Apparatus for estimating physiological phenomena and method for estimating physiological phenomena

Abstract

To alleviate spatial and physical constraints when understanding physiological phenomena such as menstrual cycles and pregnancy. [Solution] The physiological phenomenon estimation device comprises: an information acquisition unit that acquires detection data obtained by detecting over time signals that reflect changes in blood flow caused in accordance with the menstrual cycle or pregnancy state using an optical sensor or strain sensor placed on the body surface; an inference unit that takes input data including detection data or secondary data based on detection data as input and outputs information about the menstrual cycle or pregnancy state as output data obtained by the information acquisition unit, and generates inference information about the menstrual cycle or pregnancy state; and an output unit that outputs the inference information generated by the inference unit.

Description

Technical Field

[0001] The present invention relates to an apparatus for estimating physiological phenomena and a method for estimating physiological phenomena.

Background Art

[0002] Anatomical, physiological, and biochemical changes occur in the human body due to physiological phenomena such as the menstrual cycle and pregnancy. Techniques for grasping physiological phenomena from the changes occurring in the human body are known. For example, Patent Document 1 discloses an apparatus for monitoring uterine contractions of a pregnant mother. The apparatus of Patent Document 1 is a sensor unit attached to the abdomen of the mother by a belt. The sensor unit is a sensor device called a tocodynamometer, and detects a change in the tension of uterine muscles caused by uterine contractions as a pressure change acting on the sensor unit. The sensor unit needs to be fixed with an appropriate belt tension while being appropriately in contact with the abdominal wall of the mother. The sensor unit detects the contact between the abdomen of the mother and the sensor unit by a PPG sensor.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In addition to the apparatus disclosed in Patent Document 1 above, there are various apparatuses capable of grasping physiological phenomena such as the menstrual cycle and pregnancy, such as a delivery monitoring apparatus, an ultrasonic diagnostic apparatus, and a blood test apparatus. However, each apparatus has spatial constraints such as being large and difficult to carry around, or significant physical constraints on the subject, such as fixing the sensor unit with a belt.

[0005] During pregnancy, it is necessary to pay close attention to changes in the mother's and fetus's condition due to the risks of premature birth and intrauterine fetal death. Furthermore, understanding the menstrual cycle before pregnancy is expected to have applications in contraception, pregnancy, and infertility treatment. Therefore, it is desirable to alleviate the spatial and physical constraints when understanding physiological phenomena such as the menstrual cycle and pregnancy.

[0006] This disclosure is made in view of the above, and aims to provide a device and method for estimating physiological phenomena that can alleviate spatial and physical constraints when understanding physiological phenomena such as menstrual cycles and pregnancy. [Means for solving the problem]

[0007] The physiological phenomenon estimation device according to the present invention comprises: an information acquisition unit that acquires detection data obtained by detecting over time a signal reflecting changes in blood flow caused in accordance with the menstrual cycle or pregnancy state using an optical sensor or strain sensor placed on the body surface; an inference unit that takes input data including the detection data or secondary data based on the detection data as input and outputs information related to the menstrual cycle or pregnancy state as input to a trained model that produces inference information related to the menstrual cycle or pregnancy state by inputting the input data obtained by the information acquisition unit using the detection data; and an output unit that outputs the inference information generated by the inference unit.

[0008] The method for estimating physiological phenomena according to the present invention comprises the steps of: acquiring detection data obtained by detecting over time a signal that reflects changes in blood flow caused in accordance with the menstrual cycle or pregnancy state using an optical sensor or strain sensor placed on the body surface; inputting the input data obtained from the detection data into a trained model that takes input data including the detection data or secondary data based on the detection data as input and outputs information about the menstrual cycle or pregnancy state, to generate inference information about the menstrual cycle or pregnancy state; and outputting the generated inference information. [Effects of the Invention]

[0009] According to this disclosure, spatial and physical constraints can be alleviated when understanding physiological phenomena such as the menstrual cycle and pregnancy. [Brief explanation of the drawing]

[0010] [Figure 1] Figure 1 is a schematic block diagram showing a physiological phenomenon estimation device according to an embodiment. [Figure 2] Figure 2 is an explanatory diagram showing the first trained model. [Figure 3] Figure 3 is an explanatory diagram showing the second trained model. [Figure 4] Figure 4 shows an example of an inference unit that performs inference based on input data including supplementary information. [Figure 5] Figure 5 is an explanatory diagram illustrating the output of inference information. [Figure 6] Figure 6 is an explanatory diagram illustrating the generation of the first trained model (machine learning). [Figure 7] Figure 7 is an explanatory diagram illustrating the generation of a second pre-trained model (machine learning). [Figure 8] Figure 8 is a flowchart illustrating the method for estimating physiological phenomena according to this embodiment. [Figure 9] Figure 9 is an explanatory diagram showing a first configuration example of a physiological phenomenon estimation device according to the embodiment. [Figure 10] Figure 10 is an explanatory diagram showing a second configuration example of the physiological phenomenon estimation device according to the embodiment. [Figure 11] Figure 11 is an explanatory diagram showing a third configuration example of the physiological phenomenon estimation device according to the embodiment. [Figure 12] Figure 12 is a conceptual diagram showing variations in the generation of evaluation information by the evaluation unit. [Figure 13] Figure 13 is a schematic diagram illustrating the generation of individually trained models through fine-tuning. [Figure 14] Figure 14 is a schematic diagram illustrating the inference process using individually trained models. [Modes for carrying out the invention]

[0011] The following describes embodiments of the physiological phenomenon estimation apparatus and physiological phenomenon estimation method according to the present invention with reference to the drawings. However, the present invention is not limited by these embodiments. Furthermore, the components in the following embodiments include those that are easily substituted or substantially identical to those that are easily substituted by those skilled in the art. The present invention is not limited by these embodiments, and if there are multiple embodiments, they may also be constructed by combining each embodiment.

[0012] (A device for estimating physiological phenomena) Figure 1 is a schematic block diagram showing the physiological phenomenon estimation device 1 according to this embodiment. As shown in Figure 1, the estimation device 1 is a device that infers information about physiological phenomena such as the user's menstrual cycle or pregnancy status based on detection data 2 (or secondary data generated from detection data 2) detected over time by sensors 30 placed on the user's body surface, and outputs the obtained inference information 3. The estimation device 1 is a device that can be used outside of medical facilities such as hospitals, for example, at the user's home or while out and about. The user is a user who receives information about physiological phenomena such as the menstrual cycle or pregnancy status. The user is, for example, a woman in the age range in which menstruation occurs periodically.

[0013] Inference Information 3 is information regarding the menstrual cycle or pregnancy status. The menstrual cycle refers to the period from the start of menstruation to the day before the start of the next menstruation. Pregnancy status refers to the state of the mother and / or the fetus FE in the mother during pregnancy. Inference Information 3 regarding the menstrual cycle enables the provision of useful information for user US regarding pregnancy (or infertility treatment) or contraception. Inference Information 3 regarding pregnancy status may include information related to signs of perinatal complications such as premature birth, intrauterine fetal death, and placental abruption. Inference Information 3 regarding pregnancy status can be expected to enable early detection of premature birth, stillbirth, and perinatal complications, as well as medical intervention, prevention, and treatment. Furthermore, for pregnant user US, it enables a simple assessment of pregnancy status between regular checkups at medical institutions.

[0014] The detection data 2 is data obtained by detecting, over time, a signal that reflects a change in blood flow volume caused according to the menstrual cycle or pregnancy state. The menstrual cycle and pregnancy state each cause anatomical, physiological, and biochemical changes in the human body. One of the changes caused according to the menstrual cycle and pregnancy state is a change in blood flow volume. The blood flow volume is specifically the systemic blood flow volume. The detection data 2 is data that directly or indirectly indicates a change in blood flow volume. That is, the detection data 2 may be data that indicates a change in blood flow volume, or the change in blood flow volume may be grasped by secondary data obtained by analyzing the detection data 2 or the like. The detection data 2 includes time-series data in an electrical output form output by a sensor 30 (optical sensor or strain sensor), or image data indicating a time-series change in the electrical output form. The electrical output form is an electrical signal (that is, the output signal of the sensor) obtained by a detection circuit extracting a change (such as a voltage change, current change, resistance value change, etc.) occurring in the sensor 30. The detection data 2 in the electrical output form includes pair data of a measurement time and a signal intensity, and is composed of an array of pair data as time elapses. And the detection data 2 may be image data expressing such an array over time of pair data as a graph (waveform diagram).

[0015] The detection data 2 is detected by a sensor 30 disposed on the body surface of a user US who is the person to be measured. The body surface of the user US includes the surfaces of the limbs, head, and trunk of the human body. The detection data 2 may be data obtained by transcutaneously detecting a phenomenon inside the body (such as blood flow) from the body surface, or may be data obtained by detecting a signal occurring on the body surface (for example, a minute displacement of the body surface). The detection data 2 detected by the sensor 30 disposed on the body surface contains various anatomical, physiological, and biochemical information. That is, by appropriately analyzing the detection data 2 that reflects a change in blood flow volume, it is possible to grasp various information described later.

[0016] In the present embodiment, the sensor 30 is constituted by an optical sensor or a strain sensor.

[0017] An optical sensor optically detects a signal that reflects changes in blood flow. An optical sensor comprises, for example, a light emitter and a light receiver. An optical sensor is, for example, a photoelectric plethysmometer. The optical sensor emits an optical signal from the light emitter onto the body surface, and the light signal transmitted through the tissue, including blood vessels, near the body surface is detected by the light receiver. The light emitter is equipped with an LED (light-emitting diode) or the like as a light source and generates an optical signal consisting of green light, red light, or infrared light. As the light receiver, a photodiode, compound semiconductor sensor, or phototransistor can be used. A photodiode can detect an optical signal by generating a current in response to light irradiation due to the photovoltaic effect. Examples of photodiodes include CCD sensors and CMOS sensors. The photodiode may be PN type, PIN type, or other types. A compound semiconductor sensor can detect an optical signal by changing the resistance value in response to light irradiation due to the photoconductive effect. An example of a compound semiconductor sensor is a CdS (cadmium sulfide) sensor. A phototransistor is a light sensor that amplifies the output current of a photodiode using a transistor.

[0018] Strain sensors output signals corresponding to strain (minute deformation) on the body surface. When blood vessels near the body surface pulsate in response to changes in blood flow, strain occurs in surface tissues such as skin and nails. Therefore, the detection data 2 detected by the strain sensor includes a signal component that reflects tissue blood flow, which is an example of blood flow. Examples of strain sensors include resistors such as strain gauges, piezoelectric elements, MEMS (Micro Electro Mechanical Systems) sensors, and capacitance sensors. Resistors can detect strain signals due to changes in resistance caused by strain. Piezoelectric elements can detect strain signals due to changes in voltage caused by strain. MEMS sensors can utilize the above-mentioned resistor or piezoelectric MEMS pressure sensors, as well as acceleration sensors and gyroscopes to detect strain signals as displacements associated with minute deformations of surface tissues. Capacitance sensors can detect strain signals due to changes in capacitance caused by strain.

[0019] The sensor 30, which consists of an optical sensor or a strain sensor, is small and lightweight and can be worn on the user US's body as a wearable device, for example. The user US does not need to secure space in a room or elsewhere for the installation of the measuring device including the sensor 30, and can operate while wearing the sensor 30 without being limited to a specific wearing posture. Therefore, the sensor 30 configured as a wearable device can alleviate spatial and physical constraints for generating detection data 2. The location on which the sensor 30 is attached is not particularly limited. Examples of sensors 30 attached to the limbs include wearable devices such as ring-type, wristwatch (wristband)-type, and artificial nail-type sensors. Examples of sensors 30 attached to the head include wearable devices such as ear clip-type, headphone or earphone-type, and glasses-type sensors. Examples of sensors 30 attached to the torso include wearable devices such as belt-type and abdominal band (corset)-type sensors. Other forms include, for example, sensors built into smartphones. Smartphones can be used as wearable devices by fixing them to a part of the body.

[0020] Detection data 2 may include other data in addition to data reflecting changes in blood flow. For example, detection data 2 may further include information on the elasticity of the blood vessel wall, blood pressure, pulse wave, pulse rate, blood flow velocity, oxygen saturation, blood components, and body surface temperature. Information on blood components includes, for example, red blood cell count (RBC), white blood cell count (WBC), hemoglobin (Hb), hematocrit (Hct), platelet count (Plt), total protein (TP), total bilirubin (T-Bil), alkaline phosphatase (ALP), glutamate oxaloacetate transaminase (AST), glutamate pyruvate transaminase (ALT), lactate dehydrogenase (LDH), creatinine phosphokinase (CPK), leucine aminopeptidase (LAP), gamma-glutamyl transpeptidase (γ-GTP), cholinesterase (Ch-E), creatinine (Cre), uric acid (UA), blood urea nitrogen (BUN), amylase (AMY), and total uric acid. This product contains at least one of the following amounts or indicator values: sterol (T-CHO), triglycerides (TG), LDL-CHO, HDL-CHO, serum iron (Fe), unsaturated iron-binding capacity (TIBC), sodium (Na), potassium (K), chloride (Cl), calcium (Ca), C-reactive protein (CRP), brain natriuretic peptide (BNP), HbA1c, blood glucose (BS), albumin (Alb), hematocrit (Hct), partial pressure of carbon dioxide (pCO2), alcohol, GnRH (gonadotropin-releasing hormone), LH (luteinizing hormone), FSH (follicle-stimulating hormone), hCG (human chorionic gonadotropin), inhibin, prolactin (PRL), thyroid hormones (TSH, FT4, FT3), estrogen, and progesterone. The sensor 30 may detect this information using an optical sensor or a strain sensor, or it may have a sensor for detecting this information in addition to the optical sensor or strain sensor.

[0021] The physiological phenomenon estimation device 1 is composed of a computer including a processor such as a CPU (Central Processing Unit) and memory such as ROM (Read Only Memory) and RAM (Random Access Memory). The physiological phenomenon estimation device 1 includes an information acquisition unit 10, an inference unit 11, an evaluation unit 12, an output unit 13, and a storage unit 14. The information acquisition unit 10, the inference unit 11, the evaluation unit 12, and the output unit 13 are configured as functional blocks realized on the program 20, which is stored in the storage unit 14, by the processor executing the program 20. The program 20 is a program that causes the computer to function as the information acquisition unit 10, the inference unit 11, the evaluation unit 12, and the output unit 13. The information acquisition unit 10, the inference unit 11, the evaluation unit 12, and the output unit 13 may be configured by dedicated hardware.

[0022] The physiological phenomenon estimation device 1 has an interface for communicating with the sensor 30 by wire or wireless means, and receives measurement data from the sensor 30. The physiological phenomenon estimation device 1 may also be able to read data from an electromagnetic recording medium that records the detection data 2 obtained from the sensor 30.

[0023] The information acquisition unit 10 acquires detection data 2. As described above, detection data 2 is data detected over time by a sensor 30 (optical sensor or strain sensor) placed on the body surface, which reflects changes in blood flow caused by the menstrual cycle or pregnancy status. In one example, the information acquisition unit 10 acquires detection data 2 from the sensor 30 via communication. The information acquisition unit 10 acquires detection data 2 from the sensor 30 at a predetermined sampling period and stores it in the storage unit 14. The information acquisition unit 10 may also read the detection data 2 that has been detected by the sensor 30 and stored in an electromagnetic recording medium (e.g., flash memory, external server, etc.).

[0024] The inference unit 11 generates inference information 3 regarding the menstrual cycle or pregnancy status based on the detection data 2 acquired by the information acquisition unit 10. Specifically, the inference unit 11 inputs the input data 5 (see Figures 2 and 3) obtained from the detection data 2 acquired by the information acquisition unit 10 into the trained model 21 and performs an inference process to generate inference information 3 regarding the menstrual cycle or pregnancy status. The input data 5 includes the detection data 2 or secondary data based on the detection data 2. The trained model 21 outputs information regarding the menstrual cycle or pregnancy status according to the input data 5. The trained model 21 is generated by machine learning using training data that includes input data 5, which includes the temporal changes in signals reflecting changes in blood flow caused by the menstrual cycle or pregnancy status, and information regarding the menstrual cycle or pregnancy status. The trained model 21 is stored in the storage unit 14. The output inference information 3 is, for example, numerical information regarding the menstrual cycle or pregnancy status. The inference information 3 may also be, for example, label information representing a classification related to the menstrual cycle or a classification related to the pregnancy status. The inference unit 11 stores the inference information 3 regarding the menstrual cycle or pregnancy status output from the trained model 21 in the memory unit 14.

[0025] The evaluation unit 12 generates evaluation information 4 corresponding to the content of the inference information 3 generated by the inference unit 11. The evaluation unit 12 takes the inference information 3 generated by the inference unit 11 as input and generates evaluation information 4 corresponding to the input inference information 3. The evaluation information 4 is information that indicates an evaluation of the menstrual cycle or pregnancy status as grasped from the inference information 3. The evaluation unit 12 stores the generated evaluation information 4 in the storage unit 14.

[0026] The evaluation unit 12 may generate evaluation information 4 using a pre-trained evaluation model generated by machine learning, or it may generate evaluation information 4 by rule-based information processing, or it may generate evaluation information 4 by a combination (use) of a pre-trained evaluation model and rule-based information processing. As an example of rule-based information processing, the evaluation unit 12 generates evaluation information 4 corresponding to the input inference information 3 using evaluation data 22 that associates the content of the inference information 3 with the evaluation information 4.

[0027] The output unit 13 outputs inference information 3 generated by the inference unit 11. In this embodiment, in addition to the inference information 3, the output unit 13 also outputs evaluation information 4 corresponding to the inference information 3. In the example of Figure 1, the physiological phenomenon estimation device 1 can be connected to the display unit 31 by wired or wireless connection via an interface (not shown). The display unit 31 may be a standalone display device or a (built-in) display device provided in an information processing terminal such as a smartphone. The output unit 13 presents the inference information 3 and evaluation information 4 to the user US by outputting them to the display unit 31. The user US can obtain objective information (inference information 3) based on the detected data 2 regarding their menstrual cycle and pregnancy status while at home or elsewhere outside the hospital. The user US can understand the evaluation of their menstrual cycle and pregnancy status without requiring medical expertise by using the evaluation information 4, and take appropriate action, such as visiting a medical institution, if necessary. In addition to the display unit 31, the output unit 13 may also output the inference information 3 and evaluation information 4 via a network to, for example, a server or terminal at a medical institution. This allows for the sharing of objective inference information 3 and evaluation information 4 based on detection data 2 with the attending physician during consultations. This is expected to contribute to early medical intervention and improvement of treatment plans.

[0028] (Pre-trained model) In this embodiment, the trained model 21 includes a first trained model 21A that outputs inference information 3A (see Figure 2) regarding the menstrual cycle, and a second trained model 21B that outputs inference information 3B (see Figure 3) regarding the pregnancy status. The inference unit 11 selects the first trained model 21A and the second trained model 21B according to the user's selection or information regarding the presence or absence of pregnancy.

[0029] User US can choose whether to receive inference information 3A regarding menstrual cycles or inference information 3B regarding pregnancy status from the estimation device 1. Alternatively, User US can input user information to the estimation device 1 that identifies whether or not they are pregnant. The selection and input of user information to the estimation device 1 may be done using an input device (such as a keyboard) connected to the estimation device 1, or it may be done via communication from an information and communication terminal such as a smartphone. If User US selects inference information 3A regarding menstrual cycles, or if User US is not pregnant, the inference unit 11 selects the first trained model 21A and performs inference. If User US selects inference information 3B regarding pregnancy status, or if User US is pregnant, the inference unit 11 selects the second trained model 21B and performs inference. Note that regardless of whether or not User US is pregnant, the inference unit 11 may generate inference information 3A regarding menstrual cycles and inference information 3B regarding pregnancy status using both the first trained model 21A and the second trained model 21B, respectively.

[0030] The input data 5 (see Figures 2 and 3) may be the detection data 2 itself (i.e., raw data), or it may be secondary data generated by performing analysis, processing, or transformation on the detection data 2. If the input data 5 is raw data, it is time-series data or image data representing the time-series changes of the electrical output output by the sensor 30 (optical sensor or strain sensor). If the input data 5 is secondary data, it will be data in a form corresponding to the analysis, processing, or transformation process. The secondary data may be, for example, blood volume pulse (BVP) data or blood volume amplitude (BVA) data. Blood volume amplitude is data that shows the amplitude value of the volume pulse wave waveform. The secondary data may also be, for example, velocity pulse wave or acceleration pulse wave data, which are differential waveforms of the pulse wave waveform. The secondary data may also be, for example, heart rate waveform or heart rate data. As described above, the input data 5 may also be image data showing the time changes of these secondary data.

[0031] (Inferences regarding the menstrual cycle) Figure 2 is an explanatory diagram showing the first pre-trained model 21A. The first pre-trained model 21A is pre-trained by machine learning to take input data 5, which includes detected data 2 or secondary data based on detected data 2, as input and output inference information 3A regarding the menstrual cycle. The inference unit 11 infers the inference information 3A regarding the menstrual cycle by inputting the input data 5 into the first pre-trained model 21A.

[0032] In Figure 2, the input data 5 is conceptually represented by a waveform graph G1, where the vertical axis represents signal intensity and the horizontal axis represents time. The first trained model 21A outputs inference information 3A regarding the menstrual cycle. The output form of the inference information 3A regarding the menstrual cycle is not particularly limited. Figure 2 shows a graph G2 for conceptually explaining the inference information 3A regarding the menstrual cycle. In graph G2, the vertical axis represents the magnitude (or high / low) of an index related to the menstrual cycle and the horizontal axis represents time. The inference information 3A regarding the menstrual cycle may be output in the form of time-series data or time-varying waveform data, in the form of classification data into categories related to the menstrual cycle, or in the form of a single numerical value or a range of numerical values ​​related to the menstrual cycle.

[0033] For example, the inference information 3A regarding the menstrual cycle includes at least one of the following: information on physiological changes related to the menstrual cycle, information on the relevant cycle segment of the menstrual cycle, and information on the timing of ovulation or the onset of menstruation.

[0034] Physiological change information related to the menstrual cycle includes, for example, at least one of the following: hormone dynamics, basal body temperature dynamics, follicular diameter growth curves, and changes in endometrial tissue thickness. Physiological change information is output, for example, in the form of time-series data or time-varying waveform data.

[0035] Hormone dynamics refer to the changes in the production of hormones involved in the menstrual cycle. Hormones involved in the menstrual cycle include, for example, GnRH (gonadotropin-releasing hormone), LH (luteinizing hormone), FSH (follicle-stimulating hormone), hCG (human chorionic gonadotropin), inhibin, prolactin (PRL), thyroid hormones (TSH, FT4, FT3), estrogen, and progesterone, at least one of these. Increases and decreases in these hormones regulate physiological phenomena (menstruation). Hormone dynamics serve as an indicator of the timing of ovulation and other periods in the menstrual cycle, as well as an indicator of the health of hormone secretion. Graph G2 shows an example of the temporal changes in the production of LH, FSH, estrogen, and progesterone.

[0036] Basal body temperature dynamics refer to the changes in basal body temperature according to the menstrual cycle. Basal body temperature is divided into a low-temperature phase from the start of menstruation to ovulation and a high-temperature phase from after ovulation to the start of the next menstruation. Basal body temperature dynamics can be used as a guide to determine the ovulation day, for example, for pregnancy or contraception purposes. Graph G2 shows an example of the temporal changes in basal body temperature.

[0037] The follicular diameter growth curve shows the change in follicular diameter over time. Follicle diameter is an indicator of follicle size. Follicle diameter increases over time during the menstrual cycle, from the menstrual phase to the ovulation phase. After ovulation, the follicle becomes the corpus luteum. The follicular diameter growth curve can be used, for example, to determine the ovulation day for pregnancy or contraception purposes. Graph G2 shows an example of the change in follicular diameter over time.

[0038] Changes in endometrial tissue thickness refer to the change in endometrial tissue thickness over time. The endometrium increases in thickness over time, from the follicular phase to the ovulatory phase and the luteal phase. Changes in endometrial tissue thickness serve as indicators of endometrial tissue health and pregnancy rate (ease of conception). Graph G2 shows an example of changes in endometrial tissue thickness.

[0039] Information regarding the relevant periodic phase within the menstrual cycle is output, for example, in the form of classification data for the menstrual cycle phase (category). One menstrual cycle can be divided into four periods (categories), for example: menstrual phase, follicular phase, ovulation phase, and luteal phase. The number of periods is not limited to four; it may be three or fewer, or five or more. In this case, the inference information 3A regarding the menstrual cycle includes information indicating the relevant category (one of the menstrual phase, follicular phase, ovulation phase, or luteal phase). Information regarding the relevant periodic phase within the menstrual cycle may include the probability of corresponding to each of the four periodic phases. For example, assuming that data for time point t1 is input in graph G2, it will be output that time point t1 corresponds to the ovulation phase periodic phase, or the probability value corresponding to the ovulation phase periodic phase will be the largest among the probability values ​​for each periodic phase.

[0040] Information regarding the timing of ovulation or menstruation may be output, for example, in the form of a single numerical value or a range of numerical values ​​relating to the menstrual cycle. In this case, the inference information 3A regarding the menstrual cycle may include, for example, numerical information indicating the number of days from the time of inference processing to ovulation or the number of days until menstruation begins. For example, if we assume that time t1 is the expected ovulation day in graph G2, the number of days from the start of inference to time t1 will be output as information on the number of days until ovulation. The inference information 3A regarding the menstrual cycle may calculate the expected ovulation day and the expected menstruation start date as dates based on the above numerical information. The number of days (expected date) until ovulation or menstruation begins may also be calculated as a numerical range, such as between X and Y days.

[0041] Here, we will explain the principle of inferring inference information 3A regarding the menstrual cycle based on input data 5. Physiological phenomena associated with the menstrual cycle are regulated by the changes in the hormones involved in the menstrual cycle, and it is known that these changes affect systemic vascular resistance (i.e., blood flow), including peripheral blood vessels, through phenomena such as blood flow redistribution and vasodilation represented by estrogen. Graph G3 in Figure 2 illustrates the concept of changes in peripheral blood flow, with the vertical axis representing blood flow and the horizontal axis representing time.

[0042] The detection data 2 (input data 5) detected by the sensor 30 placed on the user US's body surface includes a signal component indicating changes in peripheral blood flow, as shown in graph G3. These changes in peripheral blood flow occur in conjunction with various physiological phenomena associated with the menstrual cycle, as shown in graph G2. Therefore, by acquiring features that show the relationship between the detection data 2 (input data 5), ​​as shown in graph G1, and the information regarding the menstrual cycle, as shown in graph G2, through machine learning, it becomes possible to infer menstrual cycle-related inference information 3A from the input data 5.

[0043] (Inferences regarding pregnancy status) Figure 3 is an explanatory diagram showing the second pre-trained model 21B. The second pre-trained model 21B is pre-trained by machine learning to take input data 5, which includes detected data 2 or secondary data based on detected data 2, as input and output inference information 3B regarding the pregnancy state. As shown in Figure 3, the inference unit 11 infers the inference information 3B regarding the pregnancy state by inputting the input data 5 into the second pre-trained model 21B.

[0044] In Figure 3, the concept of input data 5 is shown by the waveform graph G1. The second trained model 21B outputs inference information 3B regarding the pregnancy state. The output form of the inference information 3B regarding the pregnancy state is not particularly limited. The inference information 3B regarding the pregnancy state may be output in the form of time series data or time-varying waveform data, in the form of classification data into categories regarding the pregnancy state, or in the form of a single numerical value or a range of numerical values ​​regarding the pregnancy state.

[0045] Inference information 3B regarding the state of pregnancy includes, for example, information on at least one of the following: maternal uterine contractions, fetal intrauterine movement, maternal amniotic fluid volume, cervical ripening, gestational age, estimated fetal weight, and timing of onset of labor.

[0046] Figure 3 illustrates information about maternal uterine contractions as inferred information 3B regarding the state of pregnancy. Figure 3 shows graph G11 to conceptually explain uterine contractions. In graph G11, the vertical axis represents the intensity of uterine contractions, and the horizontal axis represents time.

[0047] Information regarding maternal uterine contractions is output, for example, in the form of time-series data or time-varying waveform data. This information includes, for example, the duration L of a uterine contraction, the interval D between contractions, the cycle of contractions, the frequency of contractions, and the intensity waveform of a uterine contraction. The duration L is the length of time from the start to the end of a single uterine contraction. The interval D is the time interval from the end of one uterine contraction to the start of the next. The cycle of contractions is the time interval from the start of one uterine contraction to the start of the next. The cycle of contractions corresponds to the sum of the duration L and the interval D. The frequency of contractions is the number of uterine contractions per period. The intensity waveform of a uterine contraction shows the time change in the intensity of a single uterine contraction, as shown in graph G11. Information regarding uterine contractions serves as an indicator of the timing of delivery and the health of the pregnancy.

[0048] Information regarding fetal intrauterine movement (fetal movement) includes, for example, either fetal movement signal waveforms or fetal movement counts. Fetal movement signal waveforms are time-series data or time-varying waveform data corresponding to the Doppler waveform of fetal movement obtained by an ultrasound Doppler device. Fetal movement counts are numerical information of the number of fetal movements that occurred per unit time. In this specification, fetal intrauterine movement refers to the movement (action) of the fetus's body within the womb and is a broad concept encompassing fetal muscle tone, fetal movement, and respiratory-like movements. Fetal movement specifically refers to the transient movements of the fetus and does not include the physiological activities of internal organs that are constantly occurring, such as the fetal heartbeat. Fetal intrauterine movement includes, for example, fetal FE muscle tone, fetal movement in the narrow sense, and respiratory-like movements. Fetal FE muscle tone refers to the flexion and extension movements of a part of the fetal FE's body (trunk or limbs). Examples of muscle tone include the movement of the spine or limbs from a flexed position to extend and return to their original flexed position, and the opening and closing movements of the palms. In the narrow sense, fetal movement refers to single or combined movements of the fetal FE's trunk and limbs, but it signifies movements greater than muscle tone. Examples of fetal movement in the narrow sense include movements such as changing body position or kicking the uterine wall. Respiratory movements of the fetal FE are movements similar to breathing. Examples of respiratory movements include intermittent movements of the fetal FE's diaphragm, abdominal wall, and rib cage that last for several seconds to tens of seconds. Information on fetal intrauterine movement (fetal movement) serves as an indicator of the health of the fetal FE in the mother's womb.

[0049] Information on maternal amniotic fluid volume includes classification information such as polyhydramnios, normal, oligohydramnios, and rupture of membranes. This information may be numerical data corresponding to the classification, or it may be probability values ​​corresponding to each classification. Amniotic fluid volume gradually increases from early to late pregnancy, and then gradually decreases. Excessive or insufficient amniotic fluid may indicate maternal and / or fetal abnormalities and serve as an indicator of the health of the pregnancy.

[0050] Information on cervical ripening includes, for example, index values ​​for cervical ripening or information on cervical length. Cervical ripening refers to the softening of the cervix as the time of delivery approaches, and the resulting changes in the morphology of the cervix. The Bishop Score is known as an index value for cervical ripening. The Bishop Score is divided into four categories (index values): 0, 1, 2, and 3. The index value for cervical ripening may be numerical information corresponding to this Bishop Score, or it may be a probability value corresponding to each of the four categories (index values). Also, cervical length shortens as the time of delivery approaches. Information on cervical length is classification information such as whether the cervical length is normal or short, and may be numerical information indicating the corresponding classification, or it may be a probability value corresponding to each classification. Information on cervical ripening serves as an indicator of the time of delivery and the possibility of premature birth.

[0051] The gestational age information is numerical information indicating the gestational age at the time the inference process is executed. The estimated fetal weight is numerical information indicating the estimated fetal weight at the time the inference process is executed. The timing of the onset of labor is numerical information indicating the number of days from the time the inference process is executed until the onset of labor. The number of days until the onset of labor may be converted to the expected date of onset of labor. The inference information 3B regarding the pregnancy status may include gestational age, estimated fetal weight, and timing of the onset of labor as numerical ranges, for example, between week X and week Y, between X grams and Y grams, and between day X and day Y.

[0052] Here, we will explain the principle of inferring inference information 3B regarding the pregnancy state based on input data 5. Uterine contractions are accompanied by vasoconstriction near the uterus, resulting in local changes in blood flow to surrounding tissues, including the uterine arteries. During uterine contractions, blood flow in the uterine arteries temporarily decreases (see, for example, the non-patent literature below). It is known that changes in local blood flow, including that of the uterine arteries, can affect blood volume in peripheral blood vessels through the blood flow redistribution phenomenon. Graph G12 in Figure 3 shows the concept of changes in blood flow in the uterine arteries, with blood flow on the vertical axis and time on the horizontal axis. Graph G13 in Figure 3 shows the concept of changes in blood flow in peripheral blood vessels, with blood flow on the vertical axis and time on the horizontal axis. Graphs G12 and G13 show that when blood flow in the uterine arteries increases or decreases, peripheral blood flow tends to increase or decrease in the opposite direction to that of the uterine arteries to compensate for this. Non-patent literature: H. Li, S. Gudmundsson, P. Olofsson, “Uterine artery blood flow velocity waveforms during uterine contractions”, Ultrasound Obstet Gynecol, 2003; 22: 578-585, [online] URL<https: / / doi.org / 10.1002 / uog.921> Furthermore, rupture of membranes, an indicator of pregnancy status and labor progression, is suggested to cause a rapid release of compression of the inferior vena cava, thereby affecting venous return. These changes in venous return lead to changes in peripheral blood flow, which can be detected by sensor 30.

[0053] The detection data 2 (input data 5) detected by the sensor 30 placed on the user US's body surface includes a signal component indicating changes in peripheral blood flow, as shown in graph G13. These changes in peripheral blood flow occur in conjunction with the blood flow in the uterine arteries, as shown in graph G12. Therefore, by acquiring features that show the relationship between detection data 2 (input data 5) and maternal uterine contractions through machine learning, it becomes possible to infer pregnancy status information 3B (information regarding uterine contractions) from input data 5.

[0054] Similarly, the information mentioned above, such as fetal intrauterine movement, maternal amniotic fluid volume, cervical ripening, gestational age, estimated fetal weight, and timing of labor onset, correlates with changes in maternal cardiovascular blood flow, and changes in maternal cardiovascular blood flow and peripheral vascular blood flow are linked by the blood flow redistribution phenomenon. For example, regarding gestational age, cardiac output (CO) is known to begin increasing at 10 weeks of gestation and to exceed pre-pregnancy CO by 30-50% between 25 and 30 weeks. Therefore, inference becomes possible by machine learning to extract features that show the relationship between input data 5 and this information.

[0055] (Preprocessing of input data) As shown in Figures 2 and 3, the inference unit 11 may perform preprocessing on the input data 5 (raw data or secondary data) and input the preprocessed input data 5 to the trained model 21. The inference unit 11 does not have to perform preprocessing. Examples of preprocessing include removal processing to remove specific signal components contained in the input data 5, smoothing processing to smooth out fluctuating components, and differentiation processing.

[0056] The removal process includes, for example, a frequency filtering process that removes signal components in a predetermined frequency band from the input data 5. The removal process also includes, for example, a process that extracts predetermined feature points from the input data 5 (removing data other than the extracted feature points). The predetermined feature points may be, for example, the baseline (baseline level), peaks (high peaks and low peaks), the start or end point of a peak (intersection with the baseline) of the waveform data of the input data 5. By removing signal components (noise) that are unnecessary for inference from the input data 5, the inference accuracy can be improved, and the processing load can be reduced by reducing the number of data points. The smoothing process is, for example, an averaging process. Examples of averaging processes include the arithmetic mean, weighted mean, geometric mean, and harmonic mean. The smoothing process may also be a fitting (approximation) process to an arbitrary function. By removing noise and outliers through the smoothing process, the inference accuracy can be improved and the variability of the inference results can be reduced.

[0057] (Additional information) The input data 5 may further include supplementary information 6 in addition to the detection data 2 or secondary data. In this case, the information acquisition unit 10 acquires supplementary information 6 related to the menstrual cycle and pregnancy status separately from the detection data 2.

[0058] Figure 4 shows an example of an inference unit 11 that performs inference based on input data 5 including supplemental information 6. In this example, the trained model 21 is constructed as a model that corresponds to multi-dimensional input data 5 so that it can accept both detection data 2 (or its secondary data) and supplemental information 6 as input. In this case, the trained model 21 is pre-trained by machine learning to output information about the menstrual cycle or pregnancy status based on the detection data 2 (or its secondary data) and supplemental information 6. Supplemental information 6 is also included in the training data in machine learning. When performing inference from the detection data 2 of the user US, the user US provides the supplemental information 6 to the information acquisition unit 10. The information acquisition unit 10 may acquire the data of supplemental information 6 from an electromagnetic recording medium. Supplemental information 6 may be recorded, for example, as part of electronic medical record information on a server of a medical institution, and the information acquisition unit 10 may read the supplemental information 6 recorded on the server via the network based on the user US's operation input. User US's operation input may be a direct operation to the estimation device 1 (information acquisition unit 10) using an input device, or it may be an indirect operation such as sending a command to the estimation device 1 from an information communication terminal such as a smartphone.

[0059] Supplementary information 6 is information about the user US other than the detected data 2, and is used for the purpose of improving the accuracy of inference. Supplementary information 6 may include, for example, the user's vital signs, basic information about the user's US, medical information about the user's US, or information about the user's US condition at the time of measurement.

[0060] Vital signs include, for example, at least one of the following: respiratory rate, body temperature, surface temperature, basal body temperature, pulse rate, blood pressure, urine output, and blood oxygen saturation. These exemplified vital signs may be, for example, data measured in the past (data from a past point in time or statistically representative values) and do not necessarily have to be data from the time of inference (real time).

[0061] User US basic information includes, for example, race, nationality, age, height, weight, BMI (Body Mass Index), subcutaneous fat thickness, and body fat percentage.

[0062] User US medical information includes obstetric and general items, but there is no need to distinguish between them. The medical information for obstetric items includes, for example, one or more of the following: pregnancy and childbirth history, gestational age, estimated weight of the most recent fetus, urinary glucose, urinary protein, presence or absence of edema, background of pregnancy (natural conception, artificial insemination, ART (Assisted Reproductive Technology)), turning points in pregnancy (miscarriage, ectopic pregnancy, simultaneous internal and external pregnancy), induced abortion, live birth / stillbirth, history of fetal reduction surgery, number of births, medical history prior to pregnancy, medication history, history of supplement use, allergy history, fetal position, fetal presentation, placental position, obstetric complications, lactation history, blood test findings (blood type, complete blood count, biochemical test values, semen analysis, AMH (anti-Müllerian hormone), hysterosalpingography, anti-sperm antibody test, hyperphospholipid antibody test (infertility test), cervical cancer screening, endometrial cancer screening, breast cancer screening, thyroid hormone test, hysteroscopy, hepatitis B and C tests, etc.), and history of sexually transmitted infections. General medical information includes any of the following: information about menstruation (menarche, cycle, regular / irregular, menstrual period, last menstrual period), sexual history, history and method of contraception, pregnancy or breastfeeding, family history, past medical history (deep vein thrombosis, pulmonary embolism, antiphospholipid syndrome, cerebrovascular disease, coronary artery disease, breast cancer, herpes zoster of pregnancy, otosclerosis, heart valve disease, hypertension, diabetes mellitus, lipid metabolism disorders (hyperlipidemia), gallbladder disease, migraine (diagnosed), cervical cancer, endometrial cancer, porphyria, liver dysfunction, epilepsy, tetany, Crohn's disease, ulcerative colitis, glaucoma, asthma, thyroid disease, renal dysfunction, collagen disease, systemic lupus erythematosus, gastric / duodenal ulcers, etc.), lifestyle (alcohol consumption, smoking, occupational history, etc.), and vaccination history.

[0063] The user's US status at the time of measurement is information about the user's US status that may affect the detected data 2 during measurement by the sensor 30. The user's US status at the time of measurement includes, for example, time information, location information (positioning information and angular velocity, angular acceleration information, etc.), posture information at the time of measurement (standing, sitting, lying down, etc.), and the degree of skin moisture at the installation site of the sensor 30.

[0064] The detection data 2 obtained from the user US may be statistically biased against the multiple training data used for learning, depending on the content of the accompanying information 6. By training the trained model 21 with the accompanying information 6 along with the input data 5 (training data) during training, the impact of the statistical bias due to the accompanying information 6 of the individual user (or detection data) on the inference results can be learned. As a result, by considering the accompanying information 6 during inference, the inference accuracy can be improved for a diverse range of users US.

[0065] Thus, in this embodiment, the trained model 21 may generate inference information 3 based on input data 5 including supplementary information 6, or it may generate inference information 3 based on input data 5 that does not include supplementary information 6.

[0066] (Output of inference results) Figure 5 is an explanatory diagram illustrating the output of inference information 3. As shown in Figure 5, the data of inference information 3 generated by the inference unit 11 is supplied to the evaluation unit 12 and the output unit 13. Inference information 3 includes either inference information 3A related to the menstrual cycle or inference information 3B related to the pregnancy state.

[0067] The evaluation unit 12 generates evaluation information 4 based on the data of the inference information 3 and the evaluation data 22 pre-stored in the storage unit 14. Specifically, the evaluation unit 12 generates evaluation information 4A regarding the menstrual cycle based on the inference information 3A regarding the menstrual cycle and the evaluation data 22A regarding the menstrual cycle. The evaluation unit 12 generates evaluation information 4B regarding the pregnancy state based on the inference information 3B regarding the pregnancy state and the evaluation data 22B regarding the pregnancy state. The evaluation information 4 (evaluation information 4A or evaluation information 4B) generated by the evaluation unit 12 is supplied to the output unit 13. In this embodiment, the estimation device 1 is shown to include an evaluation unit 12, but the estimation device 1 does not necessarily have to include an evaluation unit 12.

[0068] The evaluation data 22 includes, for example, a table that associates multiple numerical ranges (thresholds) for classifying the inference information 3 with evaluation information 4 associated with each numerical range. The evaluation data 22 also includes a variable for inputting the value of the inference information 3 and a formula for determining the evaluation category of the inference information 3. The evaluation data 22 is created in advance based on the information obtained from the inference information 3 and the knowledge of specialists in the relevant field. Evaluation data 22A related to the menstrual cycle and evaluation data 22B related to pregnancy status are created separately and pre-recorded in the storage unit 14.

[0069] The output unit 13 outputs output data, including the inference information 3 generated by the inference unit 11 and the evaluation information 4 generated by the evaluation unit 12, to the display unit 31. The output unit 13 presents this information to the user by displaying the inference information 3 and evaluation information 4 on the display unit 31. If the estimation device 1 does not have an evaluation unit 12, the output unit 13 outputs the inference information 3.

[0070] The output evaluation information 4 is, for example, output in the form of a text (message) that explains the situation and advice gathered from the inference information 3. In addition to text, evaluation information 4 may also include displays indicating the degree of goodness of the inference information 3, such as a score format (scores from 0 to 100), a ranking format (A rank, B rank, C rank, etc., from best to worst), and color identification (red, blue, green, yellow, etc.). As a result, if the inference information 3 is good, evaluation information 4 to that effect will be output. Consequently, even users without specialized knowledge can obtain an objective evaluation, reducing unnecessary visits and thus alleviating the burden on pregnant women and healthcare professionals.

[0071] The evaluation unit 12 may output evaluation information 4 based on the inference information 3, which includes information indicating the need to visit a medical institution, whether medical intervention is recommended, or the presence of signs of disease, etc. This is expected to contribute to early medical intervention for important signs and improvement of treatment strategies for perinatal complications.

[0072] For example, the evaluation information 4A regarding the menstrual cycle can include information from the inference information 3A regarding the menstrual cycle, such as information on the possibility of diseases related to the menstrual cycle, information on the period of ovulation, information on the possibility of pregnancy, and information on the likelihood of pregnancy.

[0073] Furthermore, for example, inference information 3B regarding the state of pregnancy may include signs of premature birth or placental abruption. The likelihood of premature birth is indicated by the presence or absence of true labor, which promotes cervical maturation. True labor is characterized by the strength, duration L, and interval D of the uterine contractions. In true labor, an increase in the strength and duration L of the uterine contractions is observed over time. In false labor, which does not lead to cervical maturation, the uterine contractions weaken or stop over time. Placental abruption is a serious condition, and early treatment intervention improves the perinatal outcome for both mother and child. A characteristic finding of placental abruption is the possibility of persistent uterine contractions. Therefore, assessment information 4B regarding the state of pregnancy may include, for example, information regarding the presence of signs of premature birth or placental abruption.

[0074] (Generating a pre-trained model) Figure 6 is an explanatory diagram illustrating the generation of the first pre-trained model 21A (machine learning). Figure 7 is an explanatory diagram illustrating the generation of the second pre-trained model 21B (machine learning). The learning algorithms for the pre-trained models 21 (first pre-trained model 21A, second pre-trained model 21B) are not particularly limited, but include, for example, supervised learning, unsupervised learning, and reinforcement learning, and specifically include one or more combinations of any of the following: CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), LSTM (Long Short-Term Memory), attention mechanism (including Transformer and BERT with attention mechanism), and as reinforcement learning, Q-learning, TD-learning, dynamic programming, Monte Carlo method, SARSA, deep reinforcement learning, actor-critic method, policy gradient method, multi-armed bandit method, hierarchical reinforcement learning, approximate dynamic programming, inverse reinforcement learning, multi-agent reinforcement learning, self-supervised reinforcement learning, continuous reinforcement learning, expectation maximization reinforcement learning, etc.

[0075] As shown in Figures 6 and 7, the trained model 21 is generated when the learning unit ML, a computer that performs machine learning, performs machine learning on the trained model 40 using training data. The training data includes data actually measured by collaborators (sample providers) who create the training data. The training data can be created, for example, from measurement data using specialized equipment in a hospital. Methods for acquiring training data may include one or more combinations of the following: fetal monitoring devices (intrauterine movement, uterine contractions), blood testing devices (hormone dynamics), urine component testing devices (hormone dynamics), internal examination (all items), electrocardiogram or electromyogram (intrauterine movement, uterine contractions), ultrasound imaging devices (intrauterine movement, follicular diameter, endometrial tissue thickness, amniotic fluid volume, cervical ripening), cervical mucus examination (cervical ripening), blood pressure monitor, pulse oximeter, thermometer (basal body temperature, surface temperature), and patient's chief complaint (menstrual cycle, basal body temperature, uterine contractions, intrauterine movement). Training data is created in a format that matches the output data of the trained model 21, based on data obtained by these devices (raw data) or secondary data thereof, or on the opinions of experts such as doctors regarding the obtained data.

[0076] As shown in Figure 6, the training data 41A used to generate the first trained model 21A includes pairs of input training data 42A and output training data 43A. The input training data 42A is the detection data 2 measured by the sensor 30 or its secondary data (i.e., input data 5). The input data 5 may be preprocessed, such as by frequency filtering. The output training data 43A is data created based on data measured at the same time as the input training data 42A. The output training data 43A is in the same format as the inference information 3A regarding the menstrual cycle that should be generated by the inference unit 11. The learning unit ML uses the training data 41A to cause the learning model 40 to acquire features using deep learning, which are necessary to convert the input training data 42A (input data 5) into the output of the inference information 3A regarding the menstrual cycle. The features are the weights or parameters of each node in each layer (output layer, input layer, hidden layer) that constitutes the learning model 40. The learning unit ML determines the features of the learning model 40 by performing training using multiple training datasets 41A. The learning model 40 whose features have been determined by machine learning is the first trained model 21A. Preferably, the training datasets 41A include data for at least one menstrual cycle from various data providers.

[0077] As shown in Figure 7, the training data 41B used to generate the second pre-trained model 21B includes pairs of input training data 42B and output training data 43B. The input training data 42B is the detection data 2 measured by the sensor 30 or its secondary data (i.e., input data 5). The output training data 43B is data created based on data measured at the same time as the input training data 42B. The output training data 43B is in the same format as the inference information 3B regarding pregnancy status generated by the inference unit 11. The learning unit ML uses the training data 41B to have the learning model 40 acquire features using deep learning to convert the input training data 42B (input data 5) into the output of the inference information 3B regarding pregnancy status. The learning unit ML determines the features of the learning model 40 by learning using multiple training data sets 41B. The learning model 40 whose features have been determined by machine learning is the second pre-trained model 21B.

[0078] The training data 41B may consist of data from all weeks of pregnancy from the establishment of the pregnancy, or it may consist of data from a specific period (weeks) of pregnancy, depending on the type and content of the inference information 3B to be estimated. For example, if the inference information 3B to be estimated is information about fetal intrauterine movement, the training data 41B may consist of data from a specific period (weeks), or it may consist of data from all periods starting from the minimum gestational age at which fetal movement can be detected by Doppler / echo.

[0079] As shown in Figures 6 and 7, in the generation of the trained models 21 (first trained model 21A and second trained model 21B), the input training data may include supplementary information 6. In this case, the input training data includes detection data 2 or its secondary data obtained from the sample provider and the supplementary information 6 of the sample provider. The trained model 40 uses a model that corresponds to the multi-dimensional input data 5 consisting of time series data and supplementary information 6. As a result, the trained model 40 learns features from the detection data 2 or its secondary data and the supplementary information 6 to convert them into the output of inference information 3.

[0080] The memory unit 14 of the physiological phenomenon estimation device 1 shown in Figure 1 stores the trained models 21 (first trained model 21A, second trained model 21B) generated in this manner in advance.

[0081] (Methods for estimating physiological phenomena) Next, the method for estimating physiological phenomena according to this embodiment will be described. Figure 8 is a flowchart illustrating the method for estimating physiological phenomena according to this embodiment. The method for estimating physiological phenomena is also the method for operating the physiological phenomenon estimation device 1.

[0082] As shown in Figure 8, the method for estimating physiological phenomena includes: step S10, which involves acquiring detection data 2 obtained by a sensor 30 (optical sensor or strain sensor) placed on the body surface that detects signals reflecting changes in blood flow caused by the menstrual cycle or pregnancy state over time; step S20, which involves inputting the input data 5 obtained from the acquired detection data 2 into a trained model 21 that takes the detection data 2 or secondary data based on the detection data 2 as input and outputs information about the menstrual cycle or pregnancy state, to generate inference information 3 about the menstrual cycle or pregnancy state; and step S40, which involves outputting the generated inference information 3. In the example in Figure 8, the method for estimating physiological phenomena further includes step S30, which involves generating evaluation information 4 according to the content of the inference information 3 generated by the inference unit 11. Note that in this embodiment, the step S30 for generating evaluation information 4 may be omitted.

[0083] In step S10, the information acquisition unit 10 acquires the detection data 2. The information acquisition unit 10 acquires the detection data 2 from the sensor 30 placed on the surface of the user US. The information acquisition unit 10 may also acquire the detection data 2 by reading the data from an electromagnetic recording medium that stores the detection data 2. If the input data 5 includes supplementary information 6, the information acquisition unit 10 further acquires the supplementary information 6 of the user US.

[0084] In step S20, the inference unit 11 inputs the input data 5 into the trained model 21 to infer inference information 3 regarding the menstrual cycle or pregnancy status. The inference unit 11 selects between a first trained model 21A and a second trained model 21B depending on the user US's selection or information regarding pregnancy status. The inference unit 11 selects the first trained model 21A if the user US selects inference information 3A regarding the menstrual cycle, or if the user US is not pregnant. The inference unit 11 selects the second trained model 21B if the user US selects inference information 3B regarding pregnancy status, or if the user US is pregnant. The inference unit 11 generates inference information 3 (inference information 3A regarding the menstrual cycle or inference information 3B regarding pregnancy status) by inputting the input data 5 into the selected trained model 21 and performing inference processing.

[0085] In step S30, the evaluation unit 12 generates evaluation information 4 based on the data of the inference information 3 and the evaluation data 22. The evaluation unit 12 generates evaluation information 4 according to the type of inference information 3. That is, if inference information 3A regarding the menstrual cycle is generated, the evaluation unit 12 generates evaluation information 4A regarding the menstrual cycle based on the inference information 3A regarding the menstrual cycle and the evaluation data 22A regarding the menstrual cycle. If inference information 3B regarding the pregnancy state is generated, the evaluation unit 12 generates evaluation information 4B regarding the pregnancy state based on the inference information 3B regarding the pregnancy state and the evaluation data 22B regarding the pregnancy state.

[0086] In step S40, the output unit 13 outputs output data including the inference information 3 generated by the inference unit 11 to the destination device. In this embodiment, in addition to the inference information 3, the output unit 13 also outputs evaluation information 4 generated by the evaluation unit 12. The destination device includes a display unit 31. As a result, the inference information 3 and the evaluation information 4 are displayed on the display unit 31. Depending on the type of inference information 3, the user US is provided with either a set of inference information 3A related to the menstrual cycle and evaluation information 4A related to the menstrual cycle, or a set of inference information 3B related to pregnancy status and evaluation information 4B related to pregnancy status. If the step S30 for generating evaluation information 4 is not included, in step S40, the output unit 13 outputs output data including only the inference information 3.

[0087] (Example of a configuration for a physiological phenomenon estimation device) The physiological phenomenon estimation device 1 described above can be implemented in various forms other than the example shown in Figure 1.

[0088] Figure 9 is an explanatory diagram showing a first configuration example of the physiological phenomenon estimation device 1 according to this embodiment. In the example of Figure 9, the physiological phenomenon estimation device 1 is realized by a wearable device 100. The wearable device 100 is a small device that can be attached to a part of the body. The wearable device 100 has a band (strap-shaped) form that can be attached to, for example, the wrist, ankle, upper arm, thigh, etc. Figure 9 shows an example where the wearable device 100 is a smartwatch (a wristwatch-type information terminal worn on the wrist).

[0089] The wearable device 100 comprises a control unit 102, a storage unit 14, a display unit 31, and a sensor 30. The control unit 102, storage unit 14, display unit 31, and sensor 30 are all integrated into a single housing 101.

[0090] The control unit 102 is implemented by a processor such as a CPU. The processor functions as a control unit 102 that controls the wearable device 100 by performing calculations according to the program stored in the memory unit 14. The memory unit 14 stores the program 20 of the physiological phenomenon estimation device 1. The information acquisition unit 10, inference unit 11, evaluation unit 12, and output unit 13 of the physiological phenomenon estimation device 1 are implemented by the control unit 102 of the wearable device 100 executing the program 20.

[0091] The memory unit 14 includes a non-volatile semiconductor memory. The memory unit 14 stores a program (operating system or firmware) for controlling the wearable device 100 and a program 20 (application program) for the physiological phenomenon estimation device 1. The memory unit 14 also stores the trained model 21 and evaluation data 22 shown in Figure 1.

[0092] The display unit 31 includes a self-emissive display device such as a liquid crystal display device or an organic EL (Electro-Luminescence) display device. The display unit 31 displays various information from the wearable device 100 under the control of the control unit 102. The wearable device 100 includes an input device such as a touch panel integrated with the display unit 31, and an operation unit 103 such as input buttons provided on the housing 101. The control unit 102 performs information acquisition and inference processing in response to operation input to the operation unit 103.

[0093] Sensor 30 is either an optical sensor or a strain sensor. Sensor 30 is built into the housing 101. Sensor 30 is located on the inner surface of the housing 101 (or band) that is in contact with the wearer's wrist.

[0094] In this configuration, the information acquisition unit 10 acquires detection data 2 from the sensor 30 (optical sensor or strain sensor) provided by the wearable device 100. The generation of inference information 3 and evaluation information 4 by the inference unit 11 and evaluation unit 12 is realized by computational processing by the control unit 102 of the wearable device 100 executing program 20. The output unit 13 outputs the inference information 3 and evaluation information 4 to the display unit 31 of the wearable device 100. The display unit 31 displays the inference information 3 and evaluation information 4 supplied from the output unit 13. In this way, the physiological phenomenon estimation device 1 can be realized by the wearable device 100. In other words, the wearable device 100 executes the physiological phenomenon estimation method according to this embodiment by executing program 20.

[0095] Figure 10 is an explanatory diagram showing a second configuration example of the physiological phenomenon estimation device 1 according to this embodiment. For convenience, Figure 10 shows two examples together: one in which the sensor 30 is a nail-type wearable device 110 attached to a fingernail, and another in which the sensor 30 is a ring-type wearable device 120 attached to a finger. Either the wearable device 110 or the wearable device 120 can be used individually; it is not necessary to use both.

[0096] The artificial nail-type wearable device 110 is attached to the surface of the fingernail NS by means of adhesive, adhesive tape, or a clip, for example. The wearable device 110 is equipped with a sensor 30. The wearable device 110 is connected to an information and communication terminal 130 via wired or wireless connection for bidirectional communication. The wearable device 110 outputs the detection signal from the sensor 30 to the information and communication terminal 130. The information and communication terminal 130 is, for example, a smartphone, a tablet device, or a PC (Personal Computer).

[0097] Sensor 30 includes, for example, a sheet-shaped piezoelectric element (piezoelectric film sheet). When blood flow in the peripheral blood vessels at the fingertips changes, the blood vessels near the nail NS pulsate, and this pulsation of blood vessels causes minute deformation (strain) in the nail NS. Sensor 30 detects this minute deformation in the nail NS and measures a signal that reflects the change in blood flow.

[0098] For example, a wearable device in the form of a false nail equipped with such a strain sensor can use the device disclosed in Japanese Patent Application Publication No. 2020-81689. Alternatively, in a wearable device in the form of a false nail, an optical sensor may be provided as the sensor 30. The optical sensor can detect changes in blood flow by an optical signal that passes through the nail NS and reaches the peripheral blood vessels at the fingertip.

[0099] The wearable device 110 may include a control circuit 111 in addition to the sensor 30. The control circuit 111 may include a detection circuit for detecting signals (voltage or resistance changes) output from the sensor 30, a power supply circuit, and an interface circuit for inputting and outputting data with the information and communication terminal 130.

[0100] The ring-shaped wearable device 120 has an annular shape and is worn by fitting a finger HF onto its inner circumference. The wearable device 120 is equipped with a sensor 30. The wearable device 110 is connected to the information and communication terminal 130 via wired or wireless bidirectional communication. The wearable device 110 outputs the detection signal from the sensor 30 to the information and communication terminal 130.

[0101] The sensor 30 is built into the housing of the wearable device 120. The sensor 30 is located on the inner surface of the housing that comes into contact with the wearer's finger HF. In addition to the sensor 30, the wearable device 110 may also include a control circuit 121. The control circuit 121 may include a detection circuit for detecting signals (voltage and resistance changes) output from the sensor 30, a power supply circuit, and an interface circuit for inputting and outputting data with the information and communication terminal 130.

[0102] In the example shown in Figure 10, the physiological phenomenon estimation device 1 according to the embodiment is configured by the information and communication terminal 130. The information acquisition unit 10, inference unit 11, evaluation unit 12, and output unit 13 of the physiological phenomenon estimation device 1 are realized by the processor of the information and communication terminal 130 executing the program 20.

[0103] With this configuration, the information acquisition unit 10 acquires detection data 2 from the sensor 30 (optical sensor or strain sensor) provided by the wearable device 110 or wearable device 120.

[0104] As shown in Figures 9 and 10, the sensor 30 can be installed on various types of wearable devices. The user US can move freely while wearing the wearable device, and has a high degree of freedom in their posture when acquiring detection data 2. As a result, the spatial and physical constraints associated with acquiring detection data 2 are greatly reduced.

[0105] Figure 11 is an explanatory diagram showing a third configuration example of the physiological phenomenon estimation device 1 according to this embodiment. In the example of Figure 11, the physiological phenomenon estimation device 1 is implemented by a server 200 connected to a network NW. The inference information 3 and evaluation information 4 generated by the physiological phenomenon estimation device 1 are provided to the user in the form of a so-called cloud service.

[0106] Server 200 is a computer capable of communication via a network NW. Server 200 comprises an information acquisition unit 10, an inference unit 11, an evaluation unit 12, an output unit 13, and a storage unit 14. The information acquisition unit 10, inference unit 11, evaluation unit 12, and output unit 13 of the physiological phenomenon estimation device 1 are realized on program 20 (see Figure 1) by the server 200 executing program 20 stored in the storage unit 14. Server 200 may consist of a single device or a group of multiple devices (a group of servers). For example, server 200 may consist of a server device that functions as an information acquisition unit 10, a server device that functions as an inference unit 11, a server device that functions as an evaluation unit 12, a server device that functions as a storage unit 14, etc., and the physiological phenomenon estimation device 1 may be configured by mutual communication between each server device. The output unit 13 is realized by the communication function of the server device.

[0107] In the example shown in Figure 11, user US utilizes the physiological phenomenon estimation device 1 using an information and communication terminal 210 that can communicate with server 200 via a network NW. The information and communication terminal 210 can be a smartphone, tablet, PC, smartwatch, etc. The information and communication terminal 210 comprises a control unit 211, a communication unit 212, a display unit 213, and a storage unit 214. The storage unit 214 stores an application program for utilizing the functions of the physiological phenomenon estimation device 1 through communication with server 200. The control unit 211 acquires detection data 2 from sensors 30 placed on the surface of user US via the communication unit 212. By executing the application program, the control unit 211 establishes communication with server 200 via the communication unit 212 and transmits the detection data 2 to server 200.

[0108] In the server 200, the information acquisition unit 10 acquires detection data 2 transmitted from the information communication terminal 210. The inference unit 11 inputs the detection data 2 into the trained model 21 stored in the memory unit 14 and performs inference processing to generate inference information 3. The evaluation unit 12 generates evaluation information 4 based on the inference information 3. The output unit 13 outputs the inference information 3 and evaluation information 4 to the information communication terminal 210 via the network NW. The server 200 (estimation device 1) does not necessarily have an evaluation unit 12, in which case evaluation information 4 is not generated, and the output unit 13 outputs the inference information 3 to the information communication terminal 210.

[0109] The control unit 211 of the information and communication terminal 210 receives inference information 3 from the server 200 via the communication unit 212. If evaluation information 4 is generated and output, the control unit 211 receives evaluation information 4 in addition to the inference information 3 from the server 200. The control unit 211 displays the received inference information 3 on the display unit 213. If evaluation information 4 is generated and output, the control unit 211 displays both the inference information 3 and the evaluation information 4 on the display unit 213. Thus, in the example of Figure 11, the physiological phenomenon estimation device 1 is realized by the server 200 that provides cloud services. In other words, the server 200 executes the physiological phenomenon estimation method according to this embodiment by executing the program 20.

[0110] Figure 10 shows an example where the information and communication terminal 130 functions as the physiological phenomenon estimation device 1, whereas in Figure 11, the information and communication terminal 210 functions as an operation and display device for exchanging and displaying data by communicating with the physiological phenomenon estimation device 1 (i.e., the server 200). For example, the smartwatch-type wearable device 100 shown in Figure 9 and the information and communication terminal 130 shown in Figure 10 may also function as the operation and display device shown in Figure 11 by installing an application program that utilizes the functions of the estimation device 1 through communication with the server 200, instead of installing program 20.

[0111] (effect) As described above, the physiological phenomenon estimation device 1 according to this embodiment includes: an information acquisition unit 10 that acquires detection data 2 detected over time by a sensor 30 (optical sensor or strain sensor) placed on the body surface, which reflects changes in blood flow caused according to the menstrual cycle or pregnancy state; an inference unit 11 that takes input data 5 including the detection data 2 or secondary data based on the detection data 2 as input and outputs information related to the menstrual cycle or pregnancy state to a trained model 21, which receives the input data 5 from the detection data 2 acquired by the information acquisition unit 10 to generate inference information 3 about the menstrual cycle or pregnancy state; and an output unit 13 that outputs the inference information 3 generated by the inference unit 11. As a result, inference information 3 about the menstrual cycle or pregnancy state can be generated and provided simply by acquiring detection data 2 detected over time by a sensor 30 (optical sensor or strain sensor) placed on the user's body surface, without using a sensor unit fixed to the abdomen with a belt or a large, dedicated device such as a fetal monitoring device. The sensor 30 (optical sensor or strain sensor) can be provided as a small, wearable device that does not interfere with body movements even when worn. Therefore, it is possible to alleviate spatial and physical constraints when understanding physiological phenomena such as the menstrual cycle and pregnancy.

[0112] In the physiological phenomenon estimation device 1 according to this embodiment, the detected data 2 includes time-series data of the electrical output form output by the sensor 30 (optical sensor or strain sensor), or image data (i.e., signal waveform data) showing the time-series change in the electrical output form. This makes it easy to obtain detected data 2 corresponding to a volume pulse wave reflecting changes in the user's blood flow from the time change of the electrical output of the sensor 30 (optical sensor or strain sensor). For example, by simply attaching small sensors to the limbs, head and neck (head, retina, eardrum, ear, etc.), and trunk of the user's US and detecting changes in peripheral blood flow, inference information 3 regarding the menstrual cycle or pregnancy state can be easily obtained.

[0113] In the physiological phenomenon estimation device 1 according to this embodiment, the inference information 3 regarding the menstrual cycle includes at least one of the following: physiological information related to the menstrual cycle, information regarding the relevant cycle segment of the menstrual cycle, and information regarding the timing of ovulation or menstruation onset. This allows the user US to be provided with information useful for taking measures to prevent pregnancy or contraception.

[0114] In the physiological phenomenon estimation device 1 according to this embodiment, the inference information 3B regarding the pregnancy state includes information on at least one of the following: maternal uterine contractions, fetal intrauterine movement, maternal amniotic fluid volume, cervical ripening, gestational age, estimated fetal weight, and timing of labor onset. This provides important information for monitoring the progress of pregnancy. Furthermore, since signs of premature birth or placental abruption can be identified early from the obtained information, it is expected to contribute to early medical intervention for important signs and improvement of treatment strategies for perinatal complications.

[0115] In the physiological phenomenon estimation device 1 according to this embodiment, the trained model 21 includes a first trained model 21A that outputs inference information 3 regarding the menstrual cycle, and a second trained model 21B that outputs inference information 3B regarding the pregnancy state. The inference unit 11 selects the first trained model 21A and the second trained model 21B according to the user US selection or information regarding the presence or absence of pregnancy. By separately creating (machine learning) the first trained model 21A for inference regarding the menstrual cycle and the second trained model 21B for inference regarding the pregnancy state, the estimation accuracy of each can be improved. Since the information to be provided to the user US differs depending on whether the user US is pregnant or not, selecting the first trained model 21A and the second trained model 21B makes it possible to provide appropriate information.

[0116] The physiological phenomenon estimation device 1 according to this embodiment further includes an evaluation unit 12 that generates evaluation information 4 corresponding to the content of the inference information 3 generated by the inference unit 11, and an output unit 13 outputs evaluation information 4 corresponding to the inference information 3 in addition to the inference information 3. This makes it possible to provide the user US with evaluation information 4 derived from the inference information 3 in addition to the inference information 3. This makes it possible to provide information such as information to assist a user US who does not have specialized knowledge, or advice related to the inference information 3.

[0117] In the physiological phenomenon estimation device 1 according to this embodiment, the information acquisition unit 10 acquires additional supplementary information 6 related to the menstrual cycle and pregnancy status, in addition to the detected data 2, and the input data 5 further includes the supplementary information 6 in addition to the detected data 2 or secondary data. As a result, the inference unit 11 can generate inference information 3 related to the menstrual cycle or pregnancy status based on the supplementary information 6 as well. Consequently, inference that takes into account the statistical bias of the input data 5 caused by differences in the supplementary information 6 becomes possible, thereby improving the inference accuracy.

[0118] In the physiological phenomenon estimation device 1 according to this embodiment, the information acquisition unit 10, the inference unit 11, and the output unit 13 are configured by a control unit 102 of a wearable device 100 that can be attached to a part of the body executing a program 20. The information acquisition unit 10 acquires detection data 2 from a sensor 30 (optical sensor or strain sensor) of the wearable device 100, and the output unit 13 outputs inference information 3 to a display unit 31 of the wearable device 100. Thus, the wearable device 100, which has sensing and display functions, can realize everything from acquiring detection data 2 to generating inference information 3 and outputting inference information 3. Since the wearable device 100 is a small device that can be attached to the body, it can effectively alleviate spatial and physical constraints when understanding physiological phenomena such as menstrual cycles and pregnancy.

[0119] The method for estimating physiological phenomena according to this embodiment includes: step S10, which involves acquiring detection data 2 that is detected over time by a sensor 30 (optical sensor or strain sensor) placed on the body surface, reflecting a change in blood flow caused by the menstrual cycle or pregnancy state; step S20, which involves inputting the input data 5, which is based on the acquired detection data 2, into a trained model 21 that takes the detection data 2 or secondary data based on the detection data 2 as input and outputs information about the menstrual cycle or pregnancy state, to generate inference information 3 about the menstrual cycle or pregnancy state; and step S40, which involves outputting the generated inference information 3. This makes it possible to generate and provide inference information 3 about the menstrual cycle or pregnancy state simply by acquiring detection data 2 detected over time by a sensor 30 (optical sensor or strain sensor) placed on the body surface of the user US, without using a sensor unit fixed to the abdomen with a belt or a large, dedicated device such as a fetal monitoring device. The sensor 30 (optical sensor or strain sensor) can be provided as a small wearable device 100 that does not interfere with body movements even when worn. Therefore, it is possible to alleviate spatial and physical constraints when understanding physiological phenomena such as the menstrual cycle and pregnancy.

[0120] Although embodiments of the present invention have been described above, the embodiments are not limited to those described herein. Furthermore, the aforementioned components include those that can be easily conceived by those skilled in the art, those that are substantially the same, and those that fall within the so-called equivalent range. Moreover, the aforementioned components can be combined as appropriate. Furthermore, various omissions, substitutions, or modifications of the components can be made without departing from the spirit of the embodiments described above.

[0121] For example, Figure 5 shows an example in which the evaluation unit 12 generates evaluation information 4 through rule-based information processing using evaluation data 22, but this is not the only example. Figure 12 is a conceptual diagram showing variations in the generation of evaluation information by the evaluation unit.

[0122] As shown in Figure 12, the evaluation unit 12 may generate evaluation information 4 by evaluation process VP1 using a pre-trained evaluation model 23 generated by machine learning, or by evaluation process VP2 using rule-based information processing, or by evaluation process VP3 which is a combination (usual use) of the pre-trained evaluation model 23 and rule-based information processing. The pre-trained evaluation model 23 used in evaluation process VP1 is a pre-trained model that has been trained by machine learning to take inference information 3 as input and output evaluation information 4. Examples of machine learning learning algorithms include supervised learning, unsupervised learning, and reinforcement learning, and learning algorithms similar to those described for the pre-trained model 21 are given as examples. Evaluation process VP2 is a process that generates evaluation information 4 by rule-based information processing using evaluation data 22, and is equivalent to the process shown in Figure 5. In evaluation process VP3, for example, the pre-trained evaluation model 23 may generate an intermediate evaluation corresponding to the inference information 3, and the final evaluation information 4 corresponding to the generated intermediate evaluation may be associated with the evaluation data 22. Alternatively, for example, the inference information 3 may be converted into evaluation values ​​for one or more items using evaluation data 22, and the evaluation information 4 may be generated by a pre-trained evaluation model 23 that takes the converted evaluation values ​​as input. In Figure 12, for the sake of explanation, evaluation process VP1, evaluation process VP2, and evaluation process VP3 are shown together, but the evaluation unit 12 only needs to be configured to perform any one of these processes.

[0123] Furthermore, in the above embodiment, Figure 4 shows an example of an inference unit 11 that performs inference based on input data 5 including supplementary information 6. However, the processing of input data 5 including supplementary information 6 is not limited to the example shown in Figure 4. Specifically, Figure 4 shows an example in which the trained model 21 is constructed as a model that corresponds to multi-dimensional input data 5 so that it can accept both detection data 2 (or its secondary data) and supplementary information 6 as input. However, for example, the trained model 21 may be fine-tuned according to the supplementary information 6.

[0124] Figure 13 is a schematic diagram illustrating the generation of individually trained models through fine-tuning. Figure 14 is a schematic diagram illustrating the inference process using individually trained models.

[0125] As shown in Figure 13, the trained model includes multiple individually trained models 321 that are fine-tuned using individual training data 341 classified based on the accompanying information 6, on a pre-trained pre-trained model 320. Individually trained models 321 are generated when the pre-trained model 320, which is pre-trained using the pre-trained training data 340, is fine-tuned using the individual training data 341. The trained model that has been fine-tuned using the individual training data 341 is called an individually trained model 321. In this case, the pre-trained model 320 and individually trained models 321 are not multi-dimensional models that accept input of detection data 2 and accompanying information 6, but rather trained models that have an input dimension corresponding to the detection data 2.

[0126] Based on the supplementary information 6, the training data used for machine learning is classified into one of several groups. Individual training data 341, which are groups of training data belonging to one of the groups in supplementary information 6, are created for each group.

[0127] A pre-trained model 320 is generated by machine learning (pre-training) of the learning model 40 using pre-trained data 340 that does not group data (it randomly includes data from all groups). Part or all of the generated pre-trained model 320 is then trained by fine-tuning it with individual training data 341. As a result, an individual trained model 321 is generated for each group of supplementary information 6. The number of data points in the individual training data 341 may be less than the number of data points in the pre-trained data 340.

[0128] The method for classifying (grouping) the individual training data 341 is not particularly limited. Here, we will describe an example where the accompanying information 6 is BMI. BMI can be classified into three groups, for example, high, median (values ​​between high and low), and low. The individual training data 341 consists of training data obtained from sample providers whose BMI belongs to one of each group. Then, by fine-tuning using the individual training data 341 for high BMI, individual training data 341 for median BMI, and individual training data 341 for low BMI, separate individual trained models 321 for high BMI, individual trained models 321 for median BMI, and individual trained models 321 for low BMI are created.

[0129] As shown in Figure 14, the inference unit 11 includes a selection unit 11A that selects an individually trained model 321 to input detection data 2 or secondary data based on detection data 2, based on the accompanying information 6 contained in the input data 5. The selection unit 11A selects an individually trained model 321 to input detection data 2 (or secondary data) based on the accompanying information 6 acquired by the information acquisition unit 10. The inference unit 11 generates inference information 3 for user US by inputting detection data 2 (or secondary data) to the selected individually trained model 321 and supplies it to the output unit 13.

[0130] This allows the system to select a pre-trained model 321 for inference based on whether the user's BMI falls into one of three groups: high, median, or low. As a result, the accuracy of inference can be improved according to the user's BMI. Fine-tuning based on the user's body size (BMI) is effective in improving inference accuracy because it can address statistical biases in the input data.

[0131] The number of groups into which the supplementary information 6 is classified is not particularly limited. BMI may also be classified into 2 or 4 or more groups. The same applies to the other supplementary information 6.

[0132] As mentioned above, it is not necessary to perform inference processing using the accompanying information 6. The input data 5 may not include the accompanying information 6 and may only include the detected data 2 or secondary data based on the detected data 2. In this case, the trained model 21 is machine-trained using training data that does not group (it randomly includes data from all groups), and is a trained model equivalent to the pre-trained model 320 shown in Figure 14. [Explanation of Symbols]

[0133] 1…Physiological phenomenon estimation device, 2…Detection data, 3,3A,3B…Inference information, 4,4A,4B…Evaluation information, 5…Input data, 6…Supplementary information, 10…Information acquisition unit, 11…Inference unit, 12…Evaluation unit, 13…Output unit, 14…Storage unit, 20…Program, 21…Trained model, 21A…First trained model, 21B…Second trained model, 22,22A,22B…Evaluation data, 30…Sensor, 31…Display unit 40...Learning model, 41A,41B...Training data, 42A,42B...Input training data, 43A,43B...Output training data, 100,110,120...Wearable device, 101...Housing, 102...Control unit, 103...Operation unit, 130,210...Information and communication terminal, 200...Server, 321...Individually trained model, NW...Network, US...User, FE...Fetus, VP1,VP2,VP3...Evaluation process.

Claims

1. An information acquisition unit acquires detection data that reflects changes in blood flow caused by the menstrual cycle or pregnancy status, detected over time by an optical sensor or strain sensor placed on the body surface. An inference unit takes input data including the detection data or secondary data based on the detection data as input, and outputs information about the menstrual cycle or pregnancy status as output to a trained model, and inputs the input data obtained by the information acquisition unit from the detection data to generate inference information about the menstrual cycle or pregnancy status. An output unit that outputs the inference information generated by the inference unit, A device for estimating physiological phenomena, equipped with the following features.

2. The detection data includes time-series data of the electrical output form output by a light sensor or strain sensor, or image data showing the time-series change of the electrical output form. The apparatus for estimating physiological phenomena according to claim 1.

3. The aforementioned inference information regarding the menstrual cycle is Information on physiological changes related to the menstrual cycle, Information regarding the relevant cycle segment of the menstrual cycle, Information regarding the timing of ovulation or menstruation, and at least one of the following: The apparatus for estimating physiological phenomena according to claim 1.

4. The aforementioned inference information regarding the pregnancy status is, This includes information on at least one of the following: maternal uterine contractions, fetal intrauterine movement, maternal amniotic fluid volume, cervical ripening, gestational age, estimated fetal weight, and timing of labor onset. The apparatus for estimating physiological phenomena according to claim 1.

5. The aforementioned trained model is A first trained model that outputs the aforementioned inference information regarding the menstrual cycle, A second trained model that outputs the aforementioned inference information regarding the pregnancy status, The inference unit selects the first trained model and the second trained model according to the user's selection or information regarding the presence or absence of pregnancy. The apparatus for estimating physiological phenomena according to claim 1.

6. The system further includes an evaluation unit that generates evaluation information corresponding to the content of the inference information generated by the inference unit, The output unit outputs evaluation information corresponding to the inference information, in addition to the inference information. The apparatus for estimating physiological phenomena according to claim 1.

7. The information acquisition unit acquires additional information related to the menstrual cycle and pregnancy status, separate from the detected data. The input data further includes the associated information in addition to the detection data or the secondary data. The apparatus for estimating physiological phenomena according to claim 1.

8. The information acquisition unit, the inference unit, and the output unit are configured by a control unit provided in a wearable device that can be attached to a part of the body executing a program. The information acquisition unit acquires the detection data from the optical sensor or strain sensor provided by the wearable device. The output unit outputs the inference information to the display unit of the wearable device. The apparatus for estimating physiological phenomena according to claim 1.

9. The steps include acquiring detection data obtained by detecting signals that reflect changes in blood flow caused by the menstrual cycle or pregnancy status over time using an optical sensor or strain sensor placed on the body surface, The steps include: inputting the input data obtained from the detection data into a trained model that takes the detection data or secondary data based on the detection data as input and outputs information about the menstrual cycle or pregnancy status, thereby generating inference information about the menstrual cycle or pregnancy status; The steps include outputting the generated inference information, A method for estimating physiological phenomena, comprising the following features.

Images