, 712A, 713A, ..., Which constitute the learning data group 71A, and the definitive diagnosis result 721A corresponding to each verification medical data group. It is composed of a set of a teacher data group 72A composed of 722A, 723A, .... The verification medical data groups 711A, 712A, 713A, ... Are composed of medical data A, B, and C, respectively. The medical data A, B, and C are different types of medical data.

The diagnostic function 254A performs so-called learning in which the parameter data 62A is updated so that the result of processing the training data by the neural network 61A approaches the teacher data each time the correct answer data set is given. Generally, when the rate of change of the parameter data 62A converges within the threshold value, it is determined that the learning is completed. Hereinafter, the parameter data 62A after learning is particularly referred to as learned parameter data 62At. It should be noted that the type of learning data and the type of input data during operation shown in FIG. 3 should match. For example, if the medical data of the learning data is the projection data, the input medical data at the time of operation is also the projection data.

FIG. 3 is an explanatory diagram showing an example of a data flow during operation of the diagnostic function 254A. At the time of operation, the diagnostic function 254A is input with the medical data group 81A consisting of medical data A, B, and C collected from the subject to be estimated, and uses the trained model 60A to input the medical data group 81A to be estimated. Generates an estimation result (hereinafter, referred to as estimated health information) 82A of information on the health condition.

The neural network 61A and the trained parameter data 62t constitute the trained model 60A. As a method of learning this kind and a method of constructing a trained model, various methods such as the method disclosed in Non-Patent Document 1 are known. The neural network 61A is stored in the storage circuit 23 in the form of a program. The learned parameter data 62At may be stored in the storage circuit 23, or may be stored in the storage medium connected to the processing circuit 25 via the network 100. When the trained model 60A (neural network 61A and trained parameter data 62At) is stored in the storage circuit 23, the diagnostic function 254A realized by the processor of the processing circuit 25 reads the trained model 60A from the storage circuit 23. By executing this, the estimated health information 82A of the subject to be estimated is generated based on the medical data group 81A including a plurality of medical data A, B, and C collected from the subject to be estimated.

The trained model 60A may be constructed by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

FIG. 4 is an explanatory diagram showing an example of a plurality of trained models 60A, 60B, .... Similar to the trained model 60A for the diagnostic function 254A, the trained models 60B, ..., 60X, ... Corresponding to other diagnostic functions are constructed in advance for each diagnostic purpose. That is, as shown in FIG. 4, a large number of trained models (for example, 1000) constructed for each diagnostic purpose are prepared in advance.

(Procedure during normal operation)
FIG. 5 is a flowchart showing an example of a procedure for dynamically setting the types and collection conditions of medical data to be collected and the trained model during normal operation. In FIG. 5, reference numerals with numbers attached to S indicate each step of the flowchart.

First, in step S1, the setting function 251 determines the health risk of the subject. Health risk is determined by a user, such as a physician, or automatically, depending on the subject's age, medical history, and current health status.

Next, in step S2, the setting function 251 sets the type of medical data to be collected from the subject and the collection conditions according to the health risk of the subject.

Next, in step S3, the setting function 251 controls a device for collecting medical data from the subject based on the set type of medical data and collection conditions, and collects the medical data of the subject from the device. ..

Further, in step S4, the selection function 252 selects the test subject from the plurality of trained models 60A, 60B, ..., 60X, ..., Based on the set medical data type and collection conditions. Select a trained model that will generate estimated health information. Step S3 and step S4 may be executed before or after, or may be executed in parallel. Here, it is assumed that the trained model 60A is selected.

Next, in step S5, the acquisition function 253 acquires medical data of the set type of subject from the subject DB 32, modality 35, 36, the wearable terminal 51, and the like. Here, the medical data A, B, and C corresponding to the trained model 60A are acquired.

Next, in step S6, the estimation function 254 generates estimated health information of the subject by inputting a plurality of types of medical data acquired by the acquisition function 253 into the selected trained model. Here, the estimated health information 82A is generated by inputting the medical data A, B, and C into the trained model 60A (see FIGS. 3 and 4).

Next, in step S7, the notification function 255 notifies at least one of the user and the subject of the estimated health information generated by the estimation function 254.

Next, in step S8, the setting function 251 resets the health risk of the subject based on the estimated health information generated by the estimation function 254.

For example, consider the case where the estimated health information 82A generated by the trained model 60A is information indicating whether the subject is currently suffering from the disease of the disease name α or the disease of the disease name β. If the estimated health information 82A generated by the trained model 60A is information that the subject suffers from the disease with the disease name β, the health risk of the subject suffers from the disease with the disease name β. It is determined that there is a possibility.

Next, in step S9, the setting function 251 resets the type and collection conditions of the medical data to be collected from the subject based on the reset health risk, and returns to step S3. When the learned model 60B is a model for determining whether or not the disease has the disease name β, in the above example, whether or not the subject has the disease with the disease name β is determined using the learned model 60B. The process returns to step S3 for determination. In this case, when the next step S3 is executed, the medical data A, B, and D corresponding to the trained model 60B are acquired, and when the step S4 is executed, the trained model 60B is selected (see FIG. 4). ..

As a result of the resetting, if there is medical data for which the period until the retesting is long, the notification function 255 may individually notify the user or the subject of the next scheduled test date.

With the above procedure, the type of medical data to be collected, the collection conditions, and the trained model can be dynamically set.

When medical data is collected using the wearable terminal 51, the longer the collection period, the larger the amount of storage required. If health data with 8-bit accuracy is constantly collected at 1 Hz, the daily data amount is 8 [bit] x 1 [Hz] x 60 [s] x 60 [m] x 24 [h] = 86 MB, in one year. A storage capacity of 86MBx365 [d] = 31.5GB, or 31.5GBx80 [y] = 2.5TB in 1980 (lifetime) is required.

In this regard, according to the procedure shown in FIG. 5, the type of medical data to be collected and the collection conditions can be dynamically changed according to the health risk of the subject. Therefore, since the amount of medical data required for estimating the health condition can be appropriately collected, the amount of data collected, processed, and accumulated can be reduced.

It is also clearly inefficient to collect, process, and accumulate medical data that does not contribute to the diagnosis of current health status or prediction of future onset in trained models. According to the procedure shown in FIG. 5, the type and collection conditions of medical data to be collected and the trained model can be dynamically changed according to the health risk of the subject. Therefore, it is possible to reduce the collection of medical data unnecessary for the use of the trained model suitable for the health risk of the subject, and thus the amount of data to be collected, processed, and accumulated can be reduced.

In addition, health risks vary from person to person and change with age. For this reason, fixed operations are clearly inefficient. In this regard, according to the procedure shown in FIG. 5, the type and collection conditions of medical data to be adaptively collected according to the health risk of the subject, and the trained model can be set. Therefore, the amount of data collected, processed, and stored can be reduced.

In addition, depending on the subject, the physical condition may change significantly with treatment such as surgery. Significant changes in physical condition include cases where the condition of organs changes dramatically due to organ transplantation or regenerative medicine, cases where artificial joints are placed, cases where artificial limbs or artificial limbs are attached due to amputation of limbs, etc. Can be mentioned. In this case, new data is input or there is data that will not be needed in the future. In this regard, according to the procedure shown in FIG. 5, the types and conditions of medical data to be collected adaptively according to the health risk of the subject and the trained model can be set, so that the data to be collected, processed, and accumulated can be set. The amount can be reduced.

(Identify low-impact medical data of trained models for each subject and reduce collected data)
FIG. 6 is a flowchart showing an example of a procedure for identifying low-impact medical data for each subject and reducing the collected data. In FIG. 6, reference numerals with numbers attached to S indicate each step of the flowchart.

Further, FIG. 7 is an explanatory diagram showing an example of how the low-influence medical data of the trained model 60A shown in FIG. 3 is specified for each subject.

Even in the same trained model, medical data having a low effect on estimation accuracy may differ depending on the health risk of the subject. Therefore, the procedure shown in FIG. 6 is performed for each subject.

The procedure shown in FIG. 6 is performed, for example, on the trained model used in the previous diagnosis of the subject. Hereinafter, an example of specifying low-impact medical data for each subject will be described for the trained model 60A.

In step S11, the acquisition function 253 acquires a plurality of types of medical data A, B, and C collected from the subject corresponding to the trained model 60A.

Next, in step S12, the evaluation function 256 uses the definitive diagnosis result 83 based on the medical data A, B, C of the subject for each type of the medical data A, B, C of the plurality of subjects. , The degree of influence on the estimation accuracy of the estimated health information 82A of the subject generated by the trained model 60A is obtained (see FIG. 7). Specifically, for each type of medical data A, B, and C of a plurality of types of subjects, a plurality of values different from the actually collected values are sequentially replaced, and changes in the estimated health information 82A are confirmed. ..

Next, in step S13, the evaluation function 256 identifies a type having an influence degree equal to or less than the threshold value.

Next, in step S14, the reduction function 257 reduces the amount of medical data collected from the subject of the type of medical data whose degree of influence is equal to or less than the threshold value (1) from the next time onward, (2). ) At least one of reducing the data size of the medical data of the type of subject whose degree of influence on the subject DB 32 is equal to or less than the threshold value is performed. Reducing the amount of collection includes measures that do not collect at all.

By the above procedure, it is possible to identify low-impact medical data and reduce the collected data for each subject and each trained model.

For example, for a young and relatively healthy subject, the blood pressure contained in multiple input data of a trained model is almost as accurate as the estimated health information, regardless of whether the blood pressure value is intentionally increased or decreased. It does not affect. In this case, when using the trained model for this subject in the next and subsequent tests, the blood pressure will not be collected again from the input medical data, but the previous blood pressure will be used, or the previous blood pressure will be used. Even if the average is used, the estimation accuracy of the estimated health information of the trained model is hardly affected.

As described above, according to the procedure shown in FIG. 6, it is possible to identify low-impact medical data for each subject and for each trained model and reduce the collected data. Therefore, the amount of data collected, estimated, and stored from the subject can be reduced.

(Set the type of medical data to be collected according to cost and merit)
FIG. 8 is an explanatory diagram showing an example of a state in which the priority set for each medical data is displayed on the display of the user's information processing terminal 40 based on the cost and merit of collecting the medical data.

There are three main costs involved in collecting medical data. One is the amount of collected data. For example, when 4DCT imaging is performed to collect medical data, a very large amount of data is collected. The other one is the magnitude of the physical burden on the subject. For example, when a fine needle biopsy is performed to collect medical data, the physical burden on the subject is large. The other is the magnitude of the financial burden on the subject. If high costs are required to collect medical data, the costs are high. Conversely, medical data that can be collected by low-cost, feasible tests such as blood pressure tests is low in cost.

As the cost, at least one of these three may be used, or a weighting coefficient may be set for each of the three, and the final cost may be determined by weight addition averaging or the like.

On the other hand, the merit is the high estimation accuracy of the estimated health information generated by the trained model. Even if it is a trained model that requires high-cost medical data, if the estimation accuracy of the estimated health information to be generated is high, it is concluded that this trained model is selected and high-cost medical data is collected. There may be cases.

Therefore, the setting function 251 requires a large amount of collected data, a large physical burden on the subject, and a large financial burden on the subject when the set type of medical data of the subject is temporarily applied. The type of medical data of the subject to be collected from the subject may be automatically set based on at least one of the above and the estimation accuracy of the estimated health information generated in this case. At this time, the setting function 251 sets the priority for each type of medical data collected from the subject based on the cost and merit, and the priority is set to a predetermined number of medical data or the priority from the highest priority. Medical data above the threshold value may be automatically set as the medical data of the subject to be collected from the subject. Further, as shown in FIG. 8, information on this priority may be presented to the user. In this case, the user can set the type of medical data to be collected based on the priority information.

For example, when the subject comes to the hospital, he / she chooses blood test, puncture biopsy, PET test, and which of the three to choose, based on cost merit, only blood test today. Is easily possible.

(Simplified and reconstructed trained model)
As shown in FIGS. 6 and 7, if the identification of low-impact medical data is repeated for each subject and for each trained model, a large number of correct data will be accumulated depending on the trained model. Being matured. In this case, it is advisable to review the trained model itself and build a simplified trained model.

FIG. 9 is a flowchart showing an example of a procedure for simplifying the trained model. In FIG. 9, reference numerals with numbers attached to S indicate each step of the flowchart.

The learning function 258 sets i = 1 (step S21), deletes the i-th node of the trained model, and generates a temporary trained model i (step S22). Specifically, the i-th input node and the weighting coefficient for the next node connected to this input node are deleted.

Next, the evaluation function 256 obtains the estimation accuracy of the tentative trained model i using a large number of correct answer data sets stored in the correct answer DB 31. Next, the learning function 258 adds 1 to i (step S23). When i is the number of input nodes n or less of the original trained model (YES in step S24), the process returns to step S22.

On the other hand, when i is larger than the number of input nodes n of the original trained model (NO in step S24), the verification is completed for each of the input nodes, so the process proceeds to step S26, and the estimation function 254 is the original. Replace the trained model in with a tentative trained model with the highest estimation accuracy.

By the above procedure, the trained model can be simplified without impairing the estimation accuracy. As a result, from the next time onward, a simplified trained model can be used regardless of the subject. Of course, for the simplified trained model, a process of specifying low-impact input medical data for each subject may be performed again.

The trained engine becomes inefficient as the number of types of data input increases. In this regard, according to the procedure shown in FIG. 9, the storage medium capacity occupied by the learned engine can be significantly reduced without impairing the estimation accuracy.

(Summary generation)
The medical data of the subject stored in the subject DB 32 becomes enormous with the passage of time, and the searchability deteriorates.

Therefore, the reduction function 257 may include a summary function. The summary function may be able to access information in a short time by generating heading information and link information for a part or all of the medical data of the subject stored in the subject DB 32, for example.

In addition, the summary function may generate statistical information indicated by some or all of the medical data of the subject based on a part or all of the medical data of the subject stored in the subject DB 32. Specifically, the summary function of the reduction function 257 processes old medical data collected from a subject to generate statistical information such as maximum, minimum, mean, median, variance and histogram. .. In addition, this statistical information includes feature quantity information of frequency components such as frequency analysis and fluctuation analysis, time and event information such as the date and time of occurrence of attacks and symptoms, period, and organs that have disappeared due to organ removal or transplantation. It may include event information such as the name of an organ, an operation time, a surgical procedure, and statistical information or feature amount information that characterizes the organ or organ. Further, the abnormality notification record by the abnormality monitoring function 259 may be included.

When generating statistical information, the summary feature may delete all corresponding original medical data, or delete other data, leaving the typical original medical data that well describes the summary description. You may. For example, when an arrhythmia occurs, typical original medical data is only a few seconds before and after the occurrence of the arrhythmia and the time of occurrence. In this case, the time when the arrhythmia occurred, the waveform, and how long it lasted (for example, it lasted for a week) can be grasped.

In addition, the original medical data may be deleted by using a summary of medical words (textual information) indicating a health condition suggested by a part or all of the medical data of the subject.

It is preferable to manage the created summary in association with an electronic chart or PHR.

By creating the summary, the searchability of the medical data of the subject stored in the subject DB 32 is improved. In addition, by creating a summary, the collected medical data stored in the subject DB 32 can be significantly reduced.

According to at least one embodiment described above, the amount of medical data collected or accumulated according to the subject can be reduced in consideration of the estimation accuracy regarding the health condition of the subject by the trained model.

In the above embodiment, the term "processor" refers to, for example, a dedicated or general-purpose CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or an application specific integrated circuit (ASIC). It is intended to mean a circuit such as a programmable logic device (for example, a simple programmable logic device (SPLD), a complex programmable logic device (CPLD), and an FPGA). The processor realizes various functions by reading and executing a program stored in a storage medium.

Further, in the above embodiment, an example in which a single processor of the processing circuit realizes each function is shown, but a processing circuit is formed by combining a plurality of independent processors, and each processor realizes each function. May be good. When a plurality of processors are provided, the storage medium for storing the program may be provided individually for each processor, or one storage medium collectively stores the programs corresponding to the functions of all the processors. May be good.

Although some embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention as well as the invention described in the claims and the equivalent scope thereof.

10 Medical information processing system 20 Medical information processing device 22 Display 51 Wearable terminal 81A Medical data group 82A Estimated health information 251 Setting function 252 Selection function 253 Acquisition function 254 Estimate function 255 Notification function 256 Evaluation function 257 Reduction function 258 Learning function

Claims (15)

An acquisition unit that acquires multiple types of medical data for a subject,
By inputting the plurality of types of medical data into a trained model that generates information on the health status of the subject based on the medical data of the plurality of types of the subject, the health status of the subject is related. A processing unit that generates information and
For each type of medical data of the plurality of types of the subject, an evaluation unit for evaluating the degree of influence on the estimation accuracy of the information on the health state of the subject generated by the learned model, and an evaluation unit.
Regarding the medical data of the subject whose degree of influence is equal to or less than the threshold value, (1) reduction of the amount of medical data of the type collected from the subject from the next time onward, (2) the plurality of types of the collected medical data. A reduction unit that reduces the data size of the medical data of the subject whose degree of influence is less than or equal to the threshold value in the storage unit that stores the medical data of the subject, and a reduction unit that performs at least one of them.
Medical information processing device equipped with. A notification unit that notifies at least one of the user and the subject of information on the health condition of the subject generated by the processing unit.
The medical information processing apparatus according to claim 1, further comprising. A setting unit that sets the type of medical data of the subject to be collected from the subject, and the collection conditions including at least one of data accuracy, collection frequency, and data collection method.
With more
The reduction section
For the medical data of the subject whose degree of influence is equal to or less than the threshold value, the collection conditions are changed by controlling the setting unit so that the amount of medical data collected from the next time onward is reduced.
The medical information processing device according to claim 1 or 2. The setting unit
The type and collection conditions of the medical data of the subject to be collected from the subject are reset based on the information regarding the health condition of the subject generated by the processing unit.
The medical information processing device according to claim 3. A plurality of trained models prepared according to the type of medical data and the collection condition based on the type and collection condition of the medical data of the subject to be collected from the subject set by the setting unit. Generates information on the health condition of the subject from a plurality of trained models that output information on the health condition of the subject corresponding to the medical data based on the plurality of types of medical data input. Selection part for selecting a model,
The medical information processing apparatus according to claim 3 or 4, further comprising. The setting unit
A device for collecting medical data from the subject is controlled, and the medical data of the subject is collected from the device based on the set type and collection conditions of the medical data of the subject.
The medical information processing device according to any one of claims 3 to 5. The setting unit
The type and collection conditions of the medical data of the subject to be collected from the subject are set based on the health risk of the subject.
The health risk of the subject is
Determined according to the subject's age, medical history, and current health status.
The medical information processing device according to any one of claims 3 to 6. The setting unit
Further, the size of the amount of collected data, the size of the physical burden on the subject, and the size of the financial burden on the subject, which are required when the set types of medical data of the subject are temporarily applied. The type of medical data of the subject to be collected from the subject is set based on at least one and the estimation accuracy of the information on the health state of the subject generated by the processing unit in this case.
The medical information processing device according to any one of claims 3 to 7. The setting unit
At least one of the magnitude of the collected data, the magnitude of the physical burden, and the magnitude of the financial burden, which is required when the set type of medical data of the subject is temporarily applied, and in this case, Based on the estimation accuracy of the information regarding the health condition of the subject generated by the processing unit, the collection priority is set for each type of collected data, and the information of the priority is presented to the user.
The medical information processing device according to claim 8. When the trained model that generates information about the health condition of the subject generates information about the health condition of the subject, the generation result and the medical data of the subject used for the generation process are used. Learning department to further learn the trained model,
The medical information processing apparatus according to any one of claims 1 to 9, further comprising. The evaluation unit
With respect to the trained model, among the plurality of types of input medical data, the type having an influence on the output health condition information is equal to or less than the threshold value is specified.
The learning unit
Replacing the trained model with a trained model with the specified type of medical data input removed.
The medical information processing device according to claim 10. The reduction section
Based on a part or all of the plurality of types of medical data of the subject stored in the storage unit, the statistical information indicated by the part or all of the medical data is generated.
The medical information processing device according to claim 11. The reduction section
The medical data that is the source of the statistical information is deleted from the storage unit.
The medical information processing device according to claim 12. A medical information processing system equipped with a client and a server.
Either the client or the server
Equipped with an acquisition unit that acquires multiple types of medical data of the subject,
Either the client or the server
By inputting the plurality of types of medical data into a trained model that generates information on the health status of the subject based on the medical data of the plurality of types of the subject, the health status of the subject is related. Equipped with a processing unit that generates information
Either the client or the server
For each type of medical data of the plurality of types of the subject, an evaluation unit for evaluating the degree of influence on the estimation accuracy of the information on the health state of the subject generated by the trained model is provided.
Either the client or the server
Regarding the medical data of the subject whose degree of influence is equal to or less than the threshold value, (1) reduction of the amount of medical data of the type collected from the subject from the next time onward, (2) the plurality of types of the collected medical data. It is provided with a reduction unit that reduces the data size of the medical data of the subject whose degree of influence is less than or equal to the threshold value in the storage unit that stores the medical data of the subject.
The client
A display for presenting to the user information on the health condition of the subject generated by the processing unit is provided.
Medical information processing system. Steps to acquire multiple types of medical data for subjects,
By inputting the plurality of types of medical data into a learned model that generates information on the health state of the subject based on the medical data of the plurality of types of the subject, the health state of the subject is related to the health state of the subject. Steps to generate information and
For each type of medical data of the plurality of types of the subject, a step of evaluating the degree of influence on the estimation accuracy of the information on the health state of the subject generated by the trained model, and
Regarding the medical data of the subject whose degree of influence is equal to or less than the threshold value, (1) reduction of the amount of medical data of the type collected from the subject from the next time onward, (2) the plurality of types of the collected medical data. A step of reducing the data size of the medical data of the subject whose degree of influence on the storage unit storing the medical data of the subject is equal to or less than the threshold value, and a step of performing at least one of the steps.
Medical information processing method having.

Images