JP7639274B2 - Biological information processing device, information processing device, trained model generation device, and program - Google Patents

Description

This invention relates to a biometric information processing device, an information processing device, a trained model generation device, and a program.

There is a technology for monitoring a patient's condition by continuously or at appropriate intervals measuring the patient's body temperature, pulse, respiration, oxygen saturation ( _SpO2 ), etc. Patent Document 1 discloses a technology for displaying these measured values on a display screen and performing a predetermined notification operation when an abnormality exceeding a set standard is detected.

Of these types of biometric information, in the case of biometric data that involves periodic changes over time, such as respiration, what is important is not the measured value at each moment, but the respiration rate and its pattern over a given time span. For pattern recognition, for example, a discrimination program that includes a trained model that has been trained using a machine learning model that utilizes a neural network or the like can be used.

JP 2017-192770 A

However, for safety reasons, it is not possible to use a medical device while further training a machine learning model on diagnostic information during clinical use of the medical device. On the other hand, clinical staff are often very busy, and there is an issue that it is difficult to obtain appropriate separate training data related to biomeasurements.

The object of this invention is to provide a bioinformation processing device, an information processing device, a trained model generation device, and a program that can more easily obtain appropriate training data for medical applications.

In order to achieve the above object, the present invention provides:
a setting means for externally acquiring and setting at least one of information on a waveform type indicated by waveform data relating to a measurement value accompanied by periodic time fluctuation of a subject, the measurement value being acquired from a bio-information measuring device which is a medical device in clinical use, and information on whether an emergency setting is required for the condition of the subject indicated by the waveform data, as corresponding data; and
a storage control means for storing the correspondence data and the waveform data in association with each other;
a switching means for switching between a first mode in which the storage control means stores the corresponding data and a second mode in which the storage control means does not store the corresponding data;
The present invention relates to a biometric information processing device comprising:
The present invention provides a biological information processing apparatus according to claim 1,
The waveform data includes a measurement result that deviates from a predetermined normal range of the subject.
It is characterized by:

The invention described in claim 3 is as follows:
3. The biometric information processing apparatus according to claim 1,
a calculation means for calculating a predetermined feature amount from the waveform data,
The storage control means stores the feature amount in association with the correspondence data.

The present invention provides a biometric information processing apparatus according to claim 3 ,
The apparatus further comprises an estimation means for estimating at least a part of the corresponding data corresponding to the newly acquired waveform data based on the relationship between the stored feature amount and the corresponding data.

The present invention provides a biological information processing apparatus according to claim 4 ,
The relationship is characterized in that it is obtained by learning a machine learning model.

The invention according to claim 6 further comprises:
This information processing device is characterized by comprising a database generation means for generating a database that corresponds first data stored by the memory control means in the bio-information processing device described in any one of claims 1 to 5 and second data including information of the subject obtained by a device other than the bio-information measuring device.

The present invention provides an information processing apparatus according to claim 6 ,
The database generating means generates the database by deleting personal information that identifies the subject.

The invention according to claim 8 further comprises:
The trained model generation device is characterized by comprising a generation means for generating a trained model that includes, as an input, a predetermined feature related to the waveform data contained in the database generated by the information processing device described in claim 6 or 7 , uses at least a portion of the corresponding data as teacher data, and outputs the at least a portion of the corresponding data.

The invention according to claim 9 further comprises:
Computer,
a setting means for externally acquiring and setting at least one of information on a waveform type indicated by waveform data relating to a measurement value accompanied by periodic time fluctuation of a subject, the measurement value being acquired from a bio-information measuring device which is a medical device in clinical use, and information on whether an emergency setting is required for the condition of the subject, the information being indicated by the waveform data;
a storage control means for storing the correspondence data and the waveform data in association with each other;
a switching means for switching between a first mode in which the storage control means stores the corresponding data and a second mode in which the storage control means does not store the corresponding data;
The program is characterized in that it functions as

The present invention has the effect of making it easier to obtain appropriate training data for medical applications.

1 is a diagram showing a configuration of a medical information system according to a first embodiment. FIG. 2 is a block diagram showing the functional configuration of a terminal device and a server device according to the first embodiment. FIG. 13 is a diagram showing the contents of measurement data in a clinical trial mode. FIG. 13 is a diagram showing a display example on a display screen of a terminal device. 10 is a flowchart showing a control procedure of a measurement display control process executed by the terminal device. 10 is a flowchart showing a control procedure for processing related to learning of a machine learning model executed by a server device. FIG. 11 is a block diagram showing the functional configuration of a terminal device and a server device according to a second embodiment. FIG. 11 is a diagram illustrating a display by a terminal device according to a second embodiment. 10 is a flowchart showing a control procedure of a measurement display control process executed by a terminal device of a second embodiment.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
[First embodiment]
First, a biological information processing device, an information processing device, and a trained model generation device according to the first embodiment will be described.
1 is a diagram showing a configuration of a medical information system 1 including a biological information processing device and an information processing device according to the present embodiment. In the present embodiment, the information processing device also serves as a trained model generating device.

The medical information system 1 includes a biosensor 10 (biometric information measuring device), a terminal device 20 (biometric information processing device, computer), a server device 30 (information processing device, trained model generating device), a database device 40, a network device 50, etc.

The biosensor 10 continuously measures predetermined biodata of the patient P (subject) and transmits the measurement data to the terminal device 20. The biodata here includes, for example, _SpO2 (a measurement value for calculating _SpO2 , which may be, for example, the intensity of infrared light received), pulse, and data that periodically varies over time according to the respiratory state (variation data, waveform data). The variation data includes, for example, a flow sensor that directly measures the flow of inhaled and exhaled air through a mask, an acceleration sensor that measures chest movement, an expansion/contraction sensor, a microwave sensor, and a sensor that measures the blood flow state (such as the state of hemoglobin). The periodicity referred to here includes cases in which the length, amplitude, and waveform of each period change depending on the condition of the patient P.

The terminal device 20 receives and processes the measurement data from the biosensor 10, displays the measurement results on a display screen, and transmits the measurement results to the server device 30. The measurement results include the respiratory rate per predetermined time (e.g., one minute) in addition to _SpO2 . The terminal device 20 includes, for example, a mobile terminal placed at the bedside of the patient, a mobile terminal, etc. Note that, in this example, the biosensor 10 and the terminal device 20 are illustrated as directly transmitting and receiving data by wireless communication (e.g., Bluetooth (registered trademark) etc.), but the measurement data may be acquired by a wired connection, or may be transmitted and received via a LAN (Local Area Network) via a network device 50. In addition, communication with the server device 30 may also be performed by connecting a cable to the network device 50.

The server device 30 can receive the measurement results of each patient from the terminal device 20 and display them as a list. The server device 30 can also collect, store, and analyze the measurement results. Note that there may be multiple server devices 30 (e.g., on each floor of a hospital ward) for one database device 40.

The database device 40 has a patient information database 41 (see FIG. 2) and stores and holds patient information. The patient information includes personal information of the patient (such as name, age, and sex) associated with the patient's identification information, the results of various tests, and diagnostic information. The patient information also includes the measurement results obtained from the terminal device 20.

The network device 50 controls communication between electronic devices. The network device 50 is, for example, a hub or a router, and multiple devices may be provided as necessary (for example, a hub in each hospital room). Here, the network device 50 controls communication with the terminal device 20 via wireless LAN and communication with the server device 30 and the database device 40 via LAN using cables, but is not limited to this. The router may also be connected to an external network (the Internet).

Figure 2 is a block diagram showing the functional configuration of the terminal device 20 and the server device 30. Note that the network device 50 between the terminal device 20 and the server device 30 is omitted here.

The terminal device 20 includes a control unit 21, a memory unit 22, a communication unit 23, a display unit 24, an operation reception unit 25, and an alarm operation unit 26.

The control unit 21 is a processor that has a CPU (Central Processing Unit) and the like, and controls the overall operation of each part of the terminal device 20. The control unit 21 processes the measurement data acquired from the biosensor 10 to obtain the measurement results, and also performs pre-processing thereof, such as removing noise and making adjustments when no data is acquired.

The storage unit 22 includes a RAM (Random Access Memory) and a non-volatile storage medium, such as a HDD (Hard Disk Drive) or flash memory. The RAM provides the CPU with working memory space and stores temporary data. The non-volatile storage medium stores the program 221 and various setting data. The storage unit 22 also stores measurement/processing data 222 and selection data 223 accepted by the operation acceptance unit 25. The measurement/processing data 222 and selection data 223 may be stored in the RAM or in the non-volatile storage medium depending on the period for which they are to be stored. The details of these will be described later.

The communication unit 23 controls communication with other electronic devices. The communication unit 23 includes, for example, a driver for direct communication with the biosensor 10 via Bluetooth or the like, and a network card for LAN communication.

The display unit 24 has a display screen and performs various displays based on the control of the control unit 21. The display contents include the measurement results based on the measurement data from the biosensor 10 as described above. The display screen is not particularly limited, but is, for example, a liquid crystal display screen (LCD).

The operation reception unit 25 receives an input operation from the outside and outputs it to the control unit 21 as an input signal. The operation reception unit 25 has, for example, a touch panel located over the display screen, and identifies information on the position where a touch operation has been detected and/or performed, and outputs it to the control unit 21. The operation reception unit 25 may additionally have a keyboard, a pointing device such as a mouse, a push button switch, a rotary switch, and the like. Furthermore, in the case where the terminal device 20 is a terminal installed next to a bed, the operation reception device may be located on a remote controller, and an operation input signal may be transmitted to the main body of the terminal device 20 by infrared rays, or the like.

The alarm operation unit 26 performs a predetermined alarm operation based on the control of the control unit 21. The alarm operation unit 26 may have, for example, an LED (Light Emitting Diode) lamp and may be capable of lighting or blinking. The alarm operation unit 26 may also have LED lamps of multiple colors. Alternatively, the alarm operation unit 26 may have a configuration that outputs a predetermined beep sound. The beep sound may be emitted by various well-known configurations, and may be, for example, generated directly by the operation of a semiconductor element, or may be generated by converting an electrical signal into sound and outputting it from a speaker or the like.

The server device 30 includes a control unit 31, a memory unit 32, a communication unit 33, a display unit 34, an operation reception unit 35, and an alarm operation unit 36.

The control unit 31 is a processor having a CPU (Central Processing Unit) and the like, and controls the overall operation of each part of the server device 30. In addition to displaying a list of measurement results acquired from the terminal device 20 and performing an alarm operation based on the displayed content, the control unit 31 may also be able to access the database device 40 to acquire and display patient information. The control unit 31 also integrates the measurement results and the patient information to generate learning data 321. The control unit 31 also performs a process of training a machine learning model 322 using the learning data 321.

The memory unit 32 stores and holds various data. The memory unit 32 has a RAM and a non-volatile storage medium (such as a HDD or flash memory). The RAM provides working memory space for the CPU of the control unit 31 and stores temporary data. The non-volatile storage medium stores and holds learning data 321, a machine learning model 322 (including various parameters), and a program 323.

The learning data 321 includes measurement/analysis data 3211, selection data 3212, and diagnosis data 3213. These will be described later.

The machine learning model 322 has an algorithm and data structure already defined, and the various parameters may be authenticated settings, or may be some initial values or values in the middle of learning.

The program 323 is called and executed by the control unit 31 to perform various control processes. The control processes include a learning data generation process and a learning control process, which will be described later.

The communication unit 33 controls communication with other electronic devices. The communication unit 33 includes, for example, a network card for LAN communication with other electronic devices.

The display unit 34 has a display screen and performs various displays based on the control of the control unit 31. The contents that can be displayed include a list display of the measurement results acquired from each terminal device 20 as described above. The display screen is not particularly limited, but is, for example, a liquid crystal display screen (LCD).

The operation reception unit 35 receives input operations from the outside and outputs them to the control unit 31 as input signals. The operation reception unit 35 has, for example, a pointing device such as a keyboard and a mouse. The operation reception unit 35 may also have a touch panel positioned over the display screen, and identify information on the position where a touch operation has been detected and/or performed and output the information to the control unit 31. Alternatively, the operation reception unit 35 may further have a push button switch, a rotary switch, etc.

The alarm operation unit 36 performs a predetermined alarm operation based on the control of the control unit 31. The alarm operation unit 36 may have, for example, an LED (Light Emitting Diode) lamp and may be capable of lighting or blinking. The alarm operation unit 36 may also have LED lamps of multiple colors. Alternatively, the alarm operation unit 36 may have a configuration that outputs a predetermined beep sound. The beep sound may be emitted by various well-known configurations, and may be, for example, generated directly by the operation of a semiconductor element, or may be generated by converting an electrical signal into sound and outputting it from a speaker or the like.

Next, the operation of acquiring measurement data in the medical information system 1 of this embodiment will be described. In this embodiment, as described above, biodata accompanied by time-based changes measured by the biosensor 10, particularly fluctuation data indicating changes in the respiratory state (e.g., thoracic position/amount of expansion/contraction of the chest), is processed, and the processing result is output as a measurement result. The measurement result includes the respiratory rate per predetermined time (1 minute) (i.e., the number of cycles of occurrence of changes according to breathing). Note that instead of simply counting the respiratory rate for the most recent minute, the respiratory rate may be calculated by weighting with an appropriate weighting function so that the most recent respiratory cycle is more significantly reflected. In addition, since a one-minute period is usually longer than the update interval of the display screen of the display unit 24, the counting period of the displayed respiratory rate is set to move with some overlapping portions.

Depending on the type of disease of the patient, breathing may be regular, normal, or the frequency or amplitude may be greater than normal (tachypnea, hyperpnea, tachypnea, Kussmaul) or smaller (bradypnea, hypopnea, hypopnea), or the amplitude of breathing may change irregularly (Cheyne-Stokes, Biot). These breathing patterns (waveform types) may vary depending on the patient and their condition, but can be identified by analyzing the pattern of the fluctuating waveform. Here, the type is identified using a trained machine learning model, but at present, there may be cases where the type cannot be accurately identified depending on the waveform.

In addition, continuous acquisition of biometric data is also intended to quickly detect changes in conditions that require an emergency and to inform the person in charge. If a measurement result deviates from a predetermined normal range, an emergency notification operation is performed, so that the person in charge, such as a nurse or doctor, can be informed quickly even if they are away from the patient due to other work. However, whether the occurrence of abnormal breathing requires an emergency, that is, whether a setting for performing an emergency notification operation (emergency setting) is necessary, varies depending on a combination of many aspects, such as not only the respiratory rate but also the respiratory pattern (waveform type), other measurement results, and the patient's illness. Machine learning models are also effective for judgments (estimations) with such a large number of variables, but currently, the correct answer is not always output, and the final decision is left to the person in charge.

The output of a machine learning model can be improved by repeated appropriate learning. However, due to various factors such as overlearning and inaccurate training data, the accuracy may actually decrease depending on the training data added. For this reason, there is a risk in continuing training a machine learning model for medical use after it has been officially approved as a medical device. On the other hand, in order to generate an improved machine learning model, it is essential to acquire new training data, that is, additional correct answers (trainer data) for biomeasurement data and waveform types. In the medical information system 1 of this embodiment, while normal data acquisition, judgment, and display are possible in the normal mode (second mode), it is also possible to switch to a clinical trial mode (first mode) in which training data (corresponding data) can be easily acquired when a nurse or other person in charge can input correct answer data.

FIG. 3 is a diagram showing the contents of the measurement data in the clinical trial mode.
FIG. 3A shows data of the measurement results stored in the terminal device 20. The measurement results are stored as measurement/processing data 222 in association with the set patient identification information and measurement date and time. Here, the measurement/processing data 222 includes respiratory rate and _SpO2 . In addition to these, in the clinical trial mode, the results of the person in charge's real-time judgment, such as by visually observing the patient, are acquired as input data from the outside via the operation reception unit 25, etc., and the selection results of the breathing pattern (waveform type) and the necessity of emergency notification (emergency setting) are set and stored as selection data 223 (corresponding data) corresponding to the waveform data in association with the patient identification information and the measurement date and time. In addition, the measurement data for the period corresponding to the selection data 223 is preprocessed as necessary, and similarly stored in association with the patient identification information and the measurement date and time.

The respiratory waveform data itself may be associated with patient identification information, and if the period corresponding to each selected data 223 can be identified, it may be stored separately and later analyzed for each corresponding period. In normal mode, there is no need to store the waveform data, and the measurement result data sequentially transmitted to the server device 30 may be deleted after an appropriate time has passed. Data such as body temperature and blood pressure measured at appropriate times by a nurse or other staff member may also be input directly into the terminal device 20 or via an input operation by the staff member.

FIG. 3B shows medical information data of the patient stored in the patient information database 41 of the database device 40. In addition to the measurement results of each patient, various test results, diagnosis information by doctors, and findings are recorded in the database device 40 in association with the identification information of each patient. The identification information of the patient may include personal information such as age (date of birth), sex, hospitalization/visit history (e.g., medical department, doctor in charge, period, hospital room), and other information reported by the patient (e.g., smoking and drinking information). Measurement data such as body temperature and blood pressure may be acquired by inputting it to the terminal device 20 as described above, or may be directly input to the server device 30 by the person in charge. Measurement results based on measurement data by the biosensor 10 such as respiratory rate and _SpO2 may be acquired at the timing when the body temperature and blood pressure were measured, or data for a predetermined time before and after the measurement may be averaged. Alternatively, the measurement results by the biosensor 10 are acquired periodically (intervals may be changed depending on whether or not an abnormality is detected) and/or in response to an explicit acquisition operation, and if other measurements are performed within the acquired period, these results may be integrated, and if no other measurements are performed, they may be stored as no other measurements. The medical examination record is the diagnosis by the doctor, and if there is no change, the most recent diagnosis may be acquired continuously. In addition, in the original data, it is not necessary to store data of all items by date and time in a matrix form, and each data acquired at different frequencies may be stored separately so that it can be associated with each patient and date and time. In addition, the data stored in the database device 40 is the same in the clinical trial mode and the normal mode. In other words, data used only in clinical trials is not stored in the database device 40.

3(c) is an example of the learning data 321 stored in the storage unit 32 of the server device 30. In the learning data, the data acquired in the clinical trial mode by the terminal device 20 (a part of the measurement/analysis data 3211 corresponding to the measurement/processing data 222 and the selection data 223, and the selection data 3212) and the medical information data stored in the patient information database 41 of the database device 40 (a part of the measurement/analysis data 3211 and the diagnosis data 3213) are integrated and stored. The waveform data in the data acquired from the terminal device 20 is analyzed and processed to calculate one or more feature amounts to be inputted into the machine learning model and used, and are included in and stored in the measurement/analysis data 3211. Examples of the feature amount include, but are not limited to, a simple amplitude, its rate of change, and change period, a frequency, its rate of change, and change period, and a total inhalation volume/exhalation volume for a predetermined time. Various other feature amounts may be used. The feature amount may be calculated by the terminal device 20, and simply acquired from the terminal device 20 and stored. Furthermore, in this learning data 321, information that can identify or specify an individual (personal information) may be deleted. Furthermore, data for multiple patients may be integrated and stored.

Figure 4 shows an example of a display on the display screen of the terminal device 20. Figure 4(a) shows an example of a display in normal mode, and Figure 4(b) shows an example of a display in clinical trial mode.

In FIG. 4(a), the most recent measurement results, such as respiratory rate, pulse rate, and _SpO2 , are displayed together with the patient's identification information and the current date and time. If a measurement result that falls outside the reference range Dr ( _SpO2 is the lower limit) is obtained based on the estimation result of a mechanically determined or used trained model, a predetermined notification operation is performed to notify the user of the abnormality. Here, a display example is shown in which all measurement values are within the reference range Dr. A switching indicator Df1 for the clinical trial mode is displayed in the lower right, and the clinical trial mode can be switched to with one touch by touch operation or the like. In addition, since informed consent and predetermined approval from the patient or the patient's family are required when conducting a clinical trial, a dialogue (e.g., an operation button for confirming that consent and approval have been obtained) calling attention may be displayed on the display screen when switching to the clinical trial mode.

In the clinical trial mode display shown in FIG. 4(b), the display is switched to something different from the normal mode so that a user, such as a person in charge, can clearly see that the display is different from the normal mode. Although not shown here, the words "clinical trial mode" or the like may be displayed. In addition to the numerical measurement results, the clinical trial mode also displays a waveform Dw that indicates the time variation of the measured respiratory state.

Here, an example of the display when the respiratory rate exceeds the upper limit of the reference range Dr is shown, with the respiratory rate Dma highlighted and a warning sign Ds displayed. In addition to or instead of the warning sign Ds, an emergency alert may be issued by audio output or by turning on/flashing an LED lamp of a specified color.

The display of this clinical trial mode also includes a display for accepting input operations from the user. Here, a selection screen Dc1 for selecting whether or not an emergency notification (alarm) is required, and a selection screen Dc2 for selecting a breathing pattern (waveform type) are displayed, with the options displayed as a list. When a selection made by a staff member directly observing the patient's condition is input by touch operation, the selection is stored in the terminal device 20 as selection data 223 in association with the patient identification information and the current date and time. A switch indicator Df2 for normal mode is displayed in the lower right, and the display can be returned to normal mode with a single touch operation.

Figure 5 is a flowchart showing the control procedure by the control unit 21 of the measurement display control process executed by the terminal device 20. For example, in the case of a dedicated device, this measurement display control process is started by automatically starting the program 221 when power supply starts, and in the case of a general-purpose terminal, it is started by the user starting the program 221, and is executed continuously until the power supply is turned off or the program 221 is terminated.

When the measurement display control process is started, the control unit 21 (CPU) performs various initial settings, such as acquiring patient information, starting the display screen, and establishing a communication connection with the biosensor 10, and starts processing the data acquired from the biosensor 10 (step S101). The control unit 21 is initially set to normal mode and performs data acquisition and display control (step S102).

The control unit 21 determines whether or not the clinical trial mode is in effect (step S103). If it is determined that the clinical trial mode is not in effect ("YES" in step S103), the control unit 21 acquires the latest results for each unit time and causes the display unit 24 to display the results in the normal mode described above (step S121). The control unit 21 also transmits the measurement results to the server device 30. Then, the process of the control unit 21 proceeds to step S112.

If it is determined that the clinical trial mode is selected ("YES" in step S103), the control unit 21 acquires the latest results for each unit time and causes the display unit 24 to display the results in the clinical trial mode described above (step S104). The control unit 21 also transmits the measurement results to the server device 30. The control unit 21 determines whether or not an input has been made by a selection operation related to the necessity of an emergency notification (the presence or absence of an emergency) (step S105). If it is determined that an input has been made ("YES" in step S105), the control unit 21 sets the input data of necessity as the selection data 223 and stores it in association with the current date and time (and the patient identification information) (step S106). Then, the process of the control unit 21 proceeds to step S107. If it is determined that no selection operation has been made ("NO" in step S105), the process of the control unit 21 proceeds to step S107.

When the process proceeds to step S107, the control unit 21 determines whether or not a breathing pattern has been input by a selection operation (step S107). If it is determined that a breathing pattern has been input ("YES" in step S107), the control unit 21 sets the input data of the breathing pattern as the selection data 223 and stores it in association with the current date and time (and the patient identification information) (step S108). Then, the process proceeds to step S109. If it is determined that a breathing pattern has not been input ("NO" in step S107), the process proceeds to step S109.
The processes in steps S105 to S108 constitute the setting means and storage control means of this embodiment.

When the process proceeds to step S109, the control unit 21 determines whether or not biometric data for a predetermined time or more is stored in the storage unit 22 (step S109). The biometric data here includes the measurement results after processing and the selection data 223. The determination criterion may not be the time unit, but the amount of data. If it is determined that biometric data for a predetermined time or more is not stored ("NO" in step S109), the process of the control unit 21 proceeds to step S112. If it is determined that biometric data for a predetermined time or more is stored ("YES" in step S109), the control unit 21 determines whether or not the current communication volume is equal to or less than a predetermined standard (step S110). If it is determined that the current communication volume is not equal to or less than the predetermined standard (exceeds the standard) ("NO" in step S110), the process of the control unit 21 proceeds to step S112. That is, if there is a lot of normal communication, the transmission of biometric data for learning is suspended.

If it is determined that the communication volume is equal to or less than the predetermined standard ("YES" in step S110), the control unit 21 transmits the biometric data for learning stored in the storage unit 22 to the server device 30 (step S111). The control unit 21 may delete the transmitted biometric data from the storage unit 22. Then, the process of the control unit 21 proceeds to step S112.

When the process proceeds to step S112, the control unit 21 determines whether or not there has been an input related to switching the mode setting (step S112). If it is determined that there has been an input ("YES" in step S112), the control unit 21 switches and sets the mode between the normal mode and the clinical trial mode in accordance with the input (step S113; switching means). Then, the process of the control unit 21 returns to step S103. If it is determined that there has been no input ("NO" in step S112), the process of the control unit 21 returns to step S103.

Figure 6 is a flowchart showing the control procedure by the control unit 31 of the process related to learning the machine learning model executed by the server device 30.

FIG. 6A shows a control procedure for the learning data generation process.
This process may be started periodically at a predetermined time (for example, at night), or may be started based on a predetermined input operation by the user, for example.

When the learning data generation process is started, the control unit 31 (CPU) acquires one piece of biomeasurement data (first data) acquired from the terminal device 20 (step S201). The control unit 31 acquires waveform data for a period corresponding to the biomeasurement data (step S202). The control unit 31 analyzes the acquired waveform data and calculates features that will be input for machine learning (step S203).

Based on the patient identification information and date and time information of the biomeasurement data, the control unit 31 acquires patient information (information of the subject, second data) for the corresponding date and time of the patient from the database device 40 (step S204). The control unit 31 associates the acquired biomeasurement data, features, other measurement results, and diagnostic information with the patient information and outputs it as a set of learning data (step S205; database generation means). The control unit 31 determines whether all biomeasurement data has been acquired (step S206). If it is determined that some biomeasurement data has not been acquired ("NO" in step S206), the processing of the control unit 31 returns to step S201. If it is determined that all biomeasurement data has been acquired ("YES" in step S206), the control unit 31 ends the database generation of the learning data and ends the learning data generation process.

FIG. 6( b ) shows a control procedure for learning control of the machine learning model.
The learning control process may be automatically started immediately after the learning data generation process is completed, or may be started when a start-up operation is input by the user.

The control unit 31 (CPU) inputs each set of learning data 321 into the machine learning model 322 and obtains the output (step S251). The control unit 31 compares the output result with the correct answer data (teacher data) (step S252).

The control unit 31 causes feedback based on the comparison result to be performed on each parameter of the machine learning model 322 (step S253). The feedback may be performed by a conventionally known method according to the algorithm of the machine learning model. The control unit 31 determines whether or not all data pairs of the learning data 321 have been input (step S254). If it is determined that there is a data pair that has not been input ("NO" in step S254), the processing of the control unit 31 returns to step S251. If it is determined that all data pairs have been input ("YES" in step S254), the control unit 31 ends the learning control processing.

[Second embodiment]
Next, a biological information processing apparatus and an information processing apparatus according to a second embodiment will be described.
FIG. 7 is a block diagram showing the functional configuration of a terminal device 20 which is a biometric information processing device and a server device 30 which is an information processing device according to the second embodiment.

In this second embodiment, the terminal device 20 and the server device 30 have the same configuration, except that the trained model 224 is stored in the memory unit 22 of the terminal device 20. Identical components are given the same reference numerals and descriptions thereof are omitted.

The trained model 224 is trained by the machine learning model 322 of the server device 30. In other words, by inputting data such as measurement results, the breathing pattern and the emergency alert level are output. Note that the latest parameters related to the operation of the trained model 224 may be obtained from the server device 30 as appropriate and updated.

Figure 8 is a diagram explaining the display by the terminal device 20 in this embodiment.

8A, in the terminal device 20 of this embodiment, the reference range Dr for the respiratory rate is variable, and the setting can be changed according to the input of other parameters to the trained model 224. The setting may be changed only once at the start of the clinical trial mode, or may be changed in real time. Also, a display Dp of the estimated result of the respiratory pattern is shown. The other displays are the same as those in the clinical trial mode in the terminal device 20 of the first embodiment. The request for input operation to the selection screens Dc1 and Dc2 is also performed in the same way regardless of the display content of the estimated result. Information related to the difference between the estimation result based on the trained model 224 and the judgment result input by the person in charge may be transmitted to the server device 30 in the same way as the training data, and used for performance evaluation of the trained model 224. In this case, the server device 30 may generate and store performance evaluation data separately from the training data 321. The performance evaluation data may be quantitatively evaluated based on another analysis program.

In other words, in the terminal device 20 of this embodiment, the display in the clinical trial mode can be performed by a person in charge who can appropriately judge the condition of the patient, regardless of the display content. Note that the highlighting of the respiratory rate Dma and the display of the warning sign Ds may be performed based on the standard range Dr in the normal mode based on the standard in the normal mode, and the variable standard may simply be displayed side by side, or a warning sign based on the variable standard may be determined separately and displayed.

The estimate of whether an emergency alert is necessary may be determined based on the percentage of alarms input by the person in charge that were necessary, depending on various other parameter conditions.

As shown in FIG. 8(b), the need for an emergency alert increases when the respiratory rate is too low or too high. However, the reference range Dr changes depending on the combination with the respiratory pattern, other measurement results, and the patient's illness, so the trained model sets the reference range Dr according to each of these combinations. Note that the machine learning model 322 that performs this level of learning does not necessarily have to use a neural network or the like.

Figure 9 is a flowchart showing the control procedure by the control unit 21 of the measurement display control process executed by the terminal device 20 of this embodiment. This measurement display control process differs from the measurement control process executed by the terminal device 20 of the first embodiment only in that steps S131 to S133 have been added and steps S106 and S108 have been replaced by steps S106a and S108a, with the rest of the process being the same. The same process contents are given the same reference numerals and detailed explanations will be omitted.

After the processing of step S101, the control unit 21 (CPU) acquires the latest data of each parameter of the trained model 224 from the server device 30 as the latest data for estimation (step S131). Then, the processing of the control unit 21 proceeds to step S102.

When it is determined in the determination process of step S103 that the clinical trial mode is selected ("YES" in step S103), the control unit 21 analyzes the breathing waveform data acquired from the biosensor 10 to calculate the feature amount (step S132; calculation means). The control unit 21 inputs a predetermined set of data including the feature amount into the trained model 224, and estimates the breathing pattern according to the relationship based on the learning of the machine learning model (step S133; estimation means). Note that when data such as body temperature and blood pressure are required for the estimation, the most recent measured value may be used. In this case, the most recent measured value of body temperature and blood pressure may be input to the terminal device 20 and held until the next measurement. Then, the process of the control unit 21 proceeds to step S104.

When the process of step S105 branches to "YES", the control unit 21 associates and stores the biomeasurement data, the date and time information, and the estimation result of whether or not an emergency setting is required by the trained model 224 at this date and time (according to the relationship obtained by learning the machine learning model) (step S106a). Then, the process of the control unit 21 proceeds to step S107. Also, when the judgment process of step S107 branches to "YES", the control unit 21 associates and stores the biomeasurement data, the date and time information, and the estimation result of the breathing pattern by the trained model 224 at this date and time (step S108a). Then, the process of the control unit 21 proceeds to step S109.

As described above, the terminal device 20, which is the bioinformation processing device of this embodiment, includes the control unit 21. The control unit 21, as a setting means, acquires and sets at least one of the breathing pattern (information on waveform type) indicated by waveform data relating to time variations in the measured value of the patient (subject) acquired from the biosensor 10 and information on whether emergency setting is required for the patient's condition indicated by the waveform data, as selection data 223 from the outside, here, from the operation acceptance unit 25. In addition, the control unit 21, as a storage control means, stores the selection data 223 in the storage unit 22 in association with the waveform data.
In this way, in a medical setting where it is difficult to perform measurements solely for clinical trials, information such as breathing patterns and the need for emergency notification input by a person in charge when possible can be easily acquired and stored in combination with biomeasurement data while easily checking normal data. Therefore, the terminal device 20 can more easily acquire appropriate medical learning data.

Moreover, the control unit 21 serves as a calculation means for calculating a predetermined feature amount that characterizes the waveform pattern from the waveform data, and serves as a storage control means for storing the feature amount in association with the corresponding data.
By calculating the feature amounts in advance in the terminal device 20, data correspondence can be facilitated, and the processing load of the server device 30 can be reduced.

Furthermore, the control unit 21, as an estimation means, estimates at least a part of the breathing pattern and the need for an emergency alert according to the waveform data newly acquired from the biosensor 10, based on the relationship between the stored feature amount and the selected data 223. In this way, if the person in charge, who is the user of the terminal device 20, can judge the breathing pattern and urgency by himself, and there is no problem in switching to the clinical trial mode, the estimation results by the machine learning model being trained may also be displayed. By acquiring these results together with the judgment results of the person in charge, evaluation data for the machine learning model can also be easily obtained.

The above relationship was obtained by learning the machine learning model. In other words, as described above, by acquiring this result together with the judgment result of the person in charge, evaluation data for the machine learning model can be easily obtained.

The control unit 21 also switches between a clinical trial mode in which the selected data 223 is stored as a switching means and a normal mode in which the selected data 223 is not stored as a storage control means. In this way, the control unit 21 can easily switch between a state in which a clinical trial is being conducted and a state in which a clinical trial is not being conducted in the terminal device 20, so that the clinical trial mode can be easily switched to only when there is a skilled person in charge of providing learning data available or when it is necessary, without significantly affecting normal operations.

The server device 30, which is the information processing device of this embodiment, includes a control unit 31. The control unit 31 serves as a database generating means and generates a database in which the bioinformation data stored in the terminal device 20 is associated with patient information data including measurement data of the patient acquired by a device other than the biosensor 10.
In this way, the server device 30 can obtain a large amount of learning data by generating a list by further associating the biological information data generated by each terminal device 20 with various patient data. If the server device 30 can properly access the necessary data, it can easily integrate the data, so that it can generate a large amount of necessary learning data without imposing a burden on medical personnel or with a small burden.

Furthermore, the control unit 31, as a database generation means, generates a database by deleting personal information that identifies the patient. Since the personal information itself is not necessary for the learning data, the amount of data can be reduced by deleting unnecessary data. Furthermore, since deleting personal information makes it possible to take the learning data outside the hospital, it is also possible to outsource the learning process to a third party. Furthermore, since it is possible to integrate with data from other hospitals, it is possible to obtain a large amount of learning data more efficiently and train the machine learning model 322.

In addition, the server device 30 as a trained model generation device of this embodiment has a control unit 31 as a generation means, which includes as input predetermined features related to waveform data contained in the database generated as the above-mentioned information processing device, and generates a trained model using at least a portion of the selected data 223 as teacher data and outputting the at least a portion of the selected data 223.
In this way, a trained model is generated from a database that consolidates biometric data easily obtained during normal work hours and the correct answer data for the measurement results. This makes it possible to efficiently and appropriately improve and develop machine learning models using a variety of useful training data.

In addition, the program 221 of this embodiment causes the terminal device 20 (computer) to function as a setting means for externally acquiring and setting at least one of information on waveform type (breathing pattern) indicated by waveform data relating to the time variation of the patient's measurement value acquired from the biosensor 10, and information on the need for emergency settings for the patient's condition (status) indicated by the waveform data, as selection data 223.
By installing and running such a program 221 on the terminal device 20, conventional medical operations can be continued without hindrance, while learning data can be easily collected without the need for separate hardware devices, thereby reducing the time and financial burden on medical personnel and medical corporations.

The present invention is not limited to the above-described embodiment, but may be modified in various ways.
For example, in the above embodiment, both the input of the waveform type and the emergency setting are acquired and learned, but only one of them (part) may be acquired. Also, even if both input data are collected, it is not necessary to learn both of them, and only one of them may be used for machine learning.

In addition, in the above embodiment, a single program 221 controls switching between normal mode and clinical trial mode, but it is also possible to have a program that operates in normal mode and a separate program that operates in clinical trial mode, and have the management program start one of them to control switching.

In addition, input of judgment data such as breathing patterns to the terminal device 20 is not limited to being performed via the operation reception unit 25. Input data to another terminal device may be acquired by being sent to the terminal device 20 via the communication unit 23. For example, the contents of an operation input to a tablet terminal at hand may be transmitted to the terminal device 20 installed next to the bed. In this case, a separate program may be prepared to display an input screen on the tablet terminal, or the contents displayed by the display unit 24 of the terminal device 20 may simply be transferred to the tablet terminal as well.

In addition, in the above embodiment, a display according to the output of the trained model is performed in the clinical trial mode, but the terminal device 20 does not have to have such a function. In other words, it is sufficient for the terminal device 20 to be able to easily collect training data in the clinical trial mode. In this case, there is no need for the terminal device 20 to calculate the features.

In addition, in the above embodiment, the trained model is generated by the server device 30, but the training data may be copied to another electronic device (which may be a computer, or the terminal device 20) and the trained model may be generated by the other electronic device. On the other hand, in cases where processing is limited to an authorized person on a computer within a hospital, such as the server device 30, personal information does not necessarily have to be deleted from the training data.

In addition, while the machine learning model 322 has been described as learning based on the features of the waveform data, learning may also be performed based on image data of the waveform.

In addition, in the above embodiment, respiration measurement has been used as an example of periodic fluctuation data from which waveform data is obtained, but other fluctuation data, such as measurement data of pulse waves and brain waves, may also be obtained and processed.

In addition, in the above embodiment, it has been described that the measurement results of the patient's biometric data are obtained by medical personnel, but the biometric data of the person being cared for may be measured at a care facility or at home and checked on the spot by the care personnel. Furthermore, the judgment of the care personnel is not limited to being made directly on the spot, but may be made from outside the hospital room via real-time video from a surveillance camera (a slight time lag in communication may occur).

In the above description, the storage unit 22 is a non-volatile memory including a HDD, but is not limited to this, as a computer-readable medium for the program 221 related to the processing operation of the control unit 21 according to the present invention. Portable recording media such as CD-ROMs and DVDs can be used as other computer-readable media. A carrier wave can also be used as a medium for providing data of the program according to the present invention via a communication line.
In addition, the specific configurations, contents and procedures of the processing operations, etc. shown in the above embodiments can be modified as appropriate without departing from the spirit of the present invention. The scope of the present invention includes the scope of the invention described in the claims and its equivalents.

1 Medical information system 10 Biosensor 20 Terminal device 21 Control unit 22 Memory unit 221 Program 222 Measurement/processing data 223 Selection data 224 Learned model 23 Communication unit 24 Display unit 25 Operation reception unit 26 Notification operation unit 30 Server device 31 Control unit 32 Memory unit 321 Learning data 3211 Measurement/analysis data 3212 Selection data 3213 Diagnosis data 322 Machine learning model 323 Program 33 Communication unit 34 Display unit 35 Operation reception unit 36 Notification operation unit 40 Database device 41 Patient information database 50 Network device Dc1, Dc2 Selection screen Df1, Df2 Switching indicator Dma Respiratory rate Dp Estimation result display Dr Reference range Ds Warning indicator Dw Waveform

Claims (9)

a setting means for externally acquiring and setting at least one of information on a waveform type indicated by waveform data relating to a measurement value accompanied by periodic time fluctuation of a subject, the measurement value being acquired from a bio-information measuring device which is a medical device in clinical use, and information on whether an emergency setting is required for the condition of the subject indicated by the waveform data, as corresponding data; and
a storage control means for storing the correspondence data and the waveform data in association with each other;
a switching means for switching between a first mode in which the storage control means stores the corresponding data and a second mode in which the storage control means does not store the corresponding data;
A biological information processing device comprising: The waveform data includes a measurement result that deviates from a predetermined normal range of the subject.
2. The biological information processing apparatus according to claim 1, a calculation means for calculating a predetermined feature amount from the waveform data,
3. The biometric information processing apparatus according to claim 1, wherein the memory control means stores the feature amount in association with the correspondence data. The bioinformation processing device according to claim 3, further comprising an estimation means for estimating at least a portion of the corresponding data corresponding to the newly acquired waveform data based on the relationship between the stored feature amount and the corresponding data. The biometric information processing device according to claim 4, characterized in that the relationship is obtained by learning a machine learning model. An information processing device characterized in that it comprises a database generation means for generating a database in which the first data stored by the memory control means in the bio-information processing device described in any one of claims 1 to 5 corresponds to second data including information of the subject acquired by a device other than the bio-information measuring device. 7. The information processing apparatus according to claim 6 , wherein said database generating means generates said database by deleting personal information that identifies said subject. 8. A trained model generation device comprising a generation means for generating a trained model that includes, as an input, a predetermined feature related to the waveform data contained in the database generated by the information processing device according to claim 6 or 7 , and that uses at least a portion of the corresponding data as teacher data and outputs the at least a portion of the corresponding data. Computer,
a setting means for externally acquiring and setting at least one of information on a waveform type indicated by waveform data relating to a measurement value accompanied by periodic time fluctuation of a subject, the measurement value being acquired from a bio-information measuring device which is a medical device in clinical use, and information on whether an emergency setting is required for the condition of the subject, the information being indicated by the waveform data;
a storage control means for storing the correspondence data and the waveform data in association with each other;
a switching means for switching between a first mode in which the storage control means stores the corresponding data and a second mode in which the storage control means does not store the corresponding data;
A program characterized by causing the program to function as a

Images