Delayed transient bradycardia prediction device, delayed transient bradycardia prediction system, and delayed transient bradycardia prediction program

Abstract

To enable a user to grasp occurrence of delayed transient bradycardia in a fetus prior to actual occurrence of delayed transient bradycardia in the fetus.SOLUTION: Using a temporal change in a cardiac rate of a fetus measured in a pre-occurrence period of bradycardia BDP, which is a period before a predetermined time I from a bradycardia occurrence time point tLD that delayed transient bradycardia LD previously occurred in the fetus, as learning data, a learning model 30 performs learning so as to predict whether delayed transient bradycardia LD occurs in the fetus after the predetermined time I from the pre-occurrence period of bradycardia BDP, and outputs the result. A delayed transient bradycardia prediction part 32 inputs a temporal change in a fetus cardiac rate of an object fetus in an object period TP to the learning model 30 that has learned, and predicts whether the delayed transient bradycardia occurs in the object fetus at a prediction object time point tPT on the basis of the output data of the learning model 30 for the input data. An output control part 24 notifies a user of a fact that it is predicted that the delayed transient bradycardia occurs in the object fetus, when it is predicted that the delayed transient bradycardia occurs in the fetus.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] This specification discloses a delayed transient bradycardia prediction device, a delayed transient bradycardia prediction system, and a delayed transient bradycardia prediction program. [Background technology]

[0002] Conventionally, fetal heart rate has been measured. Various methods have been proposed for measuring the fetal heart rate. For example, a method using a fetal electrocardiograph and a method using ultrasound have been proposed. In the method using ultrasound, ultrasound is transmitted toward the fetal heart, and the fetal heart rate is measured based on a Doppler signal obtained from the reflected wave reflected from the fetal heart. Since the Doppler signal detects the movement of the object to be examined, the measurement of the fetal heart rate based on the Doppler signal can be said to be a measurement method based on the period of the movement of the fetal heart.

[0003] The fetal heart rate may fluctuate. The condition of the fetus can be judged based on the fetal heart rate. Here, a high heart rate (160 bpm or more for the fetal heart rate) is called tachycardia, and a low heart rate (less than 110 bpm for the fetal heart rate) is called bradycardia. An increase in heart rate of 15 bpm or more for a period of 15 seconds or more but less than 2 minutes is called transient tachycardia, and a decrease in heart rate for a period of 15 seconds or more but less than 2 minutes is called transient bradycardia.

[0004] Conventionally, techniques have been proposed for assisting in identifying whether or not a transient tachycardia or a transient bradycardia is occurring in the fetal heart rate.

[0005] For example, Patent Document 1 discloses an information processing device that uses as learning data data in which a fetal heart rate waveform that represents the time change in fetal heart rate is labeled with labels indicating the type of transient variation indicated by the fetal heart rate waveform (e.g., transient tachycardia, early transient bradycardia, late transient bradycardia, variable transient bradycardia (mild), variable transient bradycardia (severe), persistent transient bradycardia), and uses a classifier generated based on the learning data to identify the type of transient variation. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Public Relations, Patent Publication No. 2022-146982 Summary of the Invention [Problem to be solved by the invention]

[0007] In this specification, attention is paid to transient bradycardia. In particular, there are several types of transient bradycardia, including early transient bradycardia, late transient bradycardia, variable transient bradycardia, and persistent transient bradycardia, and in this specification, attention is paid to late transient bradycardia. Late transient bradycardia is a type in which the fetal heart rate gradually decreases and becomes bradycardia with uterine contraction, and then the fetal heart rate returns to a normal value. In particular, in late transient bradycardia, the fetal heart rate reaches its nadir at least later than in early transient bradycardia and variable transient bradycardia after uterine contraction. For example, as a guideline, late transient bradycardia refers to a type in which the time from the start of bradycardia to the nadir is 30 seconds or more.

[0008] Late transient bradycardia often indicates a state in which the fetus has a low blood oxygen concentration, in other words, a state of fetal hypoxia. In particular, when severe late transient bradycardia occurs, in which the amount of decrease in the heart rate is large, a user such as a doctor may need to take urgent action, such as, for example, rapid delivery.

[0009] Therefore, it is desirable for a user such as a doctor to know as early as possible that a fetus will experience late transient bradycardia. In particular, it is desirable for a user to know that a fetus will experience late transient bradycardia before the fetus actually experiences late transient bradycardia. This allows, for example, a user to start taking measures for the fetus earlier.

[0010] The purpose of the delayed transient bradycardia prediction device, delayed transient bradycardia prediction system, and delayed transient bradycardia prediction program disclosed in this specification is to enable a user to know that a delayed transient bradycardia will occur in a fetus before the delayed transient bradycardia actually occurs in the fetus. [Means for solving the problem]

[0011] The delayed transient bradycardia prediction device of this embodiment is characterized by comprising a heart rate measurement unit that measures the heart rate of a fetus over time, a delayed transient bradycardia prediction unit that predicts whether or not delayed transient bradycardia will occur in the target fetus at a prediction target time point, which is a time point after the target period, based on output data from a learning model trained to use as learning data the time change in the heart rate of the fetus measured in a pre-bradycardia period, which is a period a predetermined time before a bradycardia occurrence time when late transient bradycardia occurred in the fetus in the past, and output the result whether or not delayed transient bradycardia will occur in the fetus a predetermined time after the pre-bradycardia period, and a notification unit that notifies a user when the delayed transient bradycardia prediction unit predicts that late transient bradycardia will occur in the target fetus at the prediction target time point.

[0012] The device further includes a uterine contraction pressure measuring unit that measures the uterine contraction pressure of the mother of the fetus over time, and the learning model is trained using learning data that further includes the time change in the uterine contraction pressure of the mother of the fetus measured in the period before the bradycardia onset, and the delayed transient bradycardia prediction unit predicts whether or not the target fetus will experience delayed transient bradycardia at the prediction target time point based on the output data of the learning model when the time change in the uterine contraction pressure of the mother of the target fetus during the target period measured by the uterine contraction pressure measuring unit is input into the learning model.

[0013] The delayed transient bradycardia prediction unit uses as learning data a time change in the heart rate of the fetus measured in a first pre-bradycardia period, which is a period before the first predetermined time from the bradycardia occurrence time, and predicts and outputs whether or not delayed transient bradycardia will occur in the fetus after the first predetermined time from the bradycardia occurrence period. The delayed transient bradycardia prediction unit outputs output data of the first learning model when a time change in the heart rate of the target fetus in a first target period, which is a period before the first predetermined time from the prediction target time, is input to the first learning model. A second learning model is trained to use as learning data the time change in the heart rate of the fetus measured in a second pre-bradycardia period, which is a period before a short second specified time, and to predict and output whether or not delayed transient bradycardia will occur in the fetus after the second specified time from the bradycardia pre-bradycardia period.The second learning model is trained to predict whether or not delayed transient bradycardia will occur in the fetus at the prediction target time based on the output data of the second learning model when the time change in the heart rate of the target fetus during a second target period, which is the period from the prediction target time to the second specified time, is input.

[0014] The notification unit may notify the user in a first manner when the first learning model outputs output data indicating that late transient bradycardia will occur in the target fetus at the predicted time, and then notify the user in a second manner different from the first manner when the second learning model outputs output data indicating that late transient bradycardia will occur in the target fetus at the predicted time.

[0015] The delayed transient bradycardia prediction system according to the present embodiment is a delayed transient bradycardia prediction system including a fetal heart rate measurement device and a delayed transient bradycardia prediction device, the fetal heart rate measurement device including a heart rate measurement unit that measures the fetal heart rate over time, and a measurement device side communication unit that transmits the measurement result of the heart rate measurement unit to the delayed transient bradycardia prediction device, and the delayed transient bradycardia prediction device uses as learning data a time change in the fetal heart rate measured in a pre-bradycardia period, which is a period a predetermined time before a bradycardia occurrence time when a delayed transient bradycardia occurred in the fetus in the past, to predict a delayed transient bradycardia prediction time from the pre-bradycardia period to the predetermined time. a delayed transient bradycardia prediction unit that predicts whether or not delayed transient bradycardia will occur in the target fetus at a prediction target time point, which is the predetermined time after the target period, based on output data of the learning model when a time change in the heart rate of the target fetus during a target period measured by the heart rate measurement unit is input, based on output data of the learning model that has been trained to predict and output whether or not delayed transient bradycardia will occur in the fetus, and a prediction device side notification unit that notifies a user when the delayed transient bradycardia prediction unit predicts that delayed transient bradycardia will occur in the target fetus at the prediction target time point.

[0016] The delayed transient bradycardia prediction device may further include a prediction device side communication unit that notifies the measurement device when the delayed transient bradycardia prediction unit predicts that delayed transient bradycardia will occur in the target fetus at the prediction target time, and the fetal heart rate measurement device may further include a measurement device side notification unit that, in response to the notification from the delayed transient bradycardia prediction device, notifies a user that delayed transient bradycardia is predicted to occur in the target fetus at the prediction target time.

[0017] The delayed transient bradycardia prediction program of this embodiment is characterized in that it operates a computer as a heart rate measurement unit that measures the heart rate of a fetus over time, a delayed transient bradycardia prediction unit that predicts whether or not delayed transient bradycardia will occur in the target fetus at a prediction target time point, which is a time point after the target period, based on output data from a learning model trained to use as learning data the time change in the heart rate of the fetus measured in a pre-bradycardia period, which is a period a predetermined time before a bradycardia occurrence time when late transient bradycardia occurred in the fetus in the past, and output the result, and the delayed transient bradycardia prediction unit predicts whether or not delayed transient bradycardia will occur in the target fetus at a prediction target time point, which is a time point after the target period, based on output data from the learning model when the time change in the heart rate of the target fetus in a target period measured by the heart rate measurement unit is input, and a notification unit that notifies a user when the delayed transient bradycardia prediction unit predicts that late transient bradycardia will occur in the target fetus at the prediction target time point. Effect of the Invention

[0018] The delayed transient bradycardia prediction device, delayed transient bradycardia prediction system, and delayed transient bradycardia prediction program disclosed in this specification enable a user to understand that a fetus will experience delayed transient bradycardia before the delayed transient bradycardia actually occurs in the fetus. [Brief description of the drawings]

[0019] [Figure 1] 1 is a schematic diagram illustrating the configuration of a delayed transient bradycardia prediction device according to the present embodiment. [Diagram 2] 1 is a conceptual diagram showing a comparison target portion used in autocorrelation processing in a Doppler signal. FIG. [Diagram 3] FIG. 2 is a conceptual diagram showing a reference signal used in autocorrelation processing. [Figure 4] 11 is a graph showing a change over time in a correlation coefficient between a comparison portion and a reference signal. [Diagram 5] FIG. 13 is a diagram showing an example of a first pre-bradycardia period. [Figure 6] FIG. 13 is a diagram showing the ROC curve of a trained learning model. [Figure 7] FIG. 13 is a conceptual diagram showing an example of a target period. [Figure 8] FIG. 13 is a diagram illustrating an example of a notification to a user. [Figure 9] FIG. 13 is a diagram showing an example of a second pre-bradycardia period. [Figure 10] FIG. 1 is a conceptual diagram showing the processing of each learning model. [Figure 11] FIG. 2 is a diagram showing an example of a first target period and a second target period. [Figure 12] 1 is a schematic diagram illustrating the configuration of a delayed transient bradycardia prediction system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] 1 is a schematic diagram of a delayed transient bradycardia prediction device 10 according to the present embodiment. The delayed transient bradycardia prediction device 10 is a medical device installed in a medical institution such as a hospital, and has a function of measuring a fetal heart rate. The delayed transient bradycardia prediction device 10 may be, for example, a labor monitoring device that measures not only the fetal heart rate but also the mother's uterine contraction pressure.

[0021] The ultrasonic probe 12 transmits ultrasonic waves toward the fetal heart and receives reflected waves of the ultrasonic waves. The ultrasonic probe 12 includes a plurality of transducer elements.

[0022] The ultrasonic probe 12 transmits ultrasonic waves from a plurality of transducer elements based on a transmission signal, which is an electrical signal, from a transmitter 14, which will be described later. The plurality of transducer elements then receive reflected waves reflected from tissues, including the fetal heart, within the transmission range of the ultrasonic waves, and convert the reflected waves into received signals, which are electrical signals, for output.

[0023] The ultrasound probe 12 is a portable device and is attached to the surface of the mother's abdomen by a belt, etc. In this state, the ultrasound probe 12 transmits ultrasound waves toward the fetal heart and receives reflected waves of the ultrasound waves.

[0024] As shown in FIG. 1, the ultrasonic probe 12 further includes a pressure sensor 12a as a uterine contraction pressure measuring section. When the ultrasonic probe 12 is attached to the surface of the mother's abdomen as described above, the pressure sensor 12a measures the pressure received from the mother's abdomen and converts it into an electrical signal. When a uterine contraction occurs, the pressure received by the pressure sensor 12a from the mother's abdomen changes, so that the pressure sensor 12a can measure the pressure received from the mother's abdomen as the mother's uterine contraction pressure. The pressure sensor 12a measures the mother's uterine contraction pressure over time. That is, the pressure sensor 12a measures the change in the mother's uterine contraction pressure over time.

[0025] The maternal uterine contraction pressure may be measured by a method other than the method using the pressure sensor 12a. For example, the maternal uterine contraction pressure may be measured by an internal measurement method in which a catheter or a small transducer is directly inserted into the uterus of the maternal.

[0026] The transmitting unit 14 transmits a transmission signal to the ultrasonic probe 12 (more specifically, a plurality of transducer elements). As a result, ultrasonic waves are transmitted from each transducer element toward the fetal heart. In this embodiment, the transmitting unit 14 causes the plurality of transducer elements to transmit ultrasonic waves having a center frequency of about 1 MHz. In this embodiment, the transmitting unit 14 causes the plurality of transducer elements to transmit ultrasonic waves simultaneously, and simultaneously stops the transmission of ultrasonic waves, and repeats this. In the ultrasonic transmission stop state, the plurality of transducer elements receive reflected waves from the fetal heart. Specifically, the transmitting unit 14 intermittently transmits a transmission signal to the plurality of transducer elements so that the ultrasonic transmission period and the transmission stop period (reception period) are repeated at a frequency of several KHz. In this embodiment, about 100 waves of ultrasonic waves are transmitted in one transmission period.

[0027] In this embodiment, the ultrasound transmitted to the fetal heart is used solely for measuring the fetal heart rate, and is not used to form an ultrasound image or to measure the velocities of tissues, blood flow, etc. of the subject.

[0028] The receiving unit 16 receives a reception signal based on the reflected waves received by the multiple transducer elements during the above-mentioned reception period from the ultrasound probe 12 (more specifically, the multiple transducer elements). The frequency of the reflected waves may have a frequency (Doppler shift frequency) that is shifted from the frequency of the transmitted ultrasound (transmission frequency) due to the Doppler effect caused by the movement of the fetal heart.

[0029] The receiving unit 16 includes an amplifier, and adds up the received signals from the multiple transducer elements and amplifies the sum in the amplifier. The receiving unit 16 outputs the amplified received signal to the Doppler signal forming unit 18.

[0030] The Doppler signal forming unit 18 includes a mixer. The mixer multiplies the received signal received from the receiving unit 16 by a reference signal having the same frequency f0 as the transmission frequency. As a result, the Doppler shift frequency f d (in other words, the difference frequency with the transmission frequency), and d A signal having a frequency of

[0031] The Doppler signal forming unit 18 also includes a low-pass filter. The low-pass filter filters out the high-frequency component 2f0+f d This removes the signal at the Doppler shift frequency f d The signal having the Doppler shift frequency f d is generally within the audible frequency range (20Hz to 20kHz).

[0032] Furthermore, the Doppler signal forming unit 18 includes an AD converter. The AD converter converts the Doppler shift frequency f d In this embodiment, the sampling frequency of the digital conversion is about 2 kHz.

[0033] The Doppler shift frequency f obtained by the Doppler signal forming unit 18 dThe signal having the above formula is a Doppler signal. The Doppler signal can be expressed as a waveform in a coordinate space defined by the horizontal axis representing time and the vertical axis representing the signal strength of the reflected wave. The Doppler signal includes a fetal heartbeat component corresponding to the movement of the fetal heart.

[0034] The heart rate measurement unit 20 measures the fetal heart rate. In this embodiment, the heart rate measurement unit 20 measures the fetal heart rate based on the Doppler signal formed by the Doppler signal formation unit 18. Specifically, the heart rate measurement unit 20 measures the fetal heart rate based on the periodicity of the Doppler signal. Hereinafter, a measurement process using autocorrelation processing on the Doppler signal will be described as a measurement process for measuring the fetal heart rate.

[0035] FIG. 2 is a conceptual diagram showing a comparison portion CP in a Doppler signal DS used in autocorrelation processing. FIG. 3 is a conceptual diagram showing a reference signal RDS used in autocorrelation processing. The heart rate measurement unit 20 calculates a correlation coefficient (hereinafter, may be simply referred to as a "correlation coefficient") between the reference signal RDS and a comparison portion CP, which is a portion of the Doppler signal DS having the same time length as the reference signal RDS. As will be described later, the reference signal RDS may be updated, but an initial reference signal RDS indicating a fetal heartbeat component is prepared in advance. When the reference signal RDS indicates a fetal heartbeat component, the reference signal RDS becomes a signal indicating the movement of the fetal heart wall or heart valve.

[0036] The heart rate measurement unit 20 repeatedly calculates the correlation coefficient while shifting the comparison portion CP little by little in the time direction (for example, by the time equivalent to one sample at the sampling frequency of the AD converter of the Doppler signal formation unit 18). By calculating the correlation coefficient while shifting the comparison portion CP in the time direction, the change in the correlation coefficient over time is calculated.

[0037] FIG. 4 is a graph showing the change in correlation coefficient over time. When the reference signal RDS indicates a fetal heartbeat component, the timing at which the correlation coefficient becomes large (for example, the timing at which it reaches a maximum value) can be said to indicate the timing at which the fetal heart wall or heart valve moves in the same way as the reference signal RDS. From this, it can be said that the period of the time change in the correlation coefficient represents the fetal heartbeat. Therefore, the heart rate measurement unit 20 can measure the fetal heart rate based on the period of the time change in the correlation coefficient. For example, the time t between the maximum values ​​of the correlation coefficient can be said to represent the time between one heartbeat of the fetus, so the heart rate measurement unit 20 obtains the fetal heartbeat per second by calculating the reciprocal of the time t.

[0038] In the above-mentioned autocorrelation process, the reference signal RDS is updated. Specifically, when the correlation coefficient satisfies a predetermined condition (for example, when it exceeds a predetermined threshold value or when the correlation coefficient is a maximum value), the comparison target portion CP corresponding to the correlation coefficient is set as a new reference signal RDS. By updating the reference signal RDS, even if the signal waveform of the Doppler signal representing the fetal heart rate component gradually changes, it is possible to suppress a decrease in the calculation accuracy of the fetal heart rate.

[0039] The heart rate measuring section 20 measures the fetal heart rate over time by the above-mentioned measurement process based on the Doppler signals successively formed based on the reception signals successively received by the ultrasonic probe 12 .

[0040] The heart rate measurement unit 20 may measure the fetal heart rate by a method other than the method using the Doppler signal. For example, the heart rate measurement unit 20 may measure the fetal heart rate over time by an internal measurement method that measures the fetal heart rate using a directly induced fetal electrocardiogram signal obtained by attaching a bipolar spiral electrode to the fetal head.

[0041] The waveform forming unit 22 forms a fetal heart rate waveform that represents the time change of the fetal heart rate measured by the heart rate measuring unit 20 over time. The waveform forming unit 22 also forms a uterine contraction pressure waveform that represents the time change of the uterine contraction pressure measured by the pressure sensor 12a over time. A combination of the fetal heart rate waveform and the uterine contraction pressure waveform is called a CTG (CardioTocoGram). FIG. 5 shows an example of a CTG consisting of a fetal heart rate waveform FHR and a uterine contraction pressure waveform UC. In the example of FIG. 5 (as well as FIGS. 7 to 9 and 11 described later), the horizontal axis represents time, and the vertical axis represents the fetal heart rate or the uterine contraction pressure.

[0042] The output control unit 24 causes the display 26 to display the fetal heart rate measured by the heart rate measurement unit 20, the uterine contraction pressure measured by the pressure sensor 12a, and the CTG formed by the waveform formation unit 22. The output control unit 24 causes the display 26 to display the fetal heart rate as a numerical value, such as beats per minute, for example. In addition to or instead of displaying the fetal heart rate on the display 26, the output control unit 24 may output the fetal heart rate as a sound from a speaker (not shown). For example, the output control unit 24 may cause the speaker to output heart sounds prepared in advance in time with the fetal heart rate. The output control unit 24 may also cause the display 26 to display the uterine contraction pressure as a numerical value.

[0043] The display 26 includes, for example, a liquid crystal display. Various screens are displayed on the display 26 according to the control from the output control unit 24.

[0044] The memory 28 includes a non-volatile memory such as an embedded multi media card (eMMC) or a read only memory (ROM), and a volatile memory such as a random access memory (RAM). A delayed transient bradycardia prediction program for operating each part of the delayed transient bradycardia prediction device 10 is stored in the non-volatile memory. The delayed transient bradycardia prediction program can also be stored in a computer-readable non-transient storage medium such as a universal serial bus (USB) memory or a CD-ROM. The delayed transient bradycardia prediction device 10 can read and execute the delayed transient bradycardia prediction program from such a storage medium.

[0045] 1, the memory 28 (specifically, a non-volatile memory) stores a learning model 30. The learning model 30 is not limited to this, but may be, for example, an image recognition model such as a Convolutional Neural Network (CNN) or a natural language processing model such as a Transformer.

[0046] The learning model 30 is trained to predict whether or not a late transient bradycardia will occur in the fetus after a predetermined time from the bradycardia pre-occurrence period, using as learning data a time change in the fetal heart rate measured in the bradycardia pre-occurrence period, which is a period before a bradycardia occurrence time when a late transient bradycardia occurred in the fetus in the past, and to output the prediction. In this embodiment, a learning process for training the learning model 30 is executed in a learning processing device, which is a device other than the late transient bradycardia prediction device 10. Then, the learning model 30 that has been trained in the learning processing device is stored in the memory 28. However, the late transient bradycardia prediction device 10 may be provided with a learning processing unit that executes the learning process described below, that is, the late transient bradycardia prediction device 10 may be a learning processing device. A learning method for the learning model 30 will be described in detail below.

[0047] First, previously acquired CTGs are stored in a CTG database accessible from the learning processing device. Then, a learning processor who performs the learning process extracts CTGs in which delayed transient bradycardia occurs from the CTGs stored in the CTG database.

[0048] In the following, it is assumed that the CTG shown in FIG. 5 has been extracted. In the fetal heart rate waveform FHR, the part where the fetal heart rate gradually drops and then recovers is the part where delayed transient bradycardia LD occurs. LD is the start time of delayed transient bradycardia LD, that is, the time when bradycardia occurs. LD The bradycardia pre-onset period BDP is determined as a period before the predetermined time I from the time t. The predetermined time I may be determined in advance. For example, the predetermined time I is determined as 1.5 minutes, 3 minutes, 6 minutes, 10 minutes, or more. In FIG. 5, the bradycardia pre-onset period BDP is determined as a period before the time t 11 From time t 12 In this specification, the time t LD The period before a certain time I from the time of the bradycardia onset period BDP is the period before the end of the bradycardia onset period BDP (i.e., time t 12 ) at time t LD This means that it is a certain time I before.

[0049] The time length of the bradycardia pre-occurrence period BDP may be determined in advance. Since doctors generally interpret the baseline of the fetal heart rate waveform FHR in a 10-minute section of the fetal heart rate waveform FHR, the time length of the bradycardia pre-occurrence period BDP is set to 10 minutes in this embodiment. However, the time length of the bradycardia pre-occurrence period BDP is not limited to this as long as the learning model 30 can learn appropriately.

[0050] The learning processor extracts the fetal heart rate waveform FHR in the bradycardia pre-onset period BDP. The extracted fetal heart rate waveform FHR in the bradycardia pre-onset period BDP becomes the learning data for the learning model 30. As described above, in this embodiment, it should be noted that the part of the fetal heart rate waveform FHR that does not include the late transient bradycardia LD (more specifically, the part that precedes the late transient bradycardia LD in time) becomes the learning data.

[0051] The processor of the learning processing device inputs the fetal heart rate waveform FHR of the bradycardia pre-onset period BDP as learning data to the learning model 30. Based on the input fetal heart rate waveform FHR of the bradycardia pre-onset period BDP, the learning model 30 predicts whether or not a delayed transient bradycardia will occur after a predetermined time I of the bradycardia pre-onset period BDP, and outputs the prediction result. The processor of the learning processing device adjusts the parameters of the learning model 30 so that the difference between the prediction result and the teacher data indicating that a delayed transient bradycardia will occur after the predetermined time I of the bradycardia pre-onset period BDP becomes small. By repeating such processing, the learning model 30 that has learned sufficiently can predict with high accuracy whether or not a delayed transient bradycardia will occur after a predetermined time I of a certain period (more specifically, the end time of the period) based on the input time change of the fetal heart rate during the certain period, and output the result.

[0052] The learning model 30 may be trained using training data further including a time change in the uterine contraction pressure of the mother of the fetus measured in the bradycardia pre-onset period BDP. That is, the learning model 30 may be trained using the CTG (fetal heart rate waveform FHR and uterine contraction pressure waveform UC) measured in the bradycardia pre-onset period BDP as training data. Because uterine contraction pressure changes periodically and delayed transient bradycardia LD occurs with an increase in uterine contraction pressure, the prediction accuracy of the learning model 30 can be improved by adding the uterine contraction pressure waveform UC to the training data.

[0053] The learning model 30 may be trained using training data further including a time change in the fetal heart rate in a non-pre-bradycardia period, which is a period before a predetermined time I from a time point when the fetus does not experience late transient bradycardia. In this case, a training processor who performs the training process extracts fetal heart rate waveforms FHR in which late transient bradycardia does not occur from the CTG database. Then, the fetal heart rate waveforms FHR in the non-pre-bradycardia period among the extracted fetal heart rate waveforms FHR are used as training data. In this case, a label indicating that late transient bradycardia occurs after a predetermined time I of the pre-bradycardia period BDP is attached to the fetal heart rate waveforms FHR in the training data in the pre-bradycardia period BDP, and a label indicating that late transient bradycardia does not occur after a predetermined time I of the non-pre-bradycardia period is attached to the fetal heart rate waveforms FHR in the non-pre-bradycardia period as training data. Such labels are used as teacher data. The learning model 30 can improve the prediction accuracy of the trained learning model 30 or the learning efficiency of the learning model 30 by training the learning model 30 using training data including not only the fetal heart rate waveform FHR in the pre-bradycardia period BDP, but also the fetal heart rate waveform FHR in the non-pre-bradycardia period.

[0054] Furthermore, the learning model 30 may be trained using training data that further includes a time change in the uterine contraction pressure of the fetus and the mother during the non-bradycardia pre-period. In other words, the learning model 30 may be trained using training data that further includes a CTG measured during the non-bradycardia pre-period. This also improves the prediction accuracy of the learning model 30 for the reasons described above.

[0055] The fact that it is possible to predict the occurrence of late transient bradycardia after a predetermined time I of the pre-bradycardia period BDP based on the fetal heart rate waveform FHR in the pre-bradycardia period BDP means that the fetal heart rate waveform FHR in the pre-bradycardia period BDP has a characteristic indicating that late transient bradycardia will occur thereafter. Specifically, as described above, late transient bradycardia indicates a hypoxic or malnutritional state of the fetus, and it is considered that a waveform pattern of the fetal heart rate waveform FHR reflecting the hypoxic or malnutritional state of the fetus appears before the actual occurrence of late transient bradycardia. Examples of such waveform patterns include the manner of fine fluctuation of the baseline of the fetal heart rate waveform FHR and the occurrence of persistent transient bradycardia.

[0056] Furthermore, as data showing the relationship between the characteristics of the CTG of the bradycardia pre-onset period BDP and the occurrence of delayed transient bradycardia after the bradycardia pre-onset period BDP, FIG. 6 shows a ROC (Receiver Operating Characteristic) curve of the trained learning model 30. In the graph of FIG. 6, the horizontal axis shows FPF (False Positive Fraction), that is, the probability that the learning model 30 has determined that delayed transient bradycardia will occur among the input data (CTG) group in which delayed transient bradycardia did not occur after the predetermined time I, and the vertical axis shows TPF (True Positive Fraction), that is, the probability that the learning model 30 has correctly determined that delayed transient bradycardia will occur among the input data (CTG) group in which delayed transient bradycardia occurred after the predetermined time I. Although multiple ROC curves are shown in FIG. 6, attention should be paid to any one of the ROC curves in the description of the ROC curves.

[0057] The learning model 30 outputs the possibility of delayed transient bradycardia occurring as output data based on the input data (CTG in this case). For example, the learning model 30 outputs a value between 0 and 1 (0 is a case where it is predicted that delayed transient bradycardia will not occur, and 1 is a case where it is predicted that delayed transient bradycardia will occur 100%) as output data. When determining whether delayed transient bradycardia will occur or not as a prediction result based on the output data of the learning model 30, a threshold is set in the range of 0 to 1, which is the range that the output data of the learning model 30 can take, and a prediction result is determined such that if the output data of the learning model 30 is equal to or greater than the threshold, delayed transient bradycardia will occur, and if the output data of the learning model 30 is less than the threshold, delayed transient bradycardia will not occur.

[0058] The ROC curve is a graph showing the change in the FPF and TPF of the learning model 30 when the threshold is changed between 0 and 1. For example, when the threshold is set to 1 (maximum value), the prediction result is determined to be that delayed transient bradycardia does not occur regardless of the output data of the learning model 30. In this case, both the FPF and the TPF are plotted at 0[%]. Conversely, when the threshold is set to 0 (minimum value), the prediction result is determined to be that delayed transient bradycardia occurs regardless of the output data of the learning model 30. In this case, both the FPF and the TPF are plotted at 100[%]. When the threshold is changed between 0 and 1, the values ​​of the FPF and the TPF also gradually change as shown in the ROC curve in FIG. 6.

[0059] Needless to say, a smaller FPF is preferable, and a larger TPF is preferable. Therefore, in a two-dimensional space with FPF on the horizontal axis and TPF on the vertical axis as shown in FIG. 6, the closer the ROC curve passes to the upper left (i.e., FPF=0[%], TPF=100[%]), the higher the prediction accuracy of the learning model 30 will be if the threshold is set appropriately. Incidentally, a parameter called AUC (Area Under Curve) has been proposed as an evaluation of the ROC curve. AUC is a parameter indicating the area of ​​the region below the ROC curve in the two-dimensional space shown in FIG. 6, and it can be said that the larger the AUC, the higher the prediction accuracy of the learning model 30. Conversely, if the prediction accuracy of the learning model 30 is quite low and, for example, the output data is output randomly, the ROC curve will be close to a straight line connecting the position where the FPF and TPF are both 0[%] and the position where the FPF and TPF are both 100[%].

[0060] As described above, multiple ROC curves are shown in FIG. 6. Each ROC curve is a function of the time period I (i.e., the pre-bradycardia period BDP, which is the learning data, and the time t LD 5. Specifically, the ROC curve indicated by the circle marker and the solid line is the ROC curve of the learning model 30 trained using learning data with a predetermined time I of 1.5 minutes, the ROC curve indicated by the triangle marker and the dashed line is the ROC curve of the learning model 30 trained using learning data with a predetermined time I of 3 minutes, the ROC curve indicated by the square marker and the solid line is the ROC curve of the learning model 30 trained using learning data with a predetermined time I of 6 minutes, and the ROC curve indicated by the circle marker and the dashed line is the ROC curve of the learning model 30 trained using learning data with a predetermined time I of 10 minutes.

[0061] As shown in FIG. 6, the ROC curve of the learning model 30 trained using the learning data with the predetermined time I of 1.5 minutes or 3 minutes has a significantly larger AUC than the straight line connecting the positions where the FPF and TPF are both 0[%] and the positions where the FPF and TPF are both 100[%]. That is, the learning model 30 trained using the CTG of the bradycardia pre-onset period BDP as the learning data can predict that a delayed transient bradycardia will occur after the CTG that is the input data. That is, it can be said that the CTG of the bradycardia pre-onset period BDP has a characteristic indicating that a delayed transient bradycardia will occur thereafter. If the CTG of the bradycardia pre-onset period BDP does not have any characteristic indicating that a delayed transient bradycardia will occur thereafter, the ROC curve of the trained learning model 30 should be close to the straight line connecting the positions where the FPF and TPF are both 0[%] and the positions where the FPF and TPF are both 100[%].

[0062] In the ROC curve shown in Fig. 6, the AUC of the ROC curve of the learning model 30 trained using learning data with a predetermined time I of 6 minutes or 10 minutes is not very large. However, in the future, it is expected that the accuracy of the learning model 30 will be improved even when the predetermined time I is long (including 10 minutes or more) by devising the structure of the learning model 30 and hyperparameters (for example, the number of epochs (the number of times one learning data is repeatedly used), the number of hidden layers, the number of nodes in each hidden layer, the number of dropouts (the number of nodes whose weights and biases are not adjusted in the learning process), or the number of batches (the number of learning data included in a subset when the learning data is divided into multiple subsets)).

[0063] Returning to FIG. 1, the delayed transient bradycardia prediction unit 32 predicts whether or not a delayed transient bradycardia will occur in a target fetus, which is a prediction target for whether or not a delayed transient bradycardia will occur, using the learned learning model 30. Specifically, first, the delayed transient bradycardia prediction unit 32 specifies the time change of the fetal heart rate in a predetermined target period. FIG. 7 is a conceptual diagram showing an example of the target period TP. FIG. 7 shows a target fetal heart rate waveform TFHR representing the time change of the fetal heart rate of the target fetus measured by the heart rate measurement unit 20, and a target uterine contraction pressure waveform TUC representing the time change of the uterine contraction pressure of the mother of the target fetus measured by the pressure sensor 12a. The time length of the target period TP may be set appropriately. For example, the time length of the target period TP may be the same as the time length of the learning data used to train the learning model 30, that is, the same as the time length of the bradycardia pre-occurrence period BDP.

[0064] The late transient bradycardia prediction unit 32 inputs the time change of the fetal heart rate of the target fetus during the target period TP to the learning model 30. Conceptually, the late transient bradycardia prediction unit 32 inputs the target fetal heart rate waveform TFHR during the target period TP shown in FIG. 7 to the learning model 30. Based on the input data, the learning model 30 calculates the time change of the fetal heart rate of the target fetus during the target period TP at a prediction target time t PT The delayed transient bradycardia prediction unit 32 predicts and outputs the possibility (for example, a value between 0 and 1) of delayed transient bradycardia occurring in the target fetus at the prediction target time t PT For example, the delayed transient bradycardia prediction unit 32 sets a predetermined threshold value, and if the output data of the learning model 30 is equal to or greater than the threshold value, predicts whether or not delayed transient bradycardia will occur in the target fetus at the prediction target time t PT If the output data of the learning model 30 is less than the threshold, the prediction target time t PT It is predicted that delayed transient bradycardia will not occur in the target fetus.

[0065] When the learning model 30 is trained using learning data further including the time change of the uterine contraction pressure of the mother of the fetus measured in the bradycardia pre-onset period BDP, or the bradycardia pre-onset period BDP and the non-bradycardia pre-onset period, the late transient bradycardia prediction unit 32 inputs the time change of the fetal heart rate of the target fetus in the target period TP and the time change of the uterine contraction pressure of the mother of the target fetus in the target period TP measured by the pressure sensor 12a to the learning model 30. Conceptually, the late transient bradycardia prediction unit 32 inputs the target fetal heart rate waveform TFHR and the target uterine contraction pressure waveform TUC of the target period TP shown in FIG. 7 to the learning model 30. Then, the late transient bradycardia prediction unit 32 calculates the prediction target time t based on the output data of the learning model 30 for the input data. PT and predicting whether or not delayed transient bradycardia will occur in the target fetus.

[0066] As the fetal heart rate of the target fetus is measured over time by the heart rate measurement unit 20, or as the uterine contraction pressure of the mother of the target fetus is measured over time by the pressure sensor 12a, the delayed transient bradycardia prediction unit 32 repeatedly inputs the time change in the fetal heart rate of the target fetus during the target period TP, or in addition, the time change in the uterine contraction pressure of the mother of the target fetus, into the learning model 30 while changing (shifting) the target period TP, as shown in FIG. PT Since the target period TP is a certain time later than I, if the target period TP is changed, the prediction target time t PT As a result, the delayed transient bradycardia prediction unit 32 changes the prediction target time t PT In other words, the delayed transient bradycardia prediction unit 32 predicts whether or not a delayed transient bradycardia will occur in the target fetus at the prediction target time t PT The purpose of this study is to predict over time whether or not the target fetus will develop delayed transient bradycardia.

[0067] The delayed transient bradycardia prediction unit 32 predicts in real time whether or not delayed transient bradycardia will occur in the target fetus at the prediction target time point. That is, by setting the end point of the target period TP to always be the current time point, the delayed transient bradycardia prediction unit 32 predicts in real time whether or not delayed transient bradycardia will occur in the target fetus at the prediction target time point t PT In this study, we will predict over time whether or not the target fetus will develop delayed transient bradycardia.

[0068] The output control unit 24 as a notification unit notifies the delayed transient bradycardia prediction unit 32 of the prediction target time t PT If it is predicted that the target fetus will have a delayed transient bradycardia at the predicted time t PT A user such as a doctor is notified that there is a possibility that the target fetus will experience delayed transient bradycardia after a predetermined time I.

[0069] The prediction target time t is a given time I from the present time PT There may be various methods for the output control unit 24 to notify the user that the target fetus may experience late transient bradycardia after a predetermined time I. For example, the output control unit 24 notifies the user by displaying on the display 26 information indicating that the target fetus may experience late transient bradycardia after a predetermined time I. For example, as shown in FIG. 8, the output control unit 24 displays a fetal heart rate waveform 40 and a uterine contraction pressure waveform 42 formed by the waveform forming unit 22 together with a predicted target time t PT A message 44 may be displayed on the display 26 indicating that the subject fetus may experience a late transient bradycardia at some point in time (3 minutes from the present time in the example of FIG. 8).

[0070] In addition, instead of or in addition to displaying message 44, the output control unit 24 may cause an illumination unit (e.g., an LED (Light-Emitting Diode)) provided in the delayed transient bradycardia prediction device 10 to emit light, or may output sound from a speaker provided in the delayed transient bradycardia prediction device 10, as a notification to the user.

[0071] In the above embodiment, the delayed transient bradycardia prediction unit 32 predicts that delayed transient bradycardia will occur in the target fetus after the predetermined time I from the current time point using a learning model 30 corresponding to one predetermined time I, but the delayed transient bradycardia prediction unit 32 may predict that delayed transient bradycardia will occur in the target fetus after each predetermined time I from the current time point using multiple learning models 30 corresponding to multiple predetermined times I. Hereinafter, a case will be described in which the delayed transient bradycardia prediction unit 32 predicts that delayed transient bradycardia will occur in the target fetus using two learning models 30 (a first learning model and a second learning model) having different predetermined times I, but the delayed transient bradycardia prediction unit 32 may predict that delayed transient bradycardia will occur in the target fetus using three or more learning models 30 corresponding to three or more predetermined times I.

[0072] First, a learning method of the first learning model will be described. The first learning model is learned by using as learning data the time change of the fetal heart rate measured in a first pre-bradycardia period, which is a period before a first predetermined time from the bradycardia occurrence time when the fetus experienced late transient bradycardia in the past. Specifically, when the CTG shown in FIG. 5 is extracted from the CTG database, the learning processor extracts a time t LD A first bradycardia pre-onset period BDP1, which is a period before the first predetermined time I1, is specified from the first bradycardia pre-onset period BDP1. Here, the first predetermined time I1 is assumed to be 3 minutes. Then, the learning processor extracts the fetal heart rate waveform FHR of the first bradycardia pre-onset period BDP1. The processor of the learning processing device trains the first learning model using the extracted fetal heart rate waveform FHR of the first bradycardia pre-onset period BDP1 as learning data to predict whether or not delayed transient bradycardia will occur in the fetus after the first predetermined time I1 from the first bradycardia pre-onset period BDP1 and output the prediction. The first learning model that has been sufficiently trained becomes able to predict with high accuracy whether or not delayed transient bradycardia will occur after the first predetermined time I1 of the period (more specifically, the end time of the period) based on the inputted time change in the fetal heart rate during the period and output the prediction.

[0073] In this case, the first learning model may be trained using training data further including a time change in the uterine contraction pressure of the mother of the fetus measured in the first pre-bradycardia period BDP1. The first learning model may be trained using training data further including a time change in the fetal heart rate in the non-pre-bradycardia period. The first learning model may be trained using training data further including a time change in the uterine contraction pressure of the mother of the fetus in the non-pre-bradycardia period.

[0074] Next, a learning method of the second learning model will be described. The second learning model is learned using as learning data the time change of the heart rate of the fetus measured in a second pre-bradycardia period, which is a period before a second predetermined time that is shorter than the first predetermined time from the bradycardia occurrence time when the fetus experienced late transient bradycardia in the past. Specifically, when the CTG shown in FIG. 9 is extracted from the CTG database, the learning processor extracts a time t LD A second bradycardia pre-onset period BDP2 is specified, which is a period before the second predetermined time I2 from the first bradycardia pre-onset period BDP2. Here, the second predetermined time I2 is assumed to be 1.5 minutes. Then, the learning processor extracts the fetal heart rate waveform FHR of the second bradycardia pre-onset period BDP2. The processor of the learning processing device trains the second learning model using the extracted fetal heart rate waveform FHR of the second bradycardia pre-onset period BDP2 as learning data to predict whether or not delayed transient bradycardia will occur in the fetus after the second predetermined time I2 from the second bradycardia pre-onset period BDP2 and output the prediction. The second learning model that has been sufficiently trained becomes able to predict with high accuracy whether or not delayed transient bradycardia will occur after the second predetermined time I2 of the period (more specifically, the end time of the period) based on the inputted time change of the fetal heart rate during the period and output the prediction.

[0075] In this case, the second learning model may be trained using training data further including a time change in the uterine contraction pressure of the mother of the fetus measured in the second pre-bradycardia period BDP2. The second learning model may be trained using training data further including a time change in the fetal heart rate in the non-pre-bradycardia period. The second learning model may be trained using training data further including a time change in the uterine contraction pressure of the mother of the fetus in the non-pre-bradycardia period.

[0076] As described above, the delayed transient bradycardia prediction unit 32 predicts the delayed transient bradycardia at the prediction target time t PT The delayed transient bradycardia prediction unit 32 predicts in real time whether or not a delayed transient bradycardia will occur in the target fetus at the target time t . Therefore, the delayed transient bradycardia prediction unit 32 shifts the target period TP (see FIG. 7) in the time direction and sequentially inputs the target period TP whose end point is the current time into the first learning model 30a and the second learning model 30b. Here, as shown in FIG. 10, when the time change of the fetal heart rate during the same target period TP is input to the first learning model 30a and the second learning model 30b, the first learning model 30a predicts the target time t PT is the prediction target time t of the second learning model 30b. PT In this example, the second learning model 30b predicts whether or not the target fetus will experience late transient bradycardia 1.5 minutes after the current time, which is after the second predetermined time I2 of the target period TP, while the first learning model 30a predicts whether or not the target fetus will experience late transient bradycardia 3 minutes after the current time, which is after the first predetermined time I1 of the target period TP.

[0077] Therefore, as shown in FIG. 11, at a given prediction target time t PT First, the delayed transient bradycardia prediction unit 32 predicts the delayed transient bradycardia at a prediction target time t PT The first learning model 30a inputs the target fetal heart rate waveform TFHR for the first target period TP1, which is the period from the first target time t PT The delayed transient bradycardia prediction unit 32 outputs output data indicating the possibility that the delayed transient bradycardia will occur in the target fetus at the prediction target time t PTThe output data of the first learning model 30a is temporarily stored in the memory 28 in association with the first target time t 1 . At this time, the delayed transient bradycardia prediction unit 32 also inputs the target fetal heart rate waveform TFHR for the first target period TP1 to the second learning model 30b. As a result, the second learning model 30b inputs the target fetal heart rate waveform TFHR for the first target time t PT Therefore, output data indicating the possibility that delayed transient bradycardia may occur in the target fetus at a time point prior to (first predetermined time I1-second predetermined time I2) is output.

[0078] At a time point (first predetermined time I1-second predetermined time I2) after the time point when the target fetal heart rate waveform TFHR for the first target period TP1 is input to the first learning model 30a (and the second learning model 30b), the delayed transient bradycardia prediction unit 32 predicts a prediction target time point t PT The target fetal heart rate waveform TFHR for the second target period TP2, which is the period from the prediction target time t PT and outputs output data indicating the possibility of delayed transient bradycardia occurring in the target fetus.

[0079] Then, the delayed transient bradycardia prediction unit 32 calculates the delayed transient bradycardia prediction time t PT and output data indicating the possibility of late transient bradycardia occurring in the target fetus at the prediction target time t PT Based on the output data indicating the possibility of delayed transient bradycardia occurring in the target fetus at the predicted target time t PT For example, when both the output data of the first learning model 30a and the output data of the second learning model 30b are equal to or greater than the above threshold, the delayed transient bradycardia prediction unit 32 predicts whether or not a delayed transient bradycardia will occur in the target fetus at the prediction target time t PT If at least one of the output data of the first learning model 30a and the output data of the second learning model 30b is less than a threshold, the prediction target time t PT It is predicted that no delayed transient bradycardia will occur in the target fetus at this point.

[0080] In addition, as shown in FIG. 6, since it is considered that the learning model 30 having a shorter predetermined time I (the second learning model 30b in the above example) has a higher prediction accuracy, the delayed transient bradycardia prediction unit 32 prepares a first weight for the first learning model 30a and a second weight for the second learning model 30b that is heavier than the first weight, and calculates the prediction target time t based on the output data of the first learning model 30a and the output data of the second learning model 30b while taking the first weight and the second weight into consideration. PT It may also be possible to predict whether or not a delayed transient bradycardia will occur in the target fetus.

[0081] In this case, when the first learning model 30a is trained using training data further including the time change of the uterine contraction pressure of the mother of the fetus measured in the bradycardia pre-occurrence period BDP, or the bradycardia pre-occurrence period BDP and the non-bradycardia pre-period, the delayed transient bradycardia prediction unit 32 may input the target fetal heart rate waveform TFHR and the target uterine contraction pressure waveform TUC in the first target period TP1 to the first learning model 30a. When the second learning model 30b is trained using training data further including the time change of the uterine contraction pressure of the mother of the fetus measured in the bradycardia pre-occurrence period BDP, or the bradycardia pre-occurrence period BDP and the non-bradycardia pre-period, the delayed transient bradycardia prediction unit 32 may input the target fetal heart rate waveform TFHR and the target uterine contraction pressure waveform TUC in the second target period TP2 to the second learning model 30b.

[0082] In addition, the output control unit 24 determines whether the first learning model 30a predicts the prediction time t PT When the output data indicating that the target fetus will experience delayed transient bradycardia at the prediction target time t is output, the user is notified in the first manner, and then the second learning model 30b predicts the occurrence of delayed transient bradycardia at the prediction target time t PT When output data indicating that the target fetus will experience delayed transient bradycardia is output, the user may be notified in a second manner different from the first manner.

[0083] For example, the output control unit 24 detects that the first learning model 30a predicts the prediction target time t PTWhen the output data indicates that the target fetus will have a delayed transient bradycardia at the prediction target time t PT When output data is output indicating that the target fetus will experience late transient bradycardia, the message 44 or the background of the fetal heart rate waveform 40 and the uterine contraction pressure waveform 42 may be displayed in red.

[0084] This allows the user to predict whether multiple learning models 30 are predictive at the time t PT It is possible to gradually grasp that output data indicating that delayed transient bradycardia will occur in the target fetus has been output.

[0085] The outline of the delayed transient bradycardia prediction device 10 according to the present embodiment has been described above. Among the above-mentioned components of the delayed transient bradycardia prediction device 10, the transmitting unit 14, the receiving unit 16, the Doppler signal forming unit 18, the heart rate measuring unit 20, the waveform forming unit 22, the output control unit 24, and the delayed transient bradycardia prediction unit 32 are configured by a processor. The processor includes at least one of a general-purpose processing device (e.g., a central processing unit (CPU)) and a dedicated processing device (e.g., a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or a programmable logic device). The processor may not be a single processing device, but may be configured by the cooperation of multiple processing devices located at physically separate locations. Moreover, each of the above-mentioned components may be realized by the cooperation of hardware such as a processor and software.

[0086] According to the delayed transient bradycardia prediction device 10, the user can know that the delayed transient bradycardia will occur in the target fetus in the future (after a predetermined time I) before the delayed transient bradycardia actually occurs in the target fetus. This allows the user to start treatment such as preparing for rapid delivery or securing a hospital if the target fetus is to be transported to another hospital earlier.

[0087] Fig. 12 is a schematic diagram of a delayed transient bradycardia prediction system 50 according to this embodiment. In the delayed transient bradycardia prediction device 10 shown in Fig. 1, the fetal heart rate measurement process and the prediction process of whether delayed transient bradycardia will occur are performed in one device, but the fetal heart rate measurement process and the prediction process of whether delayed transient bradycardia will occur may be performed in different devices.

[0088] 12 includes a fetal heart rate measuring device 52 and a late transient bradycardia prediction device 54. The fetal heart rate measuring device 52 and the late transient bradycardia prediction device 54 are communicatively connected to each other via a communication line 56 such as a LAN (Local Area Network) or a WAN (Wide Area Network). The late transient bradycardia prediction system 50 may include a plurality of fetal heart rate measuring devices 52. That is, the late transient bradycardia prediction device 54 may be communicatively connected to a plurality of fetal heart rate measuring devices 52.

[0089] The fetal heart rate measuring device 52 is installed in an examination room, a delivery room, or the like in a hospital. The late transient bradycardia prediction device 54 is installed in a space different from the fetal heart rate measuring device 52, for example, a nurse center or a central control room in a hospital. Alternatively, the fetal heart rate measuring device 52 may be installed in, for example, the pregnant woman's home, and the late transient bradycardia prediction device 54 may be a server provided remotely from the pregnant woman's home. In this case, the function performed by the late transient bradycardia prediction device 54 may be provided by a cloud service.

[0090] The fetal heart rate measuring device 52 has an ultrasonic probe 12 including a pressure sensor 12a, a transmitter 14, a receiver 16, a Doppler signal forming unit 18, a heart rate measuring unit 20, a waveform forming unit 22, an output control unit 24, a display 26, and a communication unit 60. The configurations and functions of the pressure sensor 12a, ultrasonic probe 12, transmitter 14, receiver 16, Doppler signal forming unit 18, heart rate measuring unit 20, waveform forming unit 22, output control unit 24, and display 26 of the fetal heart rate measuring device 52 are similar to those of the pressure sensor 12a, ultrasonic probe 12, transmitter 14, receiver 16, Doppler signal forming unit 18, heart rate measuring unit 20, waveform forming unit 22, output control unit 24, and display 26 in Fig. 1, so a description thereof will be omitted.

[0091] The communication unit 60 is composed of, for example, a network adapter and has a function of communicating with the late transient bradycardia prediction device 54 via a communication line 56. The communication unit 60 as a measurement device side communication unit transmits the measurement result of the heart rate measurement unit 20, in other words, information indicating the time change in the fetal heart rate of the target fetus (for example, the target fetal heart rate waveform TFHR (see FIG. 7)) to the late transient bradycardia prediction device 54. The communication unit 60 may also transmit the measurement result of the pressure sensor 12a, in other words, information indicating the time change in the uterine contraction pressure of the mother of the target fetus (for example, the target uterine contraction pressure waveform TUC (see FIG. 7)) to the late transient bradycardia prediction device 54.

[0092] The delayed transient bradycardia prediction device 54 includes a delayed transient bradycardia prediction unit 32, a memory 62 in which the learning model 30 is stored, an output control unit 64, a display 66, and a communication unit 68. The configurations and functions of the learning model 30 and the delayed transient bradycardia prediction unit 32 included in the delayed transient bradycardia prediction device 54 are similar to those of the learning model 30 and the delayed transient bradycardia prediction unit 32 in FIG. 1, and therefore will not be described.

[0093] The memory 62 includes non-volatile memory such as a hard disk drive (HDD), eMMC, and ROM, and volatile memory such as a random access memory (RAM). The display 66 includes, for example, a liquid crystal display. Various screens are displayed on the display 66 according to the control from the output control unit 64.

[0094] The delayed transient bradycardia prediction unit 32 of the delayed transient bradycardia prediction device 54 inputs the time change of the fetal heart rate of the target fetus during the target period TP (see FIG. 7 ) transmitted from each fetal heart rate measurement device 52 included in the delayed transient bradycardia prediction system 50 to the learning model 30. Then, the delayed transient bradycardia prediction unit 32 calculates the time change of the target time t PT (See Figure 7) Predict whether or not the target fetus will experience delayed transient bradycardia.

[0095] The output control unit 64 as a prediction device side notification unit notifies the delayed transient bradycardia prediction unit 32 of the prediction target time t PT If it is predicted that the target fetus will have a delayed transient bradycardia at the predicted time t PT In the above, the output control unit 64 notifies a user such as a doctor that the target fetus may experience a delayed transient bradycardia after a predetermined time I. For example, the output control unit 64 causes the display 66 to display a message 44 as shown in FIG.

[0096] The process of predicting whether or not a delayed transient bradycardia will occur by the delayed transient bradycardia prediction unit 32 using the learning model 30 requires a large amount of calculation. Therefore, the processor of the device having the learning model 30 and the delayed transient bradycardia prediction unit 32 is required to have high calculation capabilities. In the delayed transient bradycardia prediction system 50, the learning model 30 and the delayed transient bradycardia prediction unit 32 are provided in the delayed transient bradycardia prediction device 54, and are not provided in each fetal heart rate measurement device 52. Therefore, according to the delayed transient bradycardia prediction system 50, it is sufficient to provide only the delayed transient bradycardia prediction device 54 with a processor having high calculation capabilities, and it is not necessary to provide a processor having high calculation capabilities in the processor of each fetal heart rate measurement device 52. As a result, it is possible to predict whether or not a delayed transient bradycardia will occur in each target fetus that is the measurement target of each fetal heart rate measurement device 52, without increasing the cost of each fetal heart rate measurement device 52.

[0097] In addition, by displaying the judgment results of the delayed transient bradycardia prediction unit 32 for each target fetus that is the measurement target of each fetal heart rate measuring device 52 on the display 66, the user can consolidate and check multiple judgment results for each target fetus on the display 66.

[0098] The communication unit 68 is composed of, for example, a network adapter and has a function of communicating with each fetal heart rate measurement device 52 via the communication line 56. The communication unit 68 as a prediction device side communication unit receives a prediction target time t PT In this case, the output control unit 24 as a measurement device side notification unit of the fetal heart rate measurement device 52 may notify the fetal heart rate measurement device 52 of the occurrence of late transient bradycardia at the prediction target time t PT The user is notified that there is a possibility that the target fetus may experience late transient bradycardia during the

[0099] In this manner, the fetal heart rate of the target fetus is measured by the fetal heart rate measuring device 52, and the delayed transient bradycardia prediction device 54 predicts whether or not the target fetus will experience delayed transient bradycardia at the prediction target time t PT The fetal heart rate measuring device 52 may notify the user that the target fetus may have a late transient bradycardia at the prediction target time t PT It can be confirmed that there is a possibility that the target fetus may experience delayed transient bradycardia.

[0100] The above describes the delayed transient bradycardia prediction device, delayed transient bradycardia prediction system, and delayed transient bradycardia prediction program of the present embodiment. However, the delayed transient bradycardia prediction device, delayed transient bradycardia prediction system, and delayed transient bradycardia prediction program of the present embodiment are not limited to the above embodiments, and various modifications are possible as long as they do not deviate from the spirit thereof. [Explanation of symbols]

[0101] 10,54 Late transient bradycardia prediction device, 12 Ultrasound probe, 12a Pressure sensor, 14 Transmitting unit, 16 Receiving unit, 18 Doppler signal forming unit, 20 Heart rate measuring unit, 22 Waveform forming unit, 24,64 Output control unit, 26,66 Display, 28,62 Memory, 30 Learning model, 30a First learning model, 30b Second learning model, 32 Late transient bradycardia prediction unit, 50 Late transient bradycardia prediction system, 52 Fetal heart rate measuring device, 56 Communication line, 60,68 Communication unit.

Claims

1. a heart rate measuring unit that measures the fetal heart rate over time; a delayed transient bradycardia prediction unit that predicts whether or not a delayed transient bradycardia will occur in the target fetus at a prediction target time point that is a time point that is the predetermined time after the target period, based on output data from a learning model that is trained to predict and output whether or not a delayed transient bradycardia will occur in the target fetus using as learning data a time change in the heart rate of the target fetus measured in a pre-bradycardia period that is a period a predetermined time before a time point when a delayed transient bradycardia occurred in the fetus in the past; a notification unit that notifies a user when the delayed transient bradycardia prediction unit predicts that the target fetus will experience delayed transient bradycardia at the prediction target time point; A delayed transient bradycardia prediction device comprising:

2. A uterine contraction pressure measuring unit that measures the uterine contraction pressure of the mother of the fetus over time; Further equipped with The learning model is trained using training data further including a time change in uterine contraction pressure of the fetus and the mother measured in the period before the bradycardia occurrence; The delayed transient bradycardia prediction unit predicts whether or not the target fetus will experience delayed transient bradycardia at the prediction target time point based on output data of the learning model when a time change in the uterine contraction pressure of the mother of the target fetus during the target period measured by the uterine contraction pressure measurement unit is input to the learning model.

2. The delayed transient bradycardia prediction device according to claim 1 .

3. The delayed transient bradycardia prediction unit includes: a first learning model that is trained to predict and output whether or not a delayed transient bradycardia will occur in the fetus after the first predetermined time from the bradycardia pre-onset period using, as learning data, a time change in the heart rate of the fetus measured in a first pre-bradycardia period that is a period before the bradycardia onset time, and outputting the prediction data when a time change in the heart rate of the target fetus in a first target period that is a period before the first predetermined time from the prediction target time is input to the first learning model; and a second learning model that is trained to predict and output whether or not a delayed transient bradycardia will occur in the fetus after the second predetermined time from the bradycardia pre-onset period, using as learning data a time change in the heart rate of the fetus measured in a second bradycardia pre-onset period that is a period before the second predetermined time, the second predetermined time being shorter than the first predetermined time, and outputting the prediction data when a time change in the heart rate of the target fetus in a second target period that is a period before the second predetermined time from the prediction target time is input; predicting whether or not late transient bradycardia will occur in the target fetus at the prediction target time point based on the above.

2. The delayed transient bradycardia prediction device according to claim 1 .

4. The notification unit notifies the user in a first manner when the first learning model outputs output data indicating that late transient bradycardia will occur in the target fetus at the prediction target time point, and then notifies the user in a second manner different from the first manner when the second learning model outputs output data indicating that late transient bradycardia will occur in the target fetus at the prediction target time point.

4. The delayed transient bradycardia prediction device according to claim 3.

5. A fetal heart rate measuring device; A delayed transient bradycardia prediction device, A delayed transient bradycardia prediction system comprising: The fetal heart rate measuring device includes: a heart rate measuring unit that measures the fetal heart rate over time; A measurement device side communication unit that transmits the measurement result of the heart rate measurement unit to the delayed transient bradycardia prediction device; Equipped with The delayed transient bradycardia prediction device comprises: a delayed transient bradycardia prediction unit that predicts whether or not a delayed transient bradycardia will occur in the target fetus at a prediction target time point that is a time point that is the predetermined time after the target period, based on output data from a learning model that is trained to predict and output whether or not a delayed transient bradycardia will occur in the target fetus using as learning data a time change in the heart rate of the target fetus measured in a pre-bradycardia period that is a period a predetermined time before a time point when a delayed transient bradycardia occurred in the fetus in the past; a prediction device side notification unit that notifies a user when the delayed transient bradycardia prediction unit predicts that the target fetus will experience delayed transient bradycardia at the prediction target time point; Equipped with A delayed transient bradycardia prediction system comprising:

6. The delayed transient bradycardia prediction device comprises: a prediction device side communication unit that notifies the measurement device when the delayed transient bradycardia prediction unit predicts that the target fetus will experience delayed transient bradycardia at the prediction target time point; Further equipped with The fetal heart rate measuring device includes: a measurement device side notifying unit that notifies a user that a delayed transient bradycardia is predicted to occur in the target fetus at the prediction target time in response to a notification from the delayed transient bradycardia prediction device; Further comprising: The delayed transient bradycardia prediction system according to claim 5 .

7. Computer, a heart rate measuring unit that measures the fetal heart rate over time; a delayed transient bradycardia prediction unit that predicts whether or not a delayed transient bradycardia will occur in the target fetus at a prediction target time point that is a time point that is the predetermined time after the target period, based on output data from a learning model that is trained to predict and output whether or not a delayed transient bradycardia will occur in the target fetus using as learning data a time change in the heart rate of the target fetus measured in a pre-bradycardia period that is a period a predetermined time before a time point when a delayed transient bradycardia occurred in the fetus in the past; a notification unit that notifies a user when the delayed transient bradycardia prediction unit predicts that the target fetus will experience delayed transient bradycardia at the prediction target time point; A delayed transient bradycardia prediction program characterized by operating as follows.

Images