JP7599704B2 - Information processing device, information processing program, and information processing method - Google Patents

Description

The present invention relates to an information processing device, an information processing program, and an information processing method.

A fetal heart rate electrocardiogram (CTG: cardiotocography) is a waveform diagram that continuously records fetal heart rate and uterine contractions (labor pains) in chronological order, measured by a fetal monitoring device. Doctors estimate the condition of the fetus by checking the changes in uterine contractions and fetal heart rate based on the CTG. During delivery, the fetus may fall into a state of hypoxia due to various factors (e.g., placental abruption, umbilical cord compression, etc.). As a result, the fetus may develop ischemic encephalopathy due to acidosis (acidification of the blood) during the fetal and postnatal period, which may lead to physical disabilities and developmental disorders in the future. For this reason, pregnant women are continuously and intermittently observed for signs of fetal hypoxia based on the CTG during the delivery process and throughout pregnancy.

When it comes to interpreting CTG, the Japan Society of Obstetrics and Gynecology has provided criteria for judging baseline heart rate, baseline fine fluctuations in heart rate, and transient fluctuations. Doctors make their judgments based on these criteria. However, CTGs are interpreted visually by obstetricians and midwives, and there is a large degree of individual variability in interpretation based on differences in the experience of obstetricians and midwives. Therefore, despite the existence of such criteria, there are still many cases of ischemic encephalopathy.

Previously, technology has been proposed that determines parameters such as baseline fine fluctuations and transient fluctuations in heart rate based on fetal heart rate and maternal uterine activity, and indicates the level of risk to the fetus based on non-reassuring characteristics (abnormalities) defined for each parameter (see, for example, Patent Document 1).

The technology disclosed in Patent Document 1 determines non-reassuring characteristics based on standard thresholds for parameters such as baseline heart rate fluctuations and transient fluctuations. However, this technology does not disclose specific details for extracting and determining parameters such as baseline heart rate fluctuations and transient fluctuations from the fetal heart rate. Therefore, there is a demand for technology that can uniformly determine and identify the heart rate baseline, heart rate fluctuations, and transient fluctuations.

Special Publication No. 2020-500077

The present invention aims to provide auxiliary information that can be used to aid in the interpretation of CTG.

The information processing device according to the present invention is an information processing device that generates auxiliary information used to assist in the interpretation of a fetal heart rate tocogram, and includes an acquisition unit that acquires a fetal heart rate tocogram, a waveform signal extraction unit that extracts a waveform signal of a fetal heart rate waveform included in the acquired fetal heart rate tocogram, a low-frequency extraction unit that extracts a predetermined low-frequency component signal from the extracted waveform signal, and a baseline interval determination unit that determines a baseline determination target interval for determining a fetal heart rate baseline as auxiliary information based on the extracted low-frequency component signal, and is characterized in that the baseline interval determination unit extracts, in the low-frequency component signal, an interval from a predetermined first starting point to the elapse of a first predetermined time required for determining the fetal heart rate baseline as a first determination interval, and determines the first determination interval as a first constituent interval constituting the baseline determination target interval when the variance value of the low-frequency component signal in the first determination interval is smaller than a first threshold value.

The present invention can provide auxiliary information that can be used to aid in the interpretation of CTG.

FIG. 2 is a schematic diagram showing an example of a cardiotocogram used in the information processing device according to the present invention. 1 is a network connection diagram showing an embodiment of an information processing device according to the present invention; 1 is a functional block diagram showing an embodiment of an information processing apparatus according to the present invention; 4 is a functional block diagram of a noise removal unit included in the information processing device of FIG. 3. 1 is a flowchart illustrating an embodiment of an information processing method according to the present invention. 6 is a flowchart of a pre-processing included in the information processing method of FIG. 5 . 6A to 6C are schematic diagrams showing examples of signal states at each stage of the pre-processing shown in FIG. 5 . 6 is a flowchart of a portion of a baseline determination process included in the information processing method of FIG. 5 . 9 is a flowchart of another part of the baseline determination process of FIG. 8 . 9 is a schematic diagram showing an example of main processing of the baseline determination processing in FIG. 8 . 6 is a flowchart of a fine variation determination process included in the information processing method of FIG. 5 . 12A to 12C are schematic diagrams showing an example of main processes in the fine variation determination process of FIG. 11 . 6 is a flowchart of a part of a transient fluctuation learning process included in the information processing method of FIG. 5 . 14 is a flowchart of another part of the transient fluctuation learning process of FIG. 13 . 14 is a flowchart of still another part of the transient fluctuation learning process of FIG. 13 . FIG. 14 is a schematic diagram showing an example of main processing of the transient fluctuation section identification processing of FIG. 13 . 14 is a schematic diagram illustrating a method of acquiring learning data in the transient fluctuation learning process of FIG. 13 . FIG. 6 is a flowchart of a transient fluctuation identification process included in the information processing method of FIG. 5 . 19 is a schematic diagram illustrating each identification interval of the transient fluctuation identification process of FIG. 18 . FIG. 13 is a schematic diagram showing an example of a determination result when the present method is applied to an actual cardiotocography. FIG. 9 is a schematic diagram showing the result of identifying a baseline determination target section when the baseline determination process of FIG. 8 is performed on an actual cardiotocogram; FIG. 9 is a schematic diagram showing the fetal heart rate baseline determination result when the baseline determination process of FIG. 8 is performed on an actual fetal cardiotocography. 12 is a schematic diagram showing the determination result of fetal heart rate baseline fine fluctuations when the fine fluctuation determination process of FIG. 11 is performed on an actual fetal heart rate-tocogram. FIG.

Below, with reference to the drawings, an information processing device (hereinafter referred to as "the device"), an information processing program (hereinafter referred to as "the program"), and an information processing method (hereinafter referred to as "the method") according to the present invention will be described in detail.

The present invention relates to the generation of auxiliary information used to aid in the interpretation of fetal cardiotocography. The present invention also relates to the determination of fetal heart rate baseline (hereinafter referred to as "baseline"), the determination of fetal heart rate baseline fine fluctuations (hereinafter referred to as "fine fluctuations"), the generation of a classifier for transient fluctuations, and the identification of transient fluctuations.

A "cardiotocography (CTG)" is a waveform diagram that continuously records the fetal heart rate and uterine contractions (labor) measured by a labor monitoring device in a chronological order, and shows the changes in uterine contractions and fetal heart rate to estimate the state of the fetus. A cardiotocography (CTG) includes a waveform corresponding to the fetal heart rate (hereinafter referred to as the "fetal heart rate waveform") and a waveform corresponding to uterine contractions.

FIG. 1 is a schematic diagram showing an example of a CTG used in an information processing device according to the present invention.
In the figure, the upper waveform shows the fetal heart rate waveform, and the lower waveform shows the waveform corresponding to uterine contractions. The vertical axis of the upper part of the figure shows the fetal heart rate (bpm: beats per minute), and the horizontal axis shows the elapsed time (sec).

"Auxiliary information" is information used to assist in interpreting CTG. In this embodiment, the auxiliary information includes a baseline determination target section, a fine variation determination target section, and a transient variation section. In this embodiment, the auxiliary information includes not only the baseline determination results, fine variation determination results, and transient variation identification results obtained by the present invention, but also learning data required for identifying transient variations (i.e., data that assists in identification).

The "baseline (fetal heart rate baseline)" is the approximate average heart rate of the fetus over a 10-minute period. Medically, the baseline is determined by a section of the waveform corresponding to the fetal heart rate that lasts for 2 minutes or more (i.e., the baseline determination section), excluding the sections of transient fluctuations and the sections of increased fine fluctuations. In this embodiment, the baseline is determined to be one of three types: "normal," "tachycardia," or "bradycardia." The baseline is determined to be "bradycardia" if it is less than 110 bpm, "tachycardia" if it is more than 160 bpm (beats per minute), and "normal" if it is between 110 bpm and 160 bpm. The terms "normal," "tachycardia," and "bradycardia" are well known, so a description of the baseline will be omitted.

"Fetal heart rate baseline fine fluctuations" are fine fluctuations in the baseline, and by definition refer to fetal heart rate fluctuations of two or more cycles per minute that have no regularity in either amplitude or frequency. Medically, fine fluctuations are determined in sections where the baseline is legible. In this embodiment, fine fluctuations are determined as one of four types: "moderate (normal)", "increased", "decreased", or "absent". As "moderate (normal)", "increased", "decreased", and "absent" are well known in the art, a description of these fine fluctuations will be omitted.

"Transient fluctuations" are rapid fluctuations ("increases" and "decreases") in fetal heart rate. In this embodiment, transient fluctuations include five types: "transient tachycardia," "early transient bradycardia," "late transient bradycardia," "variable transient bradycardia," and "prolonged transient bradycardia." Each type of transient fluctuation is well known, so a description of them will be omitted.

The "baseline determination target section" is a section that is identified as a baseline in the present invention, and is a section for which the baseline is determined in the present invention. Details of the baseline determination target section will be described later.

The "subject section for determining fine fluctuations" is a section for determining fine fluctuations in the present invention. The subject section for determining fine fluctuations is specified within the subject section for determining the baseline. Details of the subject section for determining fine fluctuations will be described later.

A "transient fluctuation section" is a section that is identified in the present invention as a section in which a transient fluctuation occurs. Details of the transient fluctuation section will be described later.

The "classifier" in the present invention is a model based on a machine learning algorithm (for example, a known machine learning algorithm such as CNN (Convolutional Neural Network)) that identifies transient fluctuations. In the present invention, the classifier identifies "other than transient fluctuations," "transient tachycardia," "early transient bradycardia and late transient bradycardia," "variable transient bradycardia (mild)," "variable transient bradycardia (severe)," and "prolonged transient bradycardia." In this embodiment, the learning data for the classifier is generated based on the transient fluctuation section, as described below.

<<Information processing device>>
First, an embodiment of the present device will be described.

Figure 2 is a network connection diagram showing an embodiment of this device.

This device 1 generates auxiliary information used to assist in interpreting CTG. This device 1 is connected to an external device 2 via a network N. The specific configuration and operation of this device 1 will be described later.

The external device 2 is a device that generates and/or stores a CTG, for example a fetal monitoring device.

The network N is, for example, a communication network such as the Internet, a mobile communication network, a LAN (Local Area Network), a WAN (Wide Area Network), Wi-Fi (registered trademark), or Bluetooth (registered trademark).

<Configuration of information processing device>
FIG. 3 is a functional block diagram showing an embodiment of the device 1. As shown in FIG.

The device 1 is realized, for example, by a personal computer. The program runs on the device 1, and the program cooperates with the hardware resources of the device 1 to realize the method.

Here, by having a computer (not shown) execute this program, the program can cause the computer to function in the same manner as the device 1, and cause the computer to execute the method.

The device 1 includes a communication unit 11, a memory unit 12, a control unit 13, an operation unit 14, and a display unit 15.

The communication unit 11 acquires the CTG from the external device 2 via the network N. The communication unit 11 is configured, for example, by a communication module and a communication interface. The CTG is stored in the memory unit 12.

The storage unit 12 stores information (e.g., various thresholds) necessary for the device 1 to execute the method described below. The storage unit 12 is composed of, for example, a recording device such as a hard disk drive (HDD) or solid state drive (SSD) provided in the device 1, a volatile memory such as a random access memory (RAM), and/or a portable storage medium such as a flash memory.

The control unit 13 controls the operation of the entire device 1 and executes the method described below. The control unit 13 is composed of, for example, a CPU (Central Processing Unit) provided in the device 1, a volatile memory such as a RAM that functions as a working area for the CPU, and a non-volatile memory such as a ROM (Read Only Memory) that stores various information such as the program. The control unit 13 includes a preprocessing unit 131, a baseline section identification unit 132, a baseline determination unit 133, a fine fluctuation section identification unit 134, a fine fluctuation determination unit 135, a transient fluctuation section identification unit 136, a data acquisition unit 137, a classifier generation unit 138, and a classification unit 139.

The pre-processing unit 131 performs pre-processing on the CTG, i.e., acquiring the CTG, extracting a waveform signal (described later), removing noise, extracting low-frequency components, and extracting high-frequency components. The pre-processing unit 131 includes an acquisition unit 1311, a waveform signal extraction unit 1312, a noise removal unit 1313, a low-frequency extraction unit 1314, and a high-frequency extraction unit 1315.

The acquisition unit 1311 acquires the CTG from the external device 2 via the communication unit 11. In this embodiment, the acquisition unit 1311 acquires the CTG in real time. The specific operation of the acquisition unit 1311 will be described later.

The waveform signal extraction unit 1312 extracts a fetal heart rate waveform signal (hereinafter referred to as the "waveform signal") from the CTG acquired by the acquisition unit 1311. The specific operation of the waveform signal extraction unit 1312 will be described later.

Figure 4 is a functional block diagram of the noise removal unit 1313.

The noise removal unit 1313 removes noise from the waveform signal extracted by the waveform signal extraction unit 1312 (see FIG. 3). The noise removal unit 1313 includes a primary filter 1313a, a primary low-pass filter 1313b, a primary calculation unit 1313c, a secondary low-pass filter 1313d, a secondary calculation unit 1313e, and a secondary filter 1313f. In this embodiment, the noise removal unit 1313 functions as the primary filter 1313a, the primary low-pass filter 1313b, the primary calculation unit 1313c, the secondary low-pass filter 1313d, the secondary calculation unit 1313e, and the secondary filter 1313f, for example, by executing the program. The specific operation of the noise removal unit 1313 will be described later.

Return to Figure 3.
The low frequency extractor 1314 extracts a low frequency component signal in a predetermined range from the waveform signal from which noise has been removed. The specific operation of the low frequency extractor 1314 will be described later.

The high-frequency extraction unit 1315 extracts a high-frequency component signal in a predetermined range from the waveform signal from which noise has been removed. The specific operation of the high-frequency extraction unit 1315 will be described later.

The baseline section identification unit 132 identifies a baseline determination target section for determining the baseline as auxiliary information based on the extracted low-frequency component signal. The specific operation of the baseline section identification unit 132 will be described later.

The baseline determination unit 133 determines the baseline in the baseline determination target section based on the identified baseline determination target section. The specific operation of the baseline determination unit 133 will be described later.

The fine fluctuation section identification unit 134 identifies a fine fluctuation judgment target section for judging fine fluctuations as auxiliary information based on the extracted high frequency component signal. The specific operation of the fine fluctuation section identification unit 134 will be described later.

The fine fluctuation determination unit 135 determines fine fluctuations in the fine fluctuation determination target section based on the identified fine fluctuation determination target section. The specific operation of the fine fluctuation determination unit 135 will be described later.

The transient fluctuation section identifying unit 136 identifies one or more transient fluctuation sections as auxiliary information based on the extracted low-frequency component signal in order to obtain learning data for a classifier that identifies transient fluctuations. The specific operation of the transient fluctuation section identifying unit 136 will be described later.

The data acquisition unit 137 acquires learning data for the classifier based on the identified transient fluctuation section. The specific operation of the data acquisition unit 137 will be described later.

"Learning data" is training data used to train the classifier to identify transient fluctuations, and is, for example, waveform data (low-frequency component signal) of a portion of the low-frequency component signal (including at least a portion of the transient fluctuation portion).

The discriminator generation unit 138 generates a discriminator based on the training data. The specific operation of the discriminator generation unit 138 will be described later.

The identification unit 139 uses a discriminator to identify transient fluctuations. The specific operation of the identification unit 139 will be described later.

<<Information processing method>>
Next, the operation of the present device 1, i.e. the present method executed by the present device 1, will be described with reference to Figs.

Figure 5 is a flowchart illustrating an embodiment of the method.

The present device 1 executes pre-processing (S1), baseline determination processing (S2), fine fluctuation determination processing (S3), transient fluctuation learning processing (S4), and transient fluctuation identification processing (S5). That is, the present method includes pre-processing (S1), baseline determination processing (S2), fine fluctuation determination processing (S3), transient fluctuation learning processing (S4), and transient fluctuation identification processing (S5). Here, the baseline determination processing (S2), fine fluctuation determination processing (S3), and transient fluctuation identification processing (S5) are processing performed mainly in obstetrics settings. On the other hand, the transient fluctuation learning processing (S4) is processing performed in the learning setting of the classifier, and is processing performed in advance before the transient fluctuation identification processing (S5). The pre-processing (S1) is a process that is executed in advance and is common to the baseline determination process (S2), the fine fluctuation determination process (S3), the transient fluctuation learning process (S4), and the transient fluctuation identification process (S5).

<Pretreatment>
FIG. 6 is a flowchart of the pre-processing (S1).
FIG. 7 is a schematic diagram showing an example of the state of a signal at each stage of pre-processing (S1).

"Preprocessing (S1)" is a process of extracting a waveform signal from the CTG, removing noise from the waveform signal, and generating a low-frequency component signal and a high-frequency component signal.

First, the acquisition unit 1311 acquires a CTG from the external device 2 via the communication unit 11 (S101). In this embodiment, the acquisition unit 1311 continuously acquires a CTG for one pregnant woman.

Next, the waveform signal extraction unit 1312 extracts the waveform signal of the fetal heart rate waveform contained in the acquired CTG (S102).

Then, the noise removal unit 1313 applies the primary filter 1313a to the waveform signal to remove the main abrupt noise from the waveform signal (S103). The threshold of the primary filter 1313a is, for example, 10 bpm to 190 bpm. In other words, the primary filter 1313a removes abrupt noise less than 10 bpm and abrupt noise more than 190 bpm from the waveform signal.

Next, the noise removal unit 1313 applies the primary low-pass filter 1313b to the waveform signal from which the main abrupt noise has been removed (hereinafter referred to as the "primary noise-removed signal") to extract a low-frequency band component signal (hereinafter referred to as the "primary low-frequency component signal") from the primary noise-removed signal (S104). As a result, the trend of the waveform signal that does not contain the abrupt noise that was omitted from the processing S103 by the primary filter 1313a is extracted. The cutoff frequency of the primary low-pass filter 1313b is, for example, 0.03 Hz.

Next, the noise removal unit 1313 uses the primary calculation unit 1313c to calculate the absolute value of the difference between the primary noise-removed signal and the primary low-frequency component signal (hereinafter referred to as the "primary difference") (S105). As a result, the fluctuation corresponding to the baseline is canceled to a certain extent, and the noise remaining after process S104 becomes clear.

Next, the noise removal unit 1313 applies the secondary low-pass filter 1313d to the signal obtained from the absolute value of the calculated primary difference (hereinafter referred to as the "primary difference signal") to extract a low-frequency band component signal (hereinafter referred to as the "secondary low-frequency component signal") from the primary difference signal (S106). As a result, the fluctuation component corresponding to the baseline containing noise clarified by process S105 is extracted. The cutoff frequency of the secondary low-pass filter 1313d is, for example, 0.06 Hz.

Next, the noise removal unit 1313 uses the secondary calculation unit 1313e to calculate the difference between the primary difference signal and the secondary low-frequency component signal (hereinafter referred to as the "secondary difference") (S107). As a result, the aforementioned fluctuation component is cancelled. Therefore, the noise that has been made clear by the process S105 becomes even clearer and detectable.

Next, the noise removal unit 1313 applies the secondary filter 1313f to the signal obtained by the secondary difference (hereinafter referred to as the "secondary difference signal") to extract a secondary difference signal (hereinafter referred to as the "noise-removed signal") from which minute abrupt noise remaining after processing S103 has been removed (S108). Here, the threshold of the secondary filter is 10 bpm. In other words, the secondary filter 1313f removes abrupt noise exceeding 10 bpm from the secondary difference signal.

Then, the low-frequency extraction unit 1314 applies a predetermined low-pass filter to the noise-removed signal to extract a predetermined low-frequency component signal from the noise-removed signal (S109). Here, the low-pass filter applied in this process S109 is, for example, a Butterworth low-pass filter (order "4"). The cutoff frequency of the low-pass filter is, for example, 0.03 Hz. The extracted low-frequency component signal is stored, for example, in the storage unit 12.

Meanwhile, the high-frequency extraction unit 1315 applies a band-pass filter of a predetermined range to the noise-removed signal to extract a high-frequency component signal of a predetermined range from the noise-removed signal (S110). Here, the band-pass filter applied in this process S110 is, for example, a Butterworth band-pass filter (order "4"). The cut-off frequency (band-pass frequency) of the band-pass filter is 0.02 Hz to 0.8 Hz. The extracted high-frequency component signal is stored, for example, in the memory unit 12.

Here, when a doctor interprets a CTG, the doctor does not interpret the CTG by clearly distinguishing between the low-frequency components (slow fluctuations) and high-frequency components (fine fluctuations) of the fetal heart rate waveform, but rather interprets the CTG by using both components appropriately depending on the application. That is, a part of the shape of the fetal heart rate waveform for which the doctor judges the baseline overlaps with a part of the shape of the fetal heart rate waveform for which the doctor judges fine fluctuations. Therefore, in this embodiment, the high-frequency component signal includes a higher frequency band than the low-frequency component signal, and the low-frequency component signal includes a lower frequency band than the high-frequency component signal. That is, a part of the frequency band of the high-frequency component signal is common to a part of the frequency band of the low-frequency component signal. As a result, the present device 1 can achieve judgments (baseline judgment, fine fluctuation judgment) that are close to those of a doctor's visual inspection.

The pre-processing may be performed continuously in chronological order, or may be performed for each waveform signal within a given period of time (e.g., 3 minutes).

In addition, for example, the cutoff frequency of each filter provided in the noise removal unit of the present invention is not limited to the present embodiment as long as the low-frequency component signal and high-frequency component signal required for the present method (preprocessing) are obtained.

Furthermore, in preprocessing, if the remaining noise is clear enough to be detectable (e.g., to be extractable by a secondary filter) at the time when the absolute value of the first-order difference is calculated, the noise removal unit may omit steps S106 and S107.

<Baseline Determination Processing>
FIG. 8 is a flowchart of a part of the baseline determination process (S2).
FIG. 9 is a flowchart of another part of the baseline determination process (S2).
FIG. 10 is a schematic diagram showing an example of main processing steps of the baseline determination process (S2).

In FIG. 10, the first constituent section, which will be described later, is indicated by a solid arrow, and the first non-constituent section, which will be described later, is indicated by a dashed arrow. For ease of explanation, the first constituent section and the first non-constituent section are numbered chronologically from "1" to "10" in the figure. The figure shows that the first constituent sections "1" to "3" are consecutive. The figure also shows that the first non-constituent sections "4" to "10" are seven consecutive times. Furthermore, the figure shows that the first constituent sections "1" to "3" have been identified as the section subject to baseline determination.

The "baseline determination process (S2)" is a process of identifying a baseline determination target section for determining the baseline based on the low-frequency component signal, and determining the baseline in the baseline determination target section based on the baseline determination target section.

First, the baseline interval determination unit 132 reads out from the memory unit 12 a low-frequency component signal corresponding to the fetal heart rate waveform (waveform signal) for which the baseline is to be determined (S201).

Next, the baseline section identification unit 132 sets the first point on the time series in the low-frequency component signal as the starting point, and extracts the section from the starting point until the first predetermined time has elapsed as the first determination section (S202). Here, the starting point in the baseline determination process (S2) is an example of the first starting point in the present invention.

The "first predetermined time" is the time required to determine the baseline, which is two minutes in medical terms. Here, the fetal heart rate waveform in the CTG may not have a two-minute baseline determination target section. In this case, the doctor determines the baseline based on a baseline determination target section that is less than two minutes but close to two minutes (e.g., 90 seconds). Therefore, the first predetermined time is preferably about 70 to 110 seconds, and in this embodiment, the first predetermined time is set to 80 seconds. That is, in this embodiment, the first determination section is a section from the starting point to 80 seconds.

The starting point set by the baseline section determination unit in the present invention is not limited to the first point on the time series. That is, for example, the baseline section determination unit in the present invention may set a point selected by the user of the device as the starting point.

Next, the baseline section identification unit 132 calculates the variance value of the low-frequency component signal in the first judgment section (hereinafter referred to as the "first variance value") (S203).

Next, the baseline section determination unit 132 compares the first variance value with a predetermined first threshold value to determine whether the first variance value is smaller than the first threshold value (S204).

The "first threshold" is a threshold for determining whether the first determination section is appropriate as a baseline determination target section. In this embodiment, the first threshold is set to "7." The first threshold is, for example, set in advance by a user of the device 1 and stored in the memory unit 12.

When the first variance value is smaller than the first threshold value ("Y" in S204), the baseline section identification unit 132 identifies the first determination section as a section constituting the baseline determination target section (hereinafter referred to as the "first constituent section") (S205).

On the other hand, when the first variance value is equal to or greater than the first threshold value ("N" in S204), the baseline section identification unit 132 identifies the first determination section as a section that does not constitute the baseline determination target section (hereinafter referred to as the "first non-constituent section") (S206).

Next, the baseline section identification unit 132 determines whether the first non-constituent section occurs consecutively a predetermined number of times (S207).

The "predetermined number of times" is the number of times that a first non-constituent section is consecutive, causing some sections of two first constituent sections that follow one another in the time series to no longer have something in common. In other words, the predetermined number of times is the number of times that a common section disappears between a certain first constituent section and the next first constituent section in the time series (the two first constituent sections are separated). The predetermined number of times is set, for example, based on the first predetermined time. That is, for example, when the first predetermined time is 80 seconds and the second predetermined time described below is 10 seconds, the predetermined number of times is set to 9 times.

When the first non-constituent interval does not occur consecutively the predetermined number of times ("N" in S207), the baseline interval determination unit 132 determines whether or not the interval in the low-frequency component signal that is later in the time series than the first determination interval continues for a second predetermined time or more (S208). In other words, the baseline interval determination unit 132 determines whether or not the acquisition unit 1311 has continuously acquired the fetal heart rate waveform for the second predetermined time from the end point of the first determination interval.

The "second predetermined time" is a time shorter than the first predetermined time, and is set to shift the first judgment interval later in the time series. The second predetermined time is set so that two consecutive first judgment intervals in the time series have a portion in common. In this embodiment, the second predetermined time is set to 10 seconds.

Note that the second predetermined time in the present invention is not limited to 10 seconds. That is, for example, the second predetermined time in the present invention may be 5 seconds or 15 seconds.

When the later section in the time series continues for a second predetermined time or more ("Y" in S208), the baseline section identification unit 132 sets the point where the second predetermined time has elapsed from the start point as a new start point, and extracts the section from the new start point to the point where the first predetermined time has elapsed as a new first judgment section (S209). In other words, the baseline section identification unit 132 updates the position of the start point of the first judgment section to the position of a point that is the second predetermined time after that position. In other words, the baseline section identification unit 132 shifts the first judgment section back in the time series by the second predetermined time. Next, the baseline determination process (S2) returns to process S203.

When the first non-constituent intervals occur consecutively a predetermined number of times ("Y" in S207), if multiple first constituent intervals have been identified, the baseline interval identification unit 132 identifies an interval obtained by connecting first constituent intervals having a common interval among the multiple first constituent intervals as a baseline determination target interval (S210). On the other hand, if one first constituent interval has been identified, the baseline interval identification unit 132 identifies the first constituent interval as a baseline determination target interval. In this way, by identifying the baseline determination target interval, the present device 1 can identify an interval that is close to the interval in which the doctor actually determines the baseline as the baseline determination target interval. The baseline determination target interval is associated with, for example, a low-frequency component signal and a waveform signal that is the basis of the low-frequency component signal, and is stored in the storage unit 12 as auxiliary information.

Next, the baseline determination unit 133 calculates the average heart rate for each first component section that constitutes the baseline determination target section in the low-frequency component signal (S211). The average heart rate is associated with, for example, the first component section and the baseline determination target section constituted by the first component section, and is stored in the storage unit 12 as auxiliary information.

Next, the baseline determination unit 133 compares the average heart rate for each first component section with a predetermined threshold value to determine the baseline for each first component section (S212). Specifically, when the average heart rate is smaller than "110 bpm", the baseline determination unit 133 determines that "the fetal heart rate of the baseline corresponding to the first component section is bradycardia", when the average heart rate is larger than "160 bpm", the baseline determination unit 133 determines that "the fetal heart rate of the baseline corresponding to the first component section is tachycardia", and when the average heart rate is "110 bpm" or more and "160 bpm" or less, the baseline determination unit 133 determines that "the fetal heart rate of the baseline corresponding to the first component section is normal". Here, the threshold value for determining the baseline is medically defined as "110 bpm" and "160 bpm", and is stored in advance in the storage unit 12. The determination result is associated with, for example, the first component section and the baseline determination target section formed by the first component section, and is stored as auxiliary information in the storage unit 12. Next, the baseline determination process (S2) returns to process S208.

When the later section in the time series does not continue for a second predetermined time or longer ("N" in S208), the baseline section determination unit 132 determines whether the value of the standby count described below is "0" or "1" (S213).

When the standby count value is "0", the baseline section identification unit 132 adds "1" to the standby count after waiting for a second predetermined time (S214) and executes process S208. On the other hand, when the standby count value is "1", the baseline section identification unit 132 determines whether there is a first component section that has not been identified as a baseline determination target section in the sections following the section last identified as a baseline determination target section (S215).

When there is a first component section that has not been identified as a baseline determination target section ("Y" in S215), the baseline determination process (S2) returns to process S210. On the other hand, when there is no such first component section ("N" in S215), the baseline determination process (S2) ends.

<Fine fluctuation determination process>
FIG. 11 is a flowchart of the fine variation determination process (S3).
FIG. 12 is a schematic diagram showing an example of main processing of the fine variation determination process (S3).

The "fine fluctuation determination process (S3)" is a process for identifying a section in the high frequency component signal that corresponds to the baseline determination target section as the fine fluctuation determination target section, and determining fine fluctuations based on the fine fluctuation determination target section. The fine fluctuation determination process (S3) is executed, for example, after the baseline determination target section is identified in the baseline determination process (S2).

First, the fine fluctuation section identification unit 134 reads out from the memory unit 12 the high frequency component signal corresponding to the fetal heart rate waveform (waveform signal) for which fine fluctuations are to be determined, and the baseline determination target section associated with the low frequency component signal corresponding to the same waveform signal (S301).

Next, the fine fluctuation section identification unit 134 identifies a section in the high frequency component signal that corresponds to the baseline determination target section as a fine fluctuation determination target section (S302).

Next, the fine fluctuation section identification unit 134 sets the first point (start point) of the fine fluctuation judgment section as the start point, and extracts the section from the start point until the third predetermined time has elapsed as the second judgment section (S303). Here, the start point in the fine fluctuation judgment process (S3) is an example of the second start point in the present invention.

The "third predetermined time" is the time required to determine fine fluctuations, and is set to a time shorter than the first predetermined time. The third predetermined time is set, for example, to a time during which the determination results of fine fluctuations within the fine fluctuation determination target section do not change frequently. In this embodiment, the third predetermined time is set to 40 seconds. That is, in this embodiment, the second determination section is a section of 40 seconds from the starting point.

The third predetermined time in the present invention is not limited to 40 seconds. If the third predetermined time is longer than 50 seconds, small fluctuations are averaged out, and the accuracy of the judgment of small fluctuations deteriorates. On the other hand, if the third predetermined time is shorter than 30 seconds, multiple small fluctuations are judged individually, and the judgment result changes frequently. Therefore, it is desirable that the third predetermined time in the present invention be between 30 and 50 seconds.

Next, the fine variation determination unit 135 calculates the variance value of the high frequency components in the second determination section (hereinafter referred to as the "second variance value") (S304).

Next, the fine fluctuation determination unit 135 compares the second variance value with a predetermined second threshold value to determine fine fluctuations in the second determination interval (S305). Specifically, when the second variance value is 1.3 or more and 20 or less, the fine fluctuation determination unit 135 determines that "fine fluctuations in the second determination interval are moderate," when the second variance value is greater than 20 and less than 25, when the fine fluctuations in the second determination interval are increasing, when the second variance value is 0.07 or more and less than 1.3, when the second variance value is less than 0.07, when the fine fluctuations in the second determination interval are disappearing. The determination result is, for example, associated with the corresponding second determination interval and stored in the storage unit 12 as auxiliary information.

Next, the fine fluctuation section identification unit 134 determines whether a section in the baseline determination target section that is later in the time series than the second determination section continues for a fourth predetermined time or longer (S306).

The "fourth predetermined time" is a time shorter than the third predetermined time, and is a time by which the second judgment interval is shifted later in the time series. The fourth predetermined time is set so that two consecutive second judgment intervals in the time series have a portion in common. In this embodiment, the fourth predetermined time is set to 10 seconds.

Note that the fourth predetermined time in the present invention is not limited to 10 seconds. That is, for example, the fourth predetermined time in the present invention may be 5 seconds or 15 seconds.

When the latter section continues for a fourth predetermined time or more in the time series ("Y" in S306), the fine fluctuation section identification unit 134 sets the point where the fourth predetermined time has elapsed from the start point as a new start point, and extracts the section from the new start point to the third predetermined time as a new second judgment section (S307). In other words, the fine fluctuation section identification unit 134 updates the position of the start point of the second judgment section to the position of the point four predetermined times after the same position. In other words, the fine fluctuation section identification unit 134 shifts the second judgment section back by the fourth predetermined time in the time series. Then, the fine fluctuation judgment process (S3) returns to process S304. In this way, by judging fine fluctuations while shifting the second judgment section by the fourth predetermined time, the present device 1 can judge fine fluctuations in a section of the baseline that is close to the section that a doctor actually visually observes to judge fine fluctuations from the baseline.

On the other hand, if the later section in the time series does not continue for the fourth predetermined time or longer ("N" in S306), the fine fluctuation determination process (S3) ends.

The fine fluctuation determination process may be executed each time a baseline determination target section is identified. In other words, the fine fluctuation determination process may be executed in parallel with the baseline determination target process. In this case, the fine fluctuation determination process is executed, for example, each time a baseline determination target section is identified.

<Transient Fluctuation Learning Process>
FIG. 13 is a flowchart of a part of the transient fluctuation learning process (S4).
FIG. 14 is a flowchart of another part of the transient fluctuation learning process (S4).
FIG. 15 is a flowchart of still another part of the transient fluctuation learning process (S4).
FIG. 16 is a schematic diagram showing an example of main processes of the transient fluctuation learning process (S4).

The "transient fluctuation learning process (S4)" is a process for identifying a transient fluctuation section based on a low-frequency component signal, and acquiring learning data for a discriminator based on the transient fluctuation section. As described above, the transient fluctuation learning process (S4) is a process that is executed before the transient fluctuation identification process (S5).

First, the transient fluctuation section identification unit 136 reads out from the storage unit 12 a low-frequency component signal corresponding to the fetal heart rate waveform (waveform signal) from which a transient fluctuation section is to be extracted (S401). The low-frequency component signal is, for example, acquired in advance by preprocessing (S1) and stored in the storage unit 12.

Next, the transient fluctuation section identification unit 136 sets the first point on the time series in the low-frequency component signal as the starting point, and extracts the section from the starting point until the fifth predetermined time has elapsed as the third judgment section (S402). Here, the starting point in the transient fluctuation section identification process (S4) is an example of the third starting point in the present invention.

The "fifth predetermined time" is a time set for extracting a transient fluctuation section, and is set to a time shorter than the first predetermined time. Typically, a doctor judges one transient fluctuation for one transient fluctuation section. Therefore, the fifth predetermined time is set to a time during which the judgment result of the transient fluctuation within the transient fluctuation section does not fluctuate frequently, for example. In this embodiment, the fifth predetermined time is set to 40 seconds. That is, in this embodiment, the third judgment section is a section of 40 seconds from the starting point.

The fifth predetermined time in the present invention is not limited to 40 seconds. If the fifth predetermined time is longer than 50 seconds, many sections (such as the baseline) in which no transient fluctuation occurs may be included near the start and end points of the transient fluctuation section. In addition, multiple short transient fluctuation sections may be combined into one transient fluctuation section. As a result, the accuracy of extracting the transient fluctuation judgment section may deteriorate. On the other hand, if the fifth predetermined time is shorter than 30 seconds, a situation may occur in which the proportion of sections with low heart rate fluctuations in the third judgment section increases. As a result, the amount of fluctuation in the third judgment section may be averaged, and the third judgment section may not be identified as a section that constitutes a transient fluctuation section. Therefore, it is desirable for the fifth predetermined time in the present invention to be between 30 and 50 seconds.

Next, the transient fluctuation section identification unit 136 calculates the variance value of the low-frequency components in the third judgment section (hereinafter referred to as the "third variance value") (S403).

Next, the transient fluctuation section identification unit 136 compares the third variance value with a predetermined third threshold value to determine whether the third variance value is greater than the third threshold value (S404).

The "third threshold" is a threshold for determining whether the third judgment section is appropriate as a transient fluctuation section. In this embodiment, the third threshold is set to "15." The third threshold is, for example, set in advance by the user of the device 1 and stored in the memory unit 12.

When the third variance value is greater than the third threshold value ("Y" in S404), the transient fluctuation section identification unit 136 identifies the third judgment section as a second constituent section constituting the transient fluctuation section (S405).

On the other hand, when the third variance value is equal to or less than the third threshold value ("N" in S404), the transient fluctuation section identification unit 136 identifies the third judgment section as a second non-constituent section that does not constitute a transient fluctuation section (S406).

Next, the transient fluctuation section identification unit 136 determines whether the second non-constituent section occurs consecutively a predetermined number of times (S407).

The "predetermined number of times" is the number of times that a second non-constituent section successively occurs such that a portion of two adjacent second constituent sections in the time series no longer has a common section. In other words, the predetermined number of times is the number of times that a common section disappears between a second constituent section and the next second constituent section in the time series (the two second constituent sections are separated). The predetermined number of times is set, for example, based on the fifth predetermined time. That is, for example, when the fifth predetermined time is 40 seconds, the predetermined number of times is set to 3 times.

When the second non-constituent interval does not occur consecutively the predetermined number of times ("N" in S407), the transient fluctuation interval identifying unit 136 determines whether or not the interval in the low frequency component signal that is later in the time series than the third determination interval continues for a sixth predetermined time or more (S408). In other words, the transient fluctuation interval identifying unit 136 determines whether or not the acquiring unit 1311 has acquired a fetal heart rate waveform for a sixth predetermined time from the end point of the fifth determination interval.

The "sixth predetermined time" is a time shorter than the fifth predetermined time, and is set to shift the third judgment section later in the time series. The sixth predetermined time is set so that two consecutive second judgment sections in the time series have a portion in common. In this embodiment, the sixth predetermined time is set to 10 seconds.

Note that the sixth predetermined time in the present invention is not limited to 10 seconds. That is, for example, the sixth predetermined time in the present invention may be 5 seconds or 15 seconds.

When the later section in the time series continues for the second predetermined time or more ("Y" in S408), the transient fluctuation section identification unit 136 sets the point where the sixth predetermined time has elapsed from the starting point as a new starting point, and extracts the section from the new starting point to the fifth predetermined time as a new third judgment section (S409). In other words, the transient fluctuation section identification unit 136 updates the position of the starting point of the third judgment section to the position of the point six predetermined times after the same position. In other words, the transient fluctuation section identification unit 136 shifts the third judgment section back by the sixth predetermined time in the time series. Next, the transient fluctuation section identification process (S4) returns to process S403.

When the second non-constituent intervals occur consecutively a predetermined number of times ("Y" in S407), the transient fluctuation interval identifying unit 136 identifies an interval obtained by connecting second constituent intervals having a common interval among the multiple second constituent intervals as a transient fluctuation interval (S410). In other words, the transient fluctuation interval identifying unit 136 identifies an interval composed of one or more consecutive second constituent intervals as a transient fluctuation interval.

Next, the transient fluctuation section identification unit 136 extracts, from the transient fluctuation section, a section from the first point on the time series (the start point of the first second component section) until the seventh predetermined time has elapsed as a start section, and extracts, from the last point on the time series (the end point of the last second component section) until the seventh predetermined time has elapsed as an end section (S411).

The "seventh predetermined time" is a time shorter than the fifth predetermined time, and is set in order to extract a section (start section, end section) of the transient fluctuation section that may include a section in which no transient fluctuation occurs. The seventh predetermined time is set so that the transient fluctuation section does not include sections in which no transient fluctuation occurs at either end (so that such sections are minimized). In this embodiment, the seventh predetermined time is set to 10 seconds.

Note that the seventh predetermined time in the present invention is not limited to 10 seconds. That is, for example, the seventh predetermined time in the present invention may be 5 seconds or 15 seconds.

Next, the transient fluctuation section identification unit 136 calculates the variance value of the start section (hereinafter referred to as the "fourth variance value") and the variance value of the end section (hereinafter referred to as the "fifth variance value") (S412).

Next, the transient fluctuation section identification unit 136 compares the fourth variance value with a predetermined fourth threshold value and determines whether the fourth variance value is smaller than the fourth threshold value (S413).

The "fourth threshold" is a threshold that is set to determine whether or not a section in which no transient fluctuation occurs is included in the start section. In this embodiment, the fourth threshold is set to "3." The fourth threshold is, for example, set in advance by the user of the device 1 and stored in the memory unit 12.

When the fourth variance value is smaller than the fourth threshold value ("Y" in S413), the transient fluctuation section identification unit 136 excludes the start section from the transient fluctuation section (S414).

On the other hand, when the fourth variance value is equal to or greater than the fourth threshold value ("N" in S413), the transient fluctuation section identification unit 136 maintains the start section as part of the transient fluctuation section (S415).

Next, the transient fluctuation section identification unit 136 compares the fifth variance value with a predetermined fifth threshold value to determine whether the fifth variance value is smaller than the fifth threshold value (S416).

The "fifth threshold" is a threshold that is set to determine whether or not a section in which no transient fluctuation occurs is included in the end section. In this embodiment, the fifth threshold is set to "3." The fifth threshold is, for example, set in advance by a user of the device 1 and stored in the memory unit 12.

When the fifth variance value is smaller than the fifth threshold value ("Y" in S416), the transient fluctuation section identification unit 136 excludes the end section from the transient fluctuation section (S417).

On the other hand, when the fifth variance value is equal to or greater than the fifth threshold value ("N" in S416), the transient fluctuation section identification unit 136 maintains the end section as part of the transient fluctuation section (S418).

Next, the transient fluctuation interval identification unit 136 updates the transient fluctuation interval based on the results of processes S414, S415, S417, and S418 (S419). The updated transient fluctuation interval is stored in the storage unit 12 as auxiliary information. By identifying and updating the transient fluctuation interval in this way, the device 1 can identify, as the transient fluctuation interval, an interval that is close to the interval in which the doctor actually judges the transient fluctuation.

Next, the data acquisition unit 137 reads out the transient fluctuation section updated in process S419 (S420).

Next, the data acquisition unit 137 acquires, as the first learning data, the low-frequency component signal of the section from the last point on the time series of the transient fluctuation section (the end point of the transient fluctuation section) back to the eighth predetermined time (S421).

The "eighth predetermined time" is a time set in order to obtain as learning data a section in which a transient fluctuation that appears in a relatively short time occurs. Medically, a transient fluctuation is defined as ending in less than two minutes. However, in reality, a transient fluctuation may end in a relatively short time of about 90 seconds. The eighth predetermined time is a time set in order to deal with this transient fluctuation that appears in a short time, and is preferably between 80 and 100 seconds. In this embodiment, the eighth predetermined time is set to 90 seconds. The eighth predetermined time is, for example, set in advance by the user of the device 1 and stored in the memory unit 12.

Next, the data acquisition unit 137 acquires, as second learning data, the low-frequency component signal of the section from the last point on the time series of the transient fluctuation section back to the ninth predetermined time (S422).

The "ninth predetermined time" is a time set in order to obtain as learning data a section in which a transient fluctuation that appears over a relatively long period of time occurs among the transient fluctuations. Medically, a transient fluctuation is defined as ending in less than two minutes. However, in reality, a transient fluctuation may end in a relatively long period of time exceeding two minutes. The ninth predetermined time is a time set in order to deal with this long-term transient fluctuation, and is preferably between 150 and 200 seconds. In this embodiment, the ninth predetermined time is set to 180 seconds. The ninth predetermined time is, for example, set in advance by the user of the device 1 and stored in the memory unit 12.

Figure 17 is a schematic diagram explaining the method of acquiring the first learning data and the second learning data. This figure shows that for each transient fluctuation interval, the interval from the last point of the interval up to the 8th predetermined time back is acquired as the first learning data, and the interval from the last point of the interval up to the 9th predetermined time back is acquired as the second learning data.

Returning to Figures 13 to 16.
Next, the data acquiring unit 137 causes the display unit 15 to display the first learning data and the second learning data, and options for the labels corresponding to the types of transient fluctuations (S423).

The "label" is an identification mark that is set in advance for each type of transient fluctuation and stored in the memory unit 12. In this embodiment, the label of fluctuations other than transient fluctuations is set to "0", the label of transient tachycardia is set to "1", the label of early transient bradycardia and late transient bradycardia is set to "2", the label of variable transient bradycardia (mild) is set to "3", the label of variable transient bradycardia (severe) is set to "4", and the label of persistent transient bradycardia is set to "5".

Then, for example, the user of the device 1 selects labels corresponding to the first learning data and the second learning data, respectively, using, for example, the operation unit 14. At this time, the data acquisition unit 137 associates the first learning data with the label selected corresponding to the first learning data, associates the second learning data with the label selected corresponding to the second learning data, and stores them in the storage unit 12 as auxiliary information (auxiliary information that indirectly assists in identification) (S424). Next, the transient fluctuation learning process (S4) returns to process S408.

When the later section in the time series has not elapsed for the sixth predetermined time or more ("N" in S408), the transient fluctuation section identification unit 136 determines whether the value of the standby count described below is "0" or "1" (S425).

When the value of the standby count is "0", the transient fluctuation section identification unit 136 adds "1" to the standby count after waiting for the sixth predetermined time (S426) and executes process S408. On the other hand, when the value of the standby count is "1", the transient fluctuation section identification unit 136 determines whether or not there is a second constituent section that has not been identified as a transient fluctuation section in the sections following the section last identified as a transient fluctuation section (S427).

When there is a second constituent section that has not been identified as a transient fluctuation section ("Y" in S427), the transient fluctuation learning process (S4) returns to process S410. On the other hand, when there is no second constituent section ("N" in S427), the transient fluctuation learning process (S4) ends.

In the data acquisition process, the data acquisition unit of the present invention may only acquire and store the first learning data and the second learning data. In this case, the association with each label is performed, for example, by a separate process.

In addition, in the data acquisition process, the data acquisition unit of the present invention may continuously acquire the first learning data and the second learning data for multiple transient fluctuation intervals.

<Transient Fluctuation Identification Processing>
FIG. 18 is a flowchart of the transient fluctuation identification process (S5).
FIG. 19 is a schematic diagram for explaining each identification interval in the transient fluctuation identification process (S5).

The "transient fluctuation identification process (S5)" is a process for identifying transient fluctuations using a classifier generated based on the first learning data and the second learning data.

In the following description, it is assumed that the storage unit 12 stores a first classifier and a second classifier in advance. Here, the first classifier is a classifier generated in advance by the classifier generation unit 138 based on, for example, a plurality of first learning data associated with a label. The second classifier is a classifier generated in advance by the classifier generation unit 138 based on, for example, a plurality of second learning data associated with a label.

First, the classification unit 139 reads out the first classifier and the second classifier from the memory unit 12 (S501).

Next, the identification unit 139 reads out from the memory unit 12 a low-frequency component signal corresponding to the fetal heart rate waveform (waveform signal) for which transient fluctuations are to be identified (S502).

Next, the discrimination unit 139 sets the first point on the time series in the low-frequency component signal as the start point of the discrimination interval, extracts the interval from the start point until the elapse of an eighth predetermined time as the first discrimination interval, and extracts the interval from the start point until the elapse of a ninth predetermined time as the second discrimination interval (S503).

Next, the identification unit 139 identifies transient fluctuations from the low-frequency component signal in the first identification interval using the first identifier, and identifies transient fluctuations from the low-frequency component signal in the second identification interval using the second identifier (S504).

Next, the identification unit 139 determines the final identification results (final identification results) of the first and second identification intervals based on the identification result of the first classifier (first identification result) and the identification result of the second classifier (second identification result) (S505). At this time, for example, when the first and second identification results match, the identification unit 139 determines the first and second identification results as the final identification result of the first and second identification intervals. Also, for example, when the first and second identification results identify different transient fluctuations, the identification unit 139 determines both the first and second identification results as the final identification results of the first and second identification intervals. On the other hand, when only one of the first and second identification results identifies a transient fluctuation, the identification unit 139 determines the transient fluctuation as the final identification result of the first and second identification intervals. The final classification result is associated with the corresponding position (section) of the low-frequency component signal and stored as auxiliary information in the memory unit 12.

In addition, when the first and second classification results each identify a different transient fluctuation, the classification unit in the present invention may simply determine the classification result that there is a transient fluctuation as the final classification result for each of the first and second classification sections.

Next, the discrimination unit 139 determines whether a section in the low-frequency component signal that is later in the time series than the second discrimination section continues for a tenth predetermined time or longer (S506).

The "tenth predetermined time" is the time by which the first and second discrimination intervals are shifted back in the time series. The tenth predetermined time is set so that the transient fluctuation is judged without excess or deficiency. In this embodiment, the tenth predetermined time is set to three seconds.

When the later section in the time series continues for a tenth predetermined time or longer ("Y" in S506), the identification unit 139 executes process S503.

When the later section in the time series has not elapsed for the tenth predetermined time or more ("N" in S506), the identification unit 139 determines whether the value of the standby count described below is "0" or "1" (S507).

When the value of the standby count is "0", the identification unit 139 adds "1" to the standby count after waiting for the tenth predetermined time (S508) and executes process S506. On the other hand, when the value of the standby count is "1", the transient fluctuation identification process (S5) ends.

The identification unit in the present invention may execute the identification process only for the transient fluctuation section. In this case, the identification unit in the present invention may identify a transient fluctuation each time a transient fluctuation section is identified in the transient fluctuation section identification process.

<<Example>>
Next, an embodiment of the present invention will be described.

Figure 20 is a schematic diagram showing an example of the determination results when the baseline determination process (S2) and the fine variation determination process (S3) are performed on an actual CTG.

In the figure, the numbers in the upper row are underlined and indicate the baseline label ("0" is normal), and the numbers in the lower row indicate the fine fluctuation label ("0" is moderate, "2" is decreased). The figure shows that six baseline determination target sections, "a" through "f", have been identified. The figure also shows that the determination of fine fluctuations has changed within the baseline determination target sections, "b", "d", "e", and "f". Furthermore, the figure shows that the doctor has made corrections to the positions of the start and end points of the baseline determination target section, and the fine fluctuation determination results, for the baseline determination target section, "c" through "e".

Figure 21 is a schematic diagram showing the results of identifying the baseline determination target section when the baseline determination process (S2) is performed on an actual CTG.

"Total number of sections" is the number of sections that the doctor determined to be the baseline. "Number of unspecified" is the number of baseline determination target sections that the doctor determined to be the baseline but did not identify by the device 1. "Number of fine adjustments" is the number of baseline determination target sections that the doctor fine-tuned the positions of the start and/or end points of which were identified by the device 1. "Number of over-specified" is the number of baseline determination target sections that the doctor did not determine to be the baseline but identified by the device 1. "Interval accuracy rate" is the accuracy rate calculated by (total number of sections - number of over-specified) / (total number of sections + number of unspecified). As shown in Figure 21, the accuracy rate of identifying baseline determination target sections by the device 1 was approximately 96%.

Figure 22 is a schematic diagram showing the results of baseline determination when the baseline determination process (S2) is performed on an actual CTG.

The "Number of correct answers" and "Number of incorrect answers" indicate the number of correct answers and incorrect answers obtained by the device 1 in comparison with the doctor's judgment. As shown in FIG. 22, the accuracy rate of the baseline judgment by the device 1 was 100%. No bradycardia was observed in the fetal heart rate of the CTG used in this example.

Figure 23 is a schematic diagram showing the results of fine variation determination when the fine variation determination process (S3) is performed on an actual CTG.

"Number of correct answers" and "Number of incorrect answers" indicate the number of correct answers and incorrect answers given by the device 1 in comparison with the doctor's judgment. As shown in FIG. 20, the accuracy rate of the device 1 in judging fine fluctuations was approximately 99% for moderate, approximately 92% for decreased, and 100% for absent. There was no increase in fine fluctuations in the fetal heart rate of the CTG used in this example.

When the classifier generated by this device was used to classify transient fluctuations, the classification accuracy was approximately 79%. At this time, the number of training data for label "0" was 308, the number of training data for label "1" was 161, the number of training data for label "2" was 76, the number of training data for label "3" was 321, the number of training data for label "4" was 83, and the number of training data for label "5" was 33.

In this way, the present invention achieves an extremely high accuracy rate of 100% in determining the baseline and 90% to 100% in determining fine fluctuations. The present invention also achieves a high accuracy rate of approximately 79% in identifying transient fluctuations. In other words, the present invention enables uniform (automatic) and highly accurate interpretation of CTG. As a result, the present invention can assist doctors in interpreting CTG by providing them with this information as auxiliary information.

<Summary>
According to the embodiment described above, the present device 1 includes a low-frequency extraction unit 1314 that extracts a low-frequency component signal from a waveform signal, and a baseline section identification unit 132 that identifies a baseline determination target section based on the low-frequency component signal. The baseline section identification unit 132 extracts a first determination section in the low-frequency component signal, and identifies the baseline determination target section as auxiliary information based on the low-frequency component signal in the first determination section. According to this configuration, the present device 1 extracts a low-frequency component signal in which baseline characteristics are likely to appear from the CTG, and identifies the baseline determination target section as auxiliary information based on the low-frequency component signal. That is, the present device 1 can uniformly (automatically) identify auxiliary information (baseline determination target section) from the CTG. As a result, the present device 1 can provide auxiliary information used to uniformly assist in the interpretation of the CTG, regardless of the individual differences of the person who interprets the CTG.

Furthermore, according to the embodiment described above, the baseline section identification unit 132 identifies the first judgment section as the first constituent section constituting the baseline judgment target section when the variance value (first variance value) of the low-frequency component signal in the first judgment section is smaller than the first threshold value. With this configuration, the present device 1 identifies the first constituent section constituting the baseline judgment target section based on an index (variance value of the low-frequency component signal) that is different from the memory of the CTG that is visually confirmed when a doctor interprets the CTG. As a result, the present device 1 can provide auxiliary information that is used to assist in uniform interpretation of the CTG.

Furthermore, according to the embodiment described above, the baseline section identification unit 132 sets a point at which a second predetermined time shorter than the first predetermined time has elapsed from the first starting point as a new first starting point, and extracts the section from the new first starting point until the first predetermined time has elapsed as the new first judgment section. That is, the baseline section identification unit 132 shifts the first judgment section back in time by the second predetermined time. Then, when a plurality of first constituent sections are identified, the baseline section identification unit 132 identifies a section obtained by connecting first constituent sections having a common section among the plurality of first constituent sections as a baseline judgment target section. According to this configuration, the present device 1 can identify the first constituent section while shifting the first judgment section back little by little in the time series. As a result, the present device 1 can accurately identify the start point and end point of the baseline judgment target section, and identify a section that is close to the section in which the doctor actually judges the baseline as the baseline judgment target section.

Furthermore, according to the embodiment described above, the baseline determination unit 133 identifies the average heart rate for each first constituent section that constitutes the baseline determination target section in the low-frequency component signal as the baseline fetal heart rate, and determines the baseline fetal heart rate based on the average heart rate. In other words, the present device 1 always determines the baseline fetal heart rate for each first constituent section, regardless of the length of the baseline determination target section. With this configuration, the present device 1 can always capture the fluctuations in the low-frequency component signal in a constant section (first determination section), regardless of the length of the baseline determination target section. Therefore, the present device 1 can stably capture the fluctuations in the low-frequency component signal in the baseline determination target section. As a result, the present device 1 can achieve a stable and uniform determination of the baseline.

Furthermore, according to the embodiment described above, the device 1 includes a high-frequency extraction unit 1315 that extracts a high-frequency component signal from a waveform signal, and a fine fluctuation section identification unit 134 that identifies a fine fluctuation determination target section as auxiliary information based on the high-frequency component signal. The fine fluctuation section identification unit 134 identifies a section in the high-frequency component signal that corresponds to the baseline determination target section as the fine fluctuation determination target section. According to this configuration, the device 1 extracts a high-frequency component signal in which fine fluctuation characteristics are likely to appear from the CTG, and identifies the fine fluctuation determination target section as auxiliary information based on the high-frequency component signal. In other words, the device 1 can uniformly (automatically) identify auxiliary information (fine fluctuation determination target section) from the CTG. As a result, the device 1 can provide auxiliary information used to assist in uniform CTG interpretation.

Furthermore, according to the embodiment described above, the device 1 has a fine fluctuation determination unit 135 that determines fine fluctuations. The fine fluctuation determination unit 135 extracts a second determination section from the fine fluctuation determination section, and determines fine fluctuations based on the high frequency component signal in the second determination section. With this configuration, the device 1 can always capture fluctuations in the high frequency component signal in a constant section (the second determination section) regardless of the length of the fine fluctuation determination section, and can therefore stably capture fluctuations in the high frequency component signal in the fine fluctuation determination section. As a result, the device 1 can achieve stable fine fluctuation determination.

Furthermore, according to the embodiment described above, the fine fluctuation determination unit 135 sets the point at which the fourth predetermined time has elapsed from the second starting point as a new second starting point, extracts the section from the new second starting point to the section at which the third predetermined time has elapsed as a new second determination section, and determines fine fluctuations for each second determination section. That is, the fine fluctuation determination unit 135 determines fine fluctuations for each second determination section while shifting the second determination section little by little in the time series. According to this configuration, the present device 1 can always capture the fluctuation of the high-frequency component signal in a constant section (second determination section) regardless of the length of the fine fluctuation determination section, and can stably capture the fluctuation of the high-frequency component signal in the fine fluctuation determination section. In addition, by determining fine fluctuations while shifting the second determination section by the fourth predetermined time, the present device 1 can determine fine fluctuations in a section of the baseline that is close to the section that a doctor actually visually observes to determine fine fluctuations from the baseline. As a result, the present device 1 can achieve stable fine fluctuation determination.

Furthermore, according to the embodiment described above, the fine fluctuation determination unit 135 determines fine fluctuations as either "moderate," "increased," "decreased," or "disappeared" based on the variance value of the high-frequency component signal in the second determination section and a second threshold value composed of multiple threshold values. With this configuration, the device 1 determines fine fluctuations based on an index (the variance value of the high-frequency component signal) that is different from the CTG memory that is visually confirmed when a doctor interprets a CTG. As a result, the device 1 can provide auxiliary information (fine fluctuation determination results) that can be used to assist in uniform CTG interpretation.

Furthermore, according to the embodiment described above, the high frequency component signal includes a higher frequency band than the low frequency component signal. On the other hand, the low frequency component signal includes a lower frequency band than the high frequency component signal. Also, part of the frequency band of the high frequency component signal is common to part of the frequency band of the low frequency component signal. Normally, when determining the baseline or fine fluctuations, doctors do not uniformly distinguish between high and low frequency components of a waveform signal. Therefore, according to this configuration, the device 1 can obtain a determination result that is closer to a doctor's visual determination, compared to a case in which low and high frequency component signals are separated by a single threshold value.

Furthermore, according to the embodiment described above, the device 1 includes a transient fluctuation section identification unit 136 that identifies one or more transient fluctuation sections in which a transient fluctuation occurs as auxiliary information based on a low-frequency component signal. According to this configuration, the device 1 extracts a low-frequency component signal in which the characteristics of a transient fluctuation are likely to appear from the CTG, and identifies the transient fluctuation section as auxiliary information based on the low-frequency component signal. In other words, the device 1 can uniformly (automatically) identify auxiliary information (transient fluctuation section) from the CTG. As a result, the device 1 can provide auxiliary information that is used to uniformly assist in interpreting the CTG, regardless of the individual differences of the person interpreting the CTG.

Furthermore, according to the embodiment described above, when the variance value (third variance value) of the low-frequency component signal in the third judgment section is greater than the third threshold, the transient fluctuation section identification unit 136 identifies the third judgment section as the second constituent section constituting the transient fluctuation section. According to this configuration, the present device 1 identifies the second constituent section constituting the transient fluctuation section based on an index (variance value of the low-frequency component signal) different from the CTG memory visually confirmed when a doctor interprets the CTG. As a result, the present device 1 can provide auxiliary information that is used to assist in uniform CTG interpretation, regardless of the individual differences of the person interpreting the CTG.

Furthermore, according to the embodiment described above, the transient fluctuation section identification unit 136 sets the point at which a sixth predetermined time shorter than the fifth predetermined time has elapsed from the third starting point as a new third starting point, and extracts the section from the new third starting point to the fifth predetermined time as a new third judgment section. That is, the transient fluctuation section identification unit 136 shifts the third judgment section back by the sixth predetermined time in the time series. Then, when a plurality of second constituent sections are identified, the transient fluctuation section identification unit 136 identifies as a transient fluctuation section a section obtained by connecting second constituent sections having a common section among the plurality of second constituent sections. According to this configuration, the present device 1 can identify the second constituent section while shifting the third judgment section back little by little in the time series. By identifying and updating the transient fluctuation section in this way, the present device 1 can identify as a transient fluctuation section a section that is close to the section in which a doctor actually judges a transient fluctuation. As a result, the device 1 can accurately identify the start and end points of the transient fluctuation section.

Furthermore, according to the embodiment described above, the transient fluctuation section identification unit 136 extracts, from the transient fluctuation section, a section from the first point on the time series until the seventh predetermined time has elapsed as a start section, and extracts, from the last point on the time series until the seventh predetermined time has elapsed, as an end section. Then, when the fourth variance value of the start section is smaller than the fourth threshold, the transient fluctuation section identification unit 136 excludes the start section from the transient fluctuation section, and when the fifth variance value of the end section is smaller than the fourth threshold, excludes the end section from the transient fluctuation section, and updates the transient fluctuation section. According to this configuration, the present device 1 performs re-determination of sections in the transient fluctuation section that may include sections in which transient fluctuations do not occur, and can more accurately identify the start and end points of the transient fluctuation section.

Furthermore, according to the embodiment described above, the device 1 includes a data acquisition unit 137 that acquires learning data for a classifier that identifies a transient fluctuation based on the transient fluctuation section. The data acquisition unit 137 extracts, as learning data (first learning data), a low-frequency component signal from the section of the transient fluctuation section from the last point on the time series to the eighth predetermined time back, as learning data. Medically, a transient fluctuation is defined as ending in less than two minutes. Therefore, when identifying a transient fluctuation, learning data acquired from the end point of the transient fluctuation section is more likely to include the characteristics of a transient fluctuation than learning data acquired from the start point of the transient fluctuation section. Therefore, according to this configuration, the device 1 can obtain learning data including the characteristics of a transient fluctuation as auxiliary information that assists in identifying a transient fluctuation.

Furthermore, according to the embodiment described above, the present device 1 has a discrimination unit 139 that discriminates transient variations using a discriminator generated based on learning data. The discrimination unit 139 automatically discriminates transient variations based on low-frequency component signals. According to this configuration, the present device 1 acquires not only the baseline determination target section but also the discrimination result of the transient variation as auxiliary information. As a result, the present device 1 can provide auxiliary information that is used to assist in uniform CTG interpretation, regardless of the individual differences of the person interpreting the CTG.

Furthermore, according to the embodiment described above, the device 1 includes a noise removal unit 1313 that removes noise from the waveform signal. After removing the main sudden noise, the noise removal unit 1313 applies a low-pass filter and calculates the signal difference twice to clarify and remove the remaining noise. With this configuration, the device 1 can remove most of the noise that is often contained in the waveform signal of the fetal heart rate of the CTG. As a result, the device 1 can accurately identify the baseline determination target section and the transient fluctuation section, and can accurately determine the baseline and fine fluctuations.

Note that the configuration of the present device is not limited to this embodiment as long as it can identify at least the baseline determination target section as auxiliary information. That is, for example, the present device does not need to include a classifier generation unit and a classification unit, and further does not need to include a transient fluctuation section identification unit and a data acquisition unit, or further does not need to include a fine fluctuation section identification unit and a fine fluctuation determination unit.

Furthermore, the noise removal unit in the present invention only needs to be able to remove sudden noise, and the configuration of the noise removal unit in the present invention is not limited to this embodiment. That is, for example, the noise removal unit in the present invention does not need to include a secondary low-pass filter and a secondary calculation unit.

Furthermore, the control unit in the present invention may not include a noise removal unit. In this case, the present device may include, for example, a dedicated circuit that functions as a noise removal unit. Also, for example, if the waveform signal has little noise, it may not be necessary to remove noise from the waveform signal.

Furthermore, each threshold value in this method may be determined based on the accuracy rate of each judgment result and classification result, and is not limited to this embodiment.

Furthermore, the method executed by the device is not limited to this embodiment. That is, for example, the device may not execute the transient fluctuation learning process and/or the transient fluctuation identification process, and may not execute the fine fluctuation determination process.

Furthermore, in the transient fluctuation learning process, the data acquisition unit of the present invention only needs to acquire the first learning data and the second learning data, and does not need to associate the first learning data and the second learning data with the corresponding labels (types of transient fluctuations). In this case, for example, labeling of the first learning data and the second learning data may be performed separately.

Furthermore, the present device may be incorporated into an external device, or an external device may function as the present device by executing the present program.

Furthermore, in the above-described embodiment, the device 1 is configured by one computer. Alternatively, the device may be configured by multiple computers. That is, for example, the device may be configured by a group of multiple computers that function as the device. Specifically, for example, the device (group of computers) may be configured by a computer having a storage unit and a computer having a control unit that executes the method. Also, for example, the multiple computers may have the functions of a preprocessing unit, a baseline section identification unit, a baseline determination unit, a fine fluctuation determination section identification unit, a fine fluctuation determination unit, a transient fluctuation section identification unit, a data acquisition unit, a classifier generation unit, and a classification unit in a distributed manner. In this case, the multiple computers that make up the group of computers may send and receive information through a communication line, or may transfer information using a portable storage medium.

1 Information processing device 12 Storage unit 1311 Acquisition unit 1312 Waveform signal extraction unit 1314 Low frequency extraction unit 1315 High frequency extraction unit 132 Baseline section identification unit 133 Baseline determination unit 134 Fine fluctuation section identification unit 135 Fine fluctuation determination unit 136 Transient fluctuation section identification unit 137 Data acquisition unit 139 Identification unit

Claims (13)

1. An information processing device for generating auxiliary information used to assist in interpretation of a cardiotocography,
An acquisition unit for acquiring the cardiotocography;
a waveform signal extraction unit that extracts a waveform signal of a fetal heart rate waveform included in the acquired fetal cardiotocography;
a low frequency extraction unit that extracts a predetermined low frequency component signal from the waveform signal;
a baseline interval specifying unit that specifies, as the auxiliary information, a baseline determination target interval for determining a fetal heart rate baseline based on the extracted low frequency component signal;
and
The baseline section identification unit is
extracting, from the low-frequency component signal, a section from a first predetermined start point to a first predetermined time required for determining the fetal heart rate baseline as a first determination section;
when a variance value of the low frequency component signal in the first judgment section is smaller than a first threshold, the first judgment section is identified as a first constituent section constituting the baseline judgment target section;
23. An information processing apparatus comprising: The baseline section identification unit is
a point at which a second predetermined time shorter than the first predetermined time has elapsed from the first starting point is set as a new first starting point, and a section from the new first starting point to the elapse of the first predetermined time is extracted as a new first determination section;
When a plurality of the first component sections are identified, a section obtained by connecting the first component sections having a common section among the plurality of the first component sections is identified as the baseline determination target section.
2. The information processing device according to claim 1. a baseline determination unit that determines the fetal heart rate baseline in the baseline determination target section based on the identified baseline determination target section;
and
the baseline determination unit specifies, in the low frequency component signal, an average heart rate for each of the first constituent sections constituting the baseline determination target section as a fetal heart rate of the fetal heart rate baseline, and determines the fetal heart rate baseline based on the average heart rate.
3. The information processing device according to claim 1 or 2. a high frequency extraction unit that extracts a high frequency component signal within a predetermined range from the waveform signal;
a fine fluctuation section specifying unit for specifying a fine fluctuation determination target section for determining a baseline fine fluctuation of a fetal heart rate based on the extracted high frequency component signal as the auxiliary information;
and
The fine fluctuation section identification unit is
identifying a section corresponding to the baseline determination section as the fine fluctuation determination section in the extracted high frequency component signal;
2. The information processing device according to claim 1. a fine fluctuation determination unit that determines the fetal heart rate baseline fine fluctuation in the fine fluctuation determination target section based on the specified fine fluctuation determination target section;
and
The fine variation determination unit
A start point of the minute fluctuation determination target section is set as a second start point, and a section from the second start point until a third predetermined time has elapsed is extracted as a second determination section;
determining the fetal heart rate baseline fluctuation as being either moderate, increased, decreased, or absent based on a variance value of the high frequency component signal in the second judgment section and a second threshold value consisting of a plurality of threshold values;
5. The information processing device according to claim 4. The fine variation determination unit
setting a point at which a fourth predetermined time has elapsed since the second start point as a new second start point, and extracting a section from the new second start point to the time at which the third predetermined time has elapsed as a new second determination section;
determining the fetal heart rate baseline variation for each of the second determination intervals;
6. The information processing device according to claim 5. the high frequency component signal includes a higher frequency band than the low frequency component signal;
the low frequency component signal includes a frequency band lower than the high frequency component signal;
A part of a frequency band of the high frequency component signal is common to a part of a frequency band of the low frequency component signal.
5. The information processing device according to claim 4. a transient fluctuation interval identifying unit that identifies, as the auxiliary information, one or more transient fluctuation intervals in which a transient fluctuation occurs, based on the extracted low frequency component signal;
and
The transient fluctuation section identification unit is
setting the first starting point as a third starting point in the low frequency component signal, and extracting a section from the third starting point until a fifth predetermined time has elapsed as a third determination section;
when a variance value of the low frequency component signal in the third judgment section is greater than a third threshold, the third judgment section is identified as a second constituent section constituting the transient fluctuation section.
2. The information processing device according to claim 1. The transient fluctuation section identification unit is
setting a point at which a sixth predetermined time has elapsed since the third start point as a new third start point, and extracting a section from the new third start point to the point at which the fifth predetermined time has elapsed as a new third determination section;
Identifying a section constituted by one or more consecutive second constituent sections as the transient fluctuation section;
9. The information processing device according to claim 8. The transient fluctuation section identification unit is
Among the transient fluctuation sections,
A section from the first point on the time series until a seventh predetermined time has elapsed is extracted as a start section;
extracting a section from the last point on the time series back to the seventh predetermined time as an end section;
when the fourth variance value of the start section is smaller than a fourth threshold, the start section is excluded from the transient fluctuation section, and when the fifth variance value of the end section is smaller than the fourth threshold, the end section is excluded from the transient fluctuation section, and the transient fluctuation section is updated.
9. The information processing device according to claim 8. a data acquisition unit that acquires learning data for a classifier that identifies the transient fluctuation based on the transient fluctuation section;
and
The data acquisition unit extracts, as the learning data, the low-frequency component signal in a section from a last point on the time series to an eighth predetermined time back from the transient fluctuation section.
9. The information processing device according to claim 8. A computer is caused to function as the information processing device according to claim 1.
2. An information processing program comprising: 1. An information processing method executed by an information processing device for generating auxiliary information used to assist in interpretation of a cardiotocography, comprising:
The information processing device,
acquiring said cardiotocography;
A waveform extraction step of extracting a waveform signal of a fetal heart rate waveform included in the acquired fetal cardiotocography;
a low frequency extraction step of extracting a predetermined low frequency component signal from the waveform signal;
a baseline determination target section specifying step of specifying, as the auxiliary information, a baseline determination target section for determining a fetal heart rate baseline based on the extracted low frequency component signal;
Including,
The baseline determination target section specifying step includes:
extracting, from the low frequency component signal, a section from a first predetermined start point to a first predetermined time required for determining the fetal heart rate baseline as a first determination section;
identifying the baseline determination section based on the low frequency component signal in the first determination section;
Including,
23. An information processing method comprising:

Images