Menstrual cycle estimation device and menstrual cycle estimation method

In the present invention, for example, whether the subject is in a high temperature period or a low temperature period based on the body temperature of the subject (including the body surface temperature and the deep temperature, the same applies hereinafter) measured over a plurality of days. The present invention relates to a menstrual cycle estimation device and a menstrual cycle estimation method.

Conventionally, in order to grasp the hormone balance status of women for the purpose of knowing the cycle that is a guideline for pregnancy and contraception, daily fluctuations in basal body temperature are tracked and recorded over several months, and the fluctuation cycle and fluctuation range are numerically measured. The method of analysis has been carried out. It is desirable to measure the basal body temperature during bedtime, which has the lowest metabolism in the day, but it is generally difficult to measure the body temperature during bedtime. Therefore, for example, the mouth temperature at the time of waking up is measured as the body temperature, and the measured body temperature has been used instead of the basal body temperature (see, for example, Patent Document 1). However, considering the diversification of women's lives, it is difficult to keep the conditions such as the measurement time and measurement method constant every morning, and it is inconvenient to continue the measurement for a long period of time. In the measurement, artifacts (distortion of the measurement signal due to external factors) were likely to occur in the measurement results. For this reason, a sensor device that measures the temperature of the body surface (body surface temperature) is attached to the abdomen of the person to be measured who is sleeping, and the basis is based on the body surface temperature automatically measured over a long period of time by this sensor device. It is known that the representative temperature of the day corresponding to the body temperature is determined (see, for example, Patent Document 2). At this time, the representative temperature was, for example, the maximum temperature or the average temperature of the measured body surface temperature.

<patcit num="1"><text>Japanese Unexamined Patent Publication No. 11-316161</text></patcit><patcit num="2"><text>Japanese Unexamined Patent Publication No. 2004-163391</text></patcit>

<p> By the way, in the prior art, the state of close contact of the sensor device with respect to the body surface may change due to turning over of the person to be measured, and the measurement result may vary. For example, when the sensor device is attached to the abdomen of the person to be measured and the sensor device is laid on its back, the degree of adhesion of the sensor device is low, whereas when the sensor device is laid down, the degree of adhesion of the sensor device is high. Furthermore, when lying down, there is a possibility that the body surface temperature rises due to a temporary increase in blood flow due to abdominal stimulation, or there is a possibility that the body surface temperature drops due to a decrease in blood flow due to abdominal compression. In addition, there is a possibility that the body surface temperature will rise due to the use of electric blankets. As a result, the measurement result of the body surface temperature includes artifacts, and the measurement accuracy of the body surface temperature and the representative temperature tends to decrease.</p><p> Further, in the prior art, the representative temperature is determined within the range of the measured body surface temperature. At this time, even with the prior art, it is possible to grasp the fluctuation of the basal body temperature by using the representative temperature to some extent. However, in general, the body surface temperature is easily affected by the outside air, blood flow, etc., so if the maximum temperature or average temperature of the measured value is determined as the representative temperature, the representative temperature may have artifacts due to the influence of the outside air, etc. There is. Here, in grasping the biphasic nature of the basal body temperature consisting of the high temperature period and the low temperature period, it is necessary to accurately grasp the increase of + 0.3 ° C from the low temperature period to the high temperature period. On the other hand, in the prior art, there is a problem that it is difficult to distinguish between an increase in body temperature in a high temperature period and a change due to measurement artifacts.</p><p> As described above, since the representative temperature is affected by the measurement artifact, it is difficult to determine whether the subject is in the high temperature period or the low temperature period based on the representative temperature. As a result, the biphasic basal body temperature and the cycle period tended to be unclear, and it was not possible to easily grasp the hormone balance status of women.</p><p> The present invention has been made in view of the above-mentioned problems of the prior art, and an object of the present invention is menstruation in which the biphasic nature of the body temperature of the subject can be clearly grasped based on the measurement data of the body temperature over a plurality of days. It is an object of the present invention to provide a cycle estimation device and a menstrual cycle estimation method.</p>

<p> In order to solve the above-mentioned problems, the invention of claim 1 states that the subject is in either a high temperature period or a low temperature period by using a body temperature data series in which the body temperature of the subject is measured over a plurality of days. It is a menstrual cycle estimator for estimating the existence, and a hidden Markov model having two hidden states of a high temperature phase and a low temperature phase is constructed using the body temperature data series, and the hidden Markov model uses the body temperature data series as the body temperature data series. The most probable state series to be output is obtained, and the state series is used to specify the biphasic nature of the body temperature in the menstrual cycle of the subject.</p><p> In the invention of claim 2, a casing that can be always worn on the body of the person to be measured, a body temperature detecting means that is provided on the body surface side of the person to be measured and detects the body temperature, and an end time from the start time for each measurement day. A reading means for reading the body temperature detection data by the body temperature detecting means, a storage means for storing the body temperature detection data by the reading means, and a storage means stored in the storage means at predetermined time intervals up to the time. A data series generating means for generating the body temperature data series by concatenating the body temperature detection data over a plurality of days, a model building means for constructing the hidden Markov model using the body temperature data series by the data series generating means, and the said. The state series calculation means for obtaining the most probable state series from which the hidden Markov model by the model construction means outputs the body temperature data series, and the body temperature in the menstrual cycle of the subject using the state series by the state series calculation means. The configuration is provided with a state specifying means for specifying compatibility.</p><p> In the invention of claim 3, when the state specifying means performs a filter process using a median filter on the state series and measures each body temperature detection data for the state included in the state series after the filter processing. It is configured to be the state of the person to be measured.</p><p> In the invention of claim 4, the data sequence generating means deletes a value other than the preset temperature and the allowable range of temperature change from the body temperature detection data, and concatenates the remaining body temperature detection data to form a body temperature data sequence. The model building means includes a means for deleting out-of-range data for generating data, and a pre-processing means for performing filtering using a median filter on the body temperature data series generated by using the means for deleting out-of-range data. The hidden Markov model is constructed using the body temperature data series after filtering by the pretreatment means.</p><p> In the invention of claim 5, the model building means constructs the hidden Markov model in a state where the body temperature detection data of the body temperature data series has discrete values with respect to the temperature.</p><p> In the invention of claim 6, the model building means constructs the hidden Markov model in a state where the body temperature detection data of the body temperature data series has a continuous value with respect to the temperature.</p><p> In the invention of claim 7, the model building means is configured by an external processing device, and the casing is provided with a transfer means for transferring the body temperature detection data stored in the storage means to the external processing device. ..</p><p> In the invention of claim 8, the menstrual cycle estimation for estimating whether the subject is in a high temperature period or a low temperature period by using a body temperature data series in which the body temperature of the person to be measured is measured over a plurality of days. In the method, a hidden Markov model having two states of a high temperature phase and a low temperature phase is constructed using the body temperature data series, and the most probable state series in which the hidden Markov model outputs the body temperature data series is obtained. , The configuration is such that the biphasic nature of the body temperature in the menstrual cycle of the subject is specified using the state series.</p><p> In the invention of claim 9, the menstrual cycle for estimating which of the polyphasic states of the body temperature in the menstrual cycle the subject is in, using the body temperature data series obtained by measuring the body temperature of the subject over a plurality of days. This is an estimation method, in which a hidden Markov model having a plurality of hidden states corresponding to polyphasic body temperature is constructed using the body temperature data series, and the most probable state in which the hidden Markov model outputs the body temperature data series. The sequence is obtained, and the state sequence is used to specify the polyphasic nature of the body temperature in the menstrual cycle of the subject.</p>

<p> According to the inventions of claims 1 and 8, a hidden Markov model having two states of a high temperature phase and a low temperature phase is constructed using the body temperature data series, and the hidden Markov model outputs the body temperature data series with maximum likelihood. Since the state series is obtained, the state of the subject to be measured over a plurality of days for which the body temperature data series is measured can be specified by determining the state included in this state series. As a result, even when a measurement artifact occurs in the body temperature data series due to the influence of outside air or the like, the biphasic nature of the subject can be clearly grasped.</p><p> According to the invention of claim 2, since the body temperature detecting means is provided on the body surface side of the person to be measured in the casing, the reading means is said to have a predetermined time interval from the start time to the end time. The body temperature detection data is read by using the body temperature detecting means, and the storage means stores the body temperature detection data by the reading means. Then, since the data series generation means generates the body temperature data series by concatenating the body temperature detection data stored in the storage means for a plurality of days, the model construction means uses the body temperature data series by the data series generation means. The model parameters are calculated and a hidden Markov model consisting of the model parameters is constructed. As a result, the state sequence calculation means can obtain the maximum likelihood state sequence for which the hidden Markov model formed by the model construction means outputs the body temperature data series. Therefore, the state specifying means can specify the state of the person to be measured when each body temperature detection data is measured by using the state series by the state series calculation means.</p><p> According to the invention of claim 3, since the state specifying means performs the processing using the median filter on the state series, it is possible to remove the discontinuous out-of-state state included in the state series. As a result, the state does not change repeatedly every few hours or days, and the biphasic nature of the subject who repeats the high temperature period and the low temperature period with a periodicity of about one month can be clearly grasped.</p><p> According to the invention of claim 4, the out-of-range data deleting means deletes the body temperature detection data other than the preset temperature and the allowable range of temperature change as abnormal values, and comprises a value within the allowable range. The remaining body temperature detection data is concatenated to generate a body temperature data series. Generally, the body temperature measurement range of the human body is about 32 ° C to 40 ° C, so the temperature detected by the body temperature detecting means in the thermal equilibrium state does not deviate from the permissible range of, for example, 32 ° C to 40 ° C. it is conceivable that. When the body temperature detection data is a value outside this temperature range, it is conceivable that, for example, the person to be measured removes the casing from the body or receives external heat.</p><p> In addition, since the body temperature of the subject is relatively stable during bedtime, it is considered that the temperature does not change beyond the allowable range of ± 1 ° C at a measurement interval of, for example, about 10 minutes. When the body temperature detection data becomes a value other than the permissible range of this temperature change, for example, when the person to be measured temporarily wakes up, or when the person to be measured bounces off the bedding, etc. can be considered.</p><p> As described above, the body temperature detection data outside the permissible range of temperature or the permissible range of temperature change is considered to have a low contribution rate to the condition of the subject. In addition, if a hidden Markov model is constructed using body temperature detection data outside the permissible range of temperature and temperature change, the accuracy of state estimation may decrease. Therefore, the out-of-range data deleting means deletes the values other than the permissible range from the body temperature detection data and stores the body temperature data series.</p><p> Further, since the preprocessing means performs filtering processing using the median filter on the body temperature data series stored by the out-of-range data deleting means, it is possible to remove discontinuous outliers included in the body temperature data series. it can. Then, since the model building means builds a hidden Markov model using the body temperature data series after filtering by the preprocessing means, the hidden Markov model is compared with the case of using the body temperature data series including outliers. It is possible to prevent outliers from affecting the model parameters of the subject, and it is possible to improve the estimation accuracy of the state of the person to be measured.</p><p> According to the invention of claim 5, since the model building means constructs a hidden Markov model in a state where the body temperature detection data of the body temperature data series has a discrete value with respect to the temperature, the hidden Markov model is a discrete type. It has a body temperature detection data output probability consisting of a discrete probability distribution according to the body temperature detection data. Therefore, the state series can be obtained by using the body temperature data series consisting of the body temperature detection data discrete with respect to the temperature and the discrete hidden Markov model. Further, since the discrete hidden Markov model is used, the time required for the model construction process, the storage capacity, and the like can be reduced as compared with the case where the continuous hidden Markov model is used.</p><p> According to the invention of claim 6, since the model building means constructs a hidden Markov model in a state where the body temperature detection data of the body temperature data series has a continuous value with respect to the temperature, the hidden Markov model is a continuous type. It has a body temperature detection data output probability consisting of a continuous probability distribution according to the body temperature detection data. Therefore, the state series can be obtained by using the body temperature data series consisting of the body temperature detection data continuous with respect to the temperature and the continuous hidden Markov model. Further, since the continuous hidden Markov model is used, the accuracy of the state sequence can be improved as compared with the case where the discrete hidden Markov model is used.</p><p> According to the invention of claim 7, the model building means is configured by an external processing device, and the casing is provided with a transfer means for transferring the data stored in the storage means to the external processing device. Here, in order to estimate the state of the person to be measured using the body temperature data series measured over a plurality of days and to grasp the fluctuation cycle of the basal body temperature of the woman (the person to be measured), the body temperature for one cycle or more is required. It is preferable to build a hidden Markov model using the detected data. In this case, the hidden Markov model is better when an external processing device is used than when a microcomputer or mobile phone that forms a control circuit provided in a small casing to be inserted into the subject is used. Is efficient in building.</p><p> Therefore, in the present invention, the body temperature detection data stored in the storage means is transferred to an external processing device such as a server computer by using a transfer means such as a wired method or a wireless method. Therefore, the processing device can calculate and process a large amount of body temperature detection data and quickly construct a hidden Markov model.</p><p> According to the invention of claim 9, the body temperature data series is used to construct a hidden Markov model having a plurality of hidden states corresponding to the polyphasic body temperature, and the hidden Markov model outputs the body temperature data series. Since the state series is obtained, by determining the states included in this state series, it is possible to identify the state having polymorphism of the subject to be measured over a plurality of days when the body temperature data series is measured. As a result, even when a measurement artifact occurs in the body temperature data series due to the influence of outside air or the like, the polymorphism of the person to be measured can be clearly grasped. Further, for example, when a medium temperature phase is provided between the low temperature phase and the high temperature phase, the rising change point from the low temperature phase to the high temperature phase and the falling change point from the high temperature phase to the low temperature phase can be investigated in detail.</p>

Hereinafter, the menstrual cycle estimation device according to the embodiment of the present invention will be described in detail with reference to the accompanying drawings.

First, FIGS. 1 to 10 show the first embodiment, and in the figure, 1 is a wearable sensor device (hereinafter referred to as sensor device 1) that constitutes a menstrual cycle estimation device together with an external processing device 11 described later. The sensor device 1 is roughly composed of a casing 2, a body temperature detection unit 3, a control unit 5, a display unit 9, and the like, which will be described later.

Reference numeral 2 denotes a casing that forms the main body (main support) of the sensor device 1. The casing 2 is formed in a substantially oval (oval) box shape using, for example, a resin material as shown in FIGS. 1 to 4. , The circuit board 4 and the like, which will be described later, are housed inside. Further, in the casing 2, an opening 2A penetrating in a substantially circular shape is provided on the body surface side (front side) of the person to be measured H0. Here, the opening 2A is located on the center side of the casing 2.

Reference numeral 3 denotes a body temperature detecting unit (body temperature detecting means) for detecting the body temperature, and the body temperature detecting unit 3 is attached to an opening 2A located on the central side of the casing 2 as shown in FIG. Further, the body temperature detection unit 3 is composed of a temperature measuring element made of, for example, a thermistor, and a metal cover that covers the temperature measuring element. Since the body temperature detection unit 3 is arranged on the body surface side of the person to be measured H0 in the casing 2, the cover comes into contact with the body surface of the person to be measured H0, and the body temperature is electrically conducted to the temperature measuring element through the cover. As a result, the body temperature detection unit 3 outputs a signal corresponding to the body temperature of the person to be measured H0 by the temperature measuring element.

The body temperature detection unit 3 is configured to directly detect the body temperature by contacting the body surface of the person to be measured H0. However, the present invention is not limited to this. For example, when an underwear or the like is sandwiched between the body temperature detection unit 3 and the body surface of the person to be measured H0, the body temperature detection unit 3 uses the underwear or the like to pass the body temperature detection unit 3 to the person to be measured H0. It may be configured to indirectly detect the body temperature of the body. That is, the body temperature detection unit 3 does not need to detect the absolute body temperature of the subject H0, but only needs to detect the relative temperature that changes in response to the body temperature. Therefore, if the daily measurement conditions are substantially constant, the body temperature detection unit 3 may be brought into close contact with the body surface of the person to be measured H0, for example, by sandwiching one piece of underwear. Further, the body temperature detecting unit 3 is not limited to the one that detects the temperature of the body surface of the person to be measured H0, and may be configured to use, for example, a deep thermometer that detects the temperature inside the body of the person to be measured H0.

Reference numeral 4 denotes a circuit board housed inside the casing 2. As shown in FIG. 5, the circuit board 4 is equipped with a control unit 5 as a reading means including a microcomputer or the like. The input side of the control unit 5 is connected to the body temperature detection unit 3, and the output side of the control unit 5 is connected to the display unit 9, which will be described later. Further, the control unit 5 is provided with a storage unit 6 including, for example, ROM, RAM, etc. as storage means.

Here, the storage unit 6 stores in advance a program for operating the control unit 5, a start time ts, an end time te, and a time interval Δt used in the program, and the body temperature T1 described later by the operation of the control unit 5. Is memorized.

At this time, the start time ts and the end time te are set as sleeping times, for example, midnight (ts = 0:00 am) and 6:00 am (te = 6:00 am), respectively, and the time interval Δt is, for example, 10. It is set to a value of about minutes. The start time ts and end time te are set to the start time and end time of bedtime if the person to be measured H0 is, for example, a night worker. Further, the time interval Δt is not limited to 10 minutes, but is appropriately set to, for example, about 1 to 30 minutes according to the measurement conditions and the like.

Further, the control unit 5 has a timer 7 for measuring the time, and for example, a switch unit 8 composed of two button switches 8A and 8B is connected to the control unit 5. Then, the control unit 5 is driven by a power source 5A such as a coin-type lithium battery mounted on the casing 2, and operates by reading a program from the storage unit 6 by operating the switch unit 8. As a result, when the time set by the timer 7 reaches the start time ts, the control unit 5 responds to the detection temperature from the body temperature detection unit 3 at regular time intervals Δt from the start time ts to the end time te. Read the body temperature detection data D1.

Then, the control unit 5 sequentially stores the body temperature detection data D1 in the storage unit 6. As a result, a plurality of body temperature detection data D1s from the start time ts to the end time te are stored in the storage unit 6.

Further, the storage unit 6 has a value within a preset allowable temperature range (for example, 32 ° C to 40 ° C) of the body temperature detection data D1 by the control unit 5, and has a temperature change allowable range (for example, ± 1). The body temperature detection data D1 is stored with a value within ° C / 10 minutes) as a normal value and a value outside the permissible range as an abnormal value. Specifically, the normal value stores the body temperature detection data D1 as it is, and the abnormal value stores the missing data (for example, an error flag that can be distinguished from the normal value).

Here, since the body temperature measurement range of the human body is generally about 32 ° C to 40 ° C, the allowable temperature range is, for example, 32 ° C to 40 ° C when the temperature detected by the body temperature detection unit 3 in the thermal equilibrium state. It is decided based on the idea that it does not deviate from the range of. In addition, the body temperature of the subject is relatively stable during bedtime. Therefore, the permissible range of temperature change is determined based on the idea that the permissible range of temperature change does not exceed the permissible range of ± 1 ° C before and after the time interval Δt of, for example, about 10 minutes.

The allowable range of temperature and the allowable range of temperature change are not limited to the illustrated values, and may be appropriately set in consideration of measurement conditions and the like.

Further, the body temperature detection data D1 may be the detected body temperature T1 itself, or may be, for example, a difference value between a preset reference temperature value T0 and the body temperature T1. In this case, the reference temperature value T0 is preferably set to, for example, the average value of the body temperature T1 (for example, 34 ° C) from the viewpoint of reducing the data capacity.

Reference numeral 9 denotes a display unit including a liquid crystal screen or the like provided on the back side of the casing 2. The display unit 9 is connected to the control unit 5 and displays, for example, the driving state of the control unit 5. Further, the display unit 9 constitutes a transfer means for transferring the body temperature detection data D1 stored in the storage unit 6 to an external processing device 11 described later, and by operating the switch unit 8, for example, a QR code (registered trademark). Display a two-dimensional code such as. Here, this two-dimensional code includes information on the destination address such as the body temperature detection data D1 for one day (one night) stored in the storage unit 6 and the URL (Uniform Resource Locator) of the data. Therefore, the person to be measured H0 reads the information in the two-dimensional code into the mobile phone PT by using a mobile phone PT or the like having a function of reading the two-dimensional code. As a result, as shown in FIG. 6, the person to be measured H0 can access the mobile phone PT to the external processing device 11 via the Internet or the like and transfer the body temperature detection data D1 to the processing device 11. ..

Further, the person to be measured H0 may be configured to input the menstruation start date data by operating the switch unit 8 of the sensor device 1. In this case, the person to be measured H0 transfers the menstruation start date data to the processing device 11 by performing the two-dimensional code display, reading, and transmitting operations in the same manner as described above using the sensor device 1 and the mobile phone PT. Can be done.

The transfer means is configured to use a two-dimensional code display by the display unit 9. However, the present invention is not limited to this, and a connection portion such as a connector for connecting the control unit 5 and the mobile phone PT by using, for example, a wired method by a cable connection or a wireless method by infrared rays or Bluetooth is provided, and the connection portion is provided. The transfer means may be configured by.

Reference numeral 10 denotes a clip as a mounting means for attaching the casing 2 to the person to be measured H0. The clip 10 is located on the center side of the casing 2 and has a flexible strip shape and extends toward the upper part (outside). Its tip is configured to be able to hold clothes and the like. Therefore, as shown in FIG. 1, the clip 10 fixes the sensor device 1 to the abdomen of the subject H0 by sandwiching underwear such as shorts.

The sensor device 1 is not limited to the abdomen of the person to be measured H0, and may be fixed to the chest of the person to be measured H0, for example. In this case, the clip 10 is attached to underwear or the like worn at bedtime.

As shown in FIG. 6, reference numeral 11 denotes a processing device that estimates whether the subject H0 is in a high temperature period or a low temperature period using the body temperature detection data D1 measured by the sensor device 1. 11 constitutes a data series generation means, a model construction means, a state series calculation means, and a state identification means. Further, the processing device 11 is composed of, for example, a server computer or the like, and has a storage circuit including ROM, RAM, or the like.

Here, as shown in FIG. 7, the processing apparatus 11 has a low temperature phase q.₁And high temperature phase q₂Two states Q = {q₁, q₂It is equipped with a hidden Markov model HMM (hereinafter referred to as model HMM) with. At this time, the model HMM has a state transition probability matrix A = {a._ij} (Probability of transition from state i to state j), Body temperature detection data output probability matrix B = {b_j(k)} (b_j(k) is the probability that the observed data k will be in the state j) and the initial state probability matrix π = {π_i} (Probability of state i in the initial state) has model parameter λ = {A, B, π}.

Then, as shown in FIG. 8, the processing device 11 performs the menstrual cycle estimation process using the body temperature detection data D1 and the model HMM, and receives the body temperature detection data D1 for a plurality of days (for example, 6 months) measured. Estimate whether the measurer H0 is in the high temperature period or the low temperature period.

The outline of the specific procedure for menstrual cycle estimation processing is as follows. First, the processing device 11 accumulates the body temperature detection data D1 of the subject H0 transferred from the sensor device 1 over a plurality of days (for example, 6 months), and connects these body temperature detection data D1 to the body temperature data. Series O = {O₁, O₂, ..., O_T} Is generated. Next, the processing device 11 constructs a model HMM using this body temperature data series O.

Here, when constructing the model HMM, the model parameter λ = {A, B, π} of the model HMM is calculated using the body temperature data series O. At this time, the model parameter λ can be obtained by using, for example, the Baum-Welch algorithm, as will be described later. As a result, the processing device 11 can construct a model HMM having the maximum likelihood of the body temperature data series O.

In addition, the model HMM has body temperature detection data O having discrete values with respect to temperature as described later._tIs constructed using. Therefore, the body temperature detection data output probability matrix B of the model HMM is the discrete body temperature detection data O._tIt is composed of discrete probability distributions according to.

Next, the processing device 11 uses the model HMM constructed using the body temperature data series O, and may output the maximum likelihood path for the model HMM to output the body temperature data series O, that is, the body temperature data series O. Series of state changes with the highest Q^*To ask. At this time, the maximum likelihood state series Q^*Can be obtained, for example, by using the forward-backward algorithm described later.

Finally, the processing device 11 has the maximum likelihood state sequence Q.^*State contained in q_t^*(State at time t) is determined, and the body temperature detection data O included in the body temperature data series O_tThe state of the subject H0 at the time of measurement is the low temperature phase q₁And high temperature phase q₂Identify which of the above applies. As a result, the processing device 11 can estimate the state of the person to be measured H0 from the first day to the end date of the measurement date.

The processing device 11 is provided with a home page capable of displaying the estimation result of the state of the person to be measured H0. Therefore, when the person to be measured H0 accesses the homepage of the processing device 11 using, for example, the mobile phone PT, the estimation result of the state is displayed on the mobile phone PT and the computer in response to the request of the person to be measured H0. As a result, the estimation result of the state of the person to be measured H0 for 6 months is displayed on the screen of the mobile phone PT or the like.

The person to be measured H0 is not limited to the mobile phone PT, and may be configured to access the processing device 11 using various mobile terminals, computers, or the like. Further, the processing device 11 may be configured to transmit the state estimation result to the e-mail address registered in advance.

The menstrual cycle estimation device according to the present embodiment has the above-described configuration, and the menstrual cycle estimation process will be described next with reference to FIG.

Here, the subject H0 transfers the daily body temperature detection data D1 from the sensor device 1 to the processing device 11, so that the processing device 11 stores the body temperature detection data D1 for a plurality of days (6 months) in advance. It is assumed that it has been done. Then, the processing device 11 starts the menstrual cycle estimation process described later shown in FIG. 7 by receiving the request command or the like from the person to be measured H0.

First, in step 1, the missing data out of the accumulated body temperature detection data D1 for 6 months is deleted. At this time, the sensor device 1 reads the body temperature detection data D1 at 10-minute intervals for 6 hours between the start time ts and the end time te, and stores the body temperature detection data D1 in the storage unit 6. Therefore, the number of data of the body temperature detection data D1 for one day is 37, and the number of data of the body temperature detection data D1 for 6 months is about 6600. However, in reality, about 20 to 50% of missing data is generated, so the number of body temperature detection data D1 remaining after deleting the missing data is about 3300 to 5200. Therefore, the length T of the body temperature data series O is also about 3300 to 5200, corresponding to the number of data after the deletion of the missing data.

Next, in step 2, the body temperature detection data D1 for the remaining 6 months in step 1 is concatenated to generate the body temperature data series O. At this time, the body temperature data series O is configured by connecting the body temperature detection data D1 on the ts side of the start time on the first day of measurement to the body temperature detection data D1 on the te side of the end time on the end date of measurement. Therefore, the body temperature detection data O included in the body temperature data series O_t(1 t T) is arranged in chronological order from the start time ts side of the first day of measurement to the end time te side of the end date of measurement.

Next, in step 3, a filter using a median filter for the body temperature data series O as a preprocessing is performed before performing the main processing consisting of the hidden Markov model construction processing and the state series calculation processing described later. Perform processing. At this time, the median filter substitutes, for example, 11 consecutive data into the data at the center of the time series (the sixth data arranged in the time series). This makes it possible to remove obvious outliers contained in the body temperature data series O.

Next, in step 4, the body temperature detection data O_tIs discretized. In this discretization process, for example, by using the rounding method with 0.1 ° C accuracy, the vector quantization method, or the K-means method, the body temperature detection data O included in the body temperature data series O_tData clustering. As a result, the body temperature detection data O_tIs a numerical value expressed at 0.1 ° C intervals between 32 ° C and 40 ° C, for example.

Next, in step 5, the hidden Markov model shown in FIG. 9 to be described later is constructed, and the model HMM is constructed using the body temperature data series O after the pretreatment and the discretization treatment. Specifically, the optimum model parameter λopt (λopt = argmax [P (O | λ)]) of the model HMM is obtained so that the likelihood of the body temperature data series O is the highest.

Next, in step 6, state sequence arithmetic processing using, for example, a forward-backward algorithm is performed. Specifically, the maximum likelihood state sequence Q in which the model HMM constructed in step 5 outputs the body temperature data sequence O based on the equation of Equation 1 shown below.^*State contained in q_t^*Is calculated. At this time, the state series Q^*Is the path in which the model HMM is most likely to output the body temperature data series O. In addition, in the formula of Equation 1, γ_t(i) indicates the probability of being in the state i at time t, and is calculated based on the equation of equation 4 described later.

<maths num="1"><img file="JP2008264352A_D0001.tif" /></maths>

Next, in step 7, the state identification process is performed to estimate the state of the person to be measured H0 from the first day to the end date of the measurement date. Specifically, first, the maximum likelihood state series Q^*Is again filtered using the median filter. At this time, the median filter assigns these median values to the middle (sixth from the front in chronological order) states for, for example, 11 consecutive states. That is, the median filter is the low temperature phase q contained in 11 consecutive states.₁And high temperature phase q₂Substitute the state that occupies the majority of them into the middle state.

Then, the maximum likelihood state series Q output after filtering.^*Body temperature detection data O included in the body temperature data series O_tState q of the person to be measured when_t^*Is the low temperature phase q₁And high temperature phase q₂Identify which of the above applies. Specifically, the low temperature phase q on the same day₁And high temperature phase q₂Count the number of, and use the majority rule or other probabilistic method to determine the low temperature phase q₁Or high temperature phase q₂Decide which one is in. In this way, the state of the subject H0 from the first day to the end date of the measurement date is estimated, and the biphasic body temperature in the menstrual cycle of the subject H0 is specified.

Finally, in step 8, the estimation result of the state of the subject H0 from the first day to the end date of the measurement date specified in step 7 is displayed on the mobile phone PT and the computer of the subject H0.

Next, the construction process of the hidden Markov model will be described with reference to FIG. In this embodiment, a case where the model parameter λ of the model HMM is calculated by using the learning algorithm of Balm Welch will be described as an example.

In step 11, various coefficients required to build the model HMM are set. Specifically, the number of hidden states N included in the model HMM and the number of possible measured values M (body temperature detection data O)_t(Discretized number of), the length T of the body temperature data series O, and the body temperature data series O are read respectively.

At this time, the number of hidden states N, for example, the model HMM shown in FIG. 7 is the low temperature phase q.₁And high temperature phase q₂Because it contains, there are two states. In addition, the number M of possible measured values is 81 when, for example, 32 to 40 ° C is expressed with an accuracy of 0.1 ° C. In addition, the length T of the body temperature data series O is, for example, the body temperature detection data O for 6 months after deleting the missing data._tWhen it is, it will be about 3300 to 5200.

Next, in step 12, the initial value is assigned to the model parameter λ. Here, the state transition probability a_ijAs the initial value of, substitute a value of about 1 / N on the assumption that the probabilities of transitioning to all states are almost equal. However, all state transition probabilities a_ijIf the values are the same, it is difficult for the learning of the parameter λ to end (converge), so a slightly different value is substituted.

In addition, body temperature detection data output probability b_jThe initial value of (k) has almost the same probability of outputting all possible discrete body temperature values in each state, that is, the body temperature detection data O in each state._tAssuming that the value of 32 to 40 ° C is output evenly, substitute a value of about 1 / M. Furthermore, the initial state probability π_iAs the initial value of, substitute a value of about 1 / N on the assumption that the probabilities of being in all states are almost equal.

Next, in step 13, the forward variable α is based on the following equation (2)._tCalculate (i). At this time, the forward variable α_t(i) is the Partially Observatory {O from time 1 to time t₁, O₂, ..., O_t} After observing, the probability of being in the state i is shown at time t.

<maths num="2"><img file="JP2008264352A_D0002.tif" /></maths>

Next, in step 14, the backward variable β is based on the following equation (3)._tCalculate (i). At this time, the backward variable β_t(i) is the partial observation sequence {O from time t + 1 to the end in state i at time t._{t + 1}, O_{t + 2}, ..., O_T} Shows the probability of observing.

<maths num="3"><img file="JP2008264352A_D0003.tif" /></maths>

Next, in step 15, the probability γ of being in the state i at time t is based on the following equation of equation 4._tCalculate (i).

<maths num="4"><img file="JP2008264352A_D0004.tif" /></maths>

Next, in step 16, the probability γ_tUpdate parameter λ using (i). The specific explanation is as follows. First, the probability of being in state i at time t and being in state j at time t + 1 is ξ_tIf (i, j), this probability ξ_t(i, j) is expressed by the following equation of equation 5.

<maths num="5"><img file="JP2008264352A_D0005.tif" /></maths>

Here, the forward variable α_t(i) and backward variable β_tFrom the definition of (i), the probability γ_t(i) and probability ξ_tThere is a relationship with (i, j) shown in the following equation (6).

<maths num="6"><img file="JP2008264352A_D0006.tif" /></maths>

From the above, when the body temperature detection data D1 is discretized, the initial state probability π after the re-estimation calculation._i, State transition probability a_ijAnd body temperature detection data output probability b_j(k) can be expressed by the following equations 7, 8, and 9, respectively.

<maths num="7"><img file="JP2008264352A_D0007.tif" /></maths>

<maths num="8"><img file="JP2008264352A_D0008.tif" /></maths>

<maths num="9"><img file="JP2008264352A_D0009.tif" /></maths>

In Balm Welch's learning algorithm, the probability Pr (O | λ) of outputting the data series O by the model parameter λ after re-estimation by re-estimating outputs the data series O by the model parameter λ before re-estimation. It is guaranteed to be higher than the probability Pr (O | λ). Therefore, by repeating the above steps 13 to 16, the model parameter λ based on the body temperature data series O can be obtained.

When step 16 is completed, the process proceeds to step 17. Then, in step 17, in order to determine whether or not the model parameter λ has converged, the logarithmic value η of the probability Pr (O | λ) by the model parameter λ after re-estimation is based on the following equation of equation tens. Is calculated. The logarithmic operation is performed because the value of the probability Pr (O | λ) tends to be too small, and it is easier to determine the convergence by using the logarithm.

<maths num="10"><img file="JP2008264352A_D0010.tif" /></maths>

Next, in step 18, it is determined whether or not the difference Δη of the logarithmic value η before and after the re-estimation is larger than a predetermined convergence test value δ. Then, when it is determined as "YES" in step 18, the difference between the probabilities Pr (O | λ) before and after re-estimation is large, and the parameter λ has not converged, so the process from step 13 is repeated.

On the other hand, when it is determined as "NO" in step 18, it is considered that the difference between the probabilities Pr (O | λ) before and after re-estimation is small and the parameter λ has converged. Therefore, the process proceeds to step 19, the calculation result of the re-estimation at this time is output as the optimum model parameter λopt, and the result is returned in step 20.

In the present embodiment, the above-mentioned menstrual cycle estimation process and hidden Markov model construction process are performed, and next, the estimation result of the state of the subject H0 when these processes are used will be described.

Here, when the body temperature detection data D1 is measured for a specific subject H0 over a measurement period of 6 months, the subject H0 during the measurement period is estimated using the body temperature detection data D1 for this measurement period. This case will be described as an example.

First, the processing device 11 deletes the missing data with respect to the body temperature detection data D1 accumulated in advance for 6 months, and concatenates the remaining data D1. As a result, as shown in the characteristic line A1 in FIG. 10, a body temperature data series O in which about 5200 data D1 are concatenated is generated.

Next, the body temperature data series O is preprocessed and discretized, and then the hidden Markov model is constructed. As a result, the maximum likelihood model HMM that outputs the body temperature data series O is constructed. Therefore, the maximum likelihood state series Q that outputs the body temperature data series O using this constructed model HMM is constructed.^*To ask. As a result, as shown in the characteristic line A2 in FIG. 10, the state series Q having a periodicity of about one month corresponding to the body temperature data series O.^*Can be obtained. However, the state series Q^*Outliers occur throughout the measurement period, and the low temperature phase q₁And high temperature phase q₂There is a part where and is abruptly switched like the delta function.

Therefore, the processing device 11 uses the model HMM to obtain the state sequence Q.^*On the other hand, state identification processing is performed using a median filter or other probabilistic methods. As a result, the outliers can be removed. Therefore, as shown in the characteristic line A3 in FIG. 10, the state series Q^*Is the cold phase q₁And high temperature phase q₂Can be switched with periodicity in units of about one month, and the biphasic nature of the menstrual cycle can be clearly grasped.

Also, the high temperature phase q₂From low temperature phase q₁The point of change to is a signal of the beginning of menstruation. Therefore, the state sequence Q estimated by the processing device 11^*In the characteristic line A3 in Fig. 10, the high temperature phase q₂From low temperature phase q₁We compared the time when we switched to and the time when menstruation actually occurred. As a result, as shown in FIG. 10, the state series Q^*High temperature phase in q₂From low temperature phase q₁It was confirmed that the time of switching to and the time of menstruation were almost the same, and that the estimation of the menstrual cycle using the model HMM was effective.

Thus, according to this embodiment, the cold phase q using the body temperature data sequence O₁And high temperature phase q₂A model HMM having the two states of is constructed, and the model HMM outputs the body temperature data series O. The maximum likelihood state series Q^*Since we ask for this state series Q^*State contained in q_t^*By determining, the state of the subject H0 over a plurality of days when the body temperature data series O was measured can be specified. As a result, the biphasic nature of the subject H0 can be clearly grasped even when the measurement artifact occurs in the body temperature data series O due to the influence of the outside air or the like.

Further, the processing device 11 has a maximum likelihood state sequence Q.^*Since post-processing is performed on the median filter and other probabilistic methods, the state series Q^*It is possible to remove the discontinuous disengagement state contained in. As a result, the state does not change frequently every few hours or days, and the low temperature phase q has a periodicity of about one month.₁And high temperature phase q₂The biphasic nature of the subject H0, which repeats the above steps, can be clearly grasped.

Further, in the present embodiment, the processing device 11 generates the body temperature data series O by deleting at least one of the preset temperature and the temperature change in the body temperature detection data D1 that is out of the permissible range. To do. Therefore, when constructing the model HMM, the body temperature detection data D1 outside the permissible range of temperature or the permissible range of temperature change is not used. As a result, the abnormal value of the body temperature detection data D1 does not affect the model HMM, and when the state of the subject H0 is estimated using the model HMM, the accuracy of this estimation result can be improved.

Further, since the processing device 11 performs filtering processing using the median filter on the body temperature data series O before constructing the model HMM, it is possible to remove discontinuous outliers included in the body temperature data series O. it can. Then, since the treatment device 11 constructs the model HMM using the body temperature data series O after performing this filtering process, the model parameters of the model HMM are compared with the case where the body temperature data series O including the outliers is used. It is possible to prevent outliers from affecting λ, and it is possible to improve the estimation accuracy of the state of the subject H0.

Further, the processing device 11 uses the body temperature detection data O of the body temperature data series O._tSince the model HMM was constructed with the values discrete with respect to the temperature, the model HMM is the discrete body temperature detection data O._tIt has a body temperature detection data output probability B consisting of a discrete probability distribution according to. Therefore, body temperature detection data O that is discrete with respect to temperature_tMaximum likelihood state sequence Q using body temperature data sequence O consisting of^*Can be sought. Further, since the discrete hidden Markov model HMM is used, the time and storage capacity required for the construction process of the hidden Markov model can be reduced as compared with the case where the continuous hidden Markov model is used.

Further, in the present embodiment, the casing 2 of the sensor device 1 is provided with a display unit 9 for transferring the body temperature detection data D1 stored in the storage unit 6 to the external processing device 11. Therefore, using the two-dimensional code displayed on the display unit 9, the body temperature detection data D1 in the storage unit 6 can be transferred to an external processing device 11, and the model HMM can be constructed using the processing device 11. it can.

Therefore, for example, even when constructing a model HMM using the body temperature detection data D1 for several months for the subject H0, the processing device 11 stores a large amount of the body temperature detection data D1 in a large-capacity storage circuit. Since it can be accumulated and processed at high speed, model HMM construction and maximum likelihood state sequence Q^*Can be calculated quickly.

In the first embodiment, steps 1 to 4 in FIG. 8 are specific examples of data series generation means, steps 1 and 2 are specific examples of out-of-range data deletion means, and steps 3 and 4 are preprocessing means. Specific examples are shown respectively. Further, steps 5 in FIG. 8 and steps 11 to 20 in FIG. 9 show specific examples of the model construction means, step 6 shows a specific example of the state series calculation means, and step 7 shows a specific example of the state identification means. ..

Further, in the preprocessing in step 3 and the state identification processing in step 7, a median filter is used, but for example, an outlier (outlier state) may be removed, and a low-pass filter or the like is used. It may be configured.

Furthermore, the median filter used for preprocessing and state identification processing is configured to perform filtering using, for example, 11 data, but the number of data used for filtering is determined by the time interval for detecting body temperature detection data D1. It is set as appropriate according to the situation.

Next, FIGS. 11 and 12 show a second embodiment of the present invention, and a feature of this embodiment is that the processing apparatus has a body temperature detection data of a body temperature data series having a continuous value with respect to temperature. It is in the configuration that the hidden Markov model is constructed in the state. In this embodiment, the same components as those in the first embodiment described above are designated by the same reference numerals, and the description thereof will be omitted.

Reference numeral 21 denotes a processing device according to the present embodiment, and the processing device 21 constitutes a data series generation means, a model construction means, a state series calculation means, and a state identification means in substantially the same manner as the processing device 11 according to the first embodiment. To do. However, as shown in FIG. 12, the processing device 21 uses the body temperature detection data O of the body temperature data series O._tIt differs from the processing apparatus 11 according to the first embodiment in that the hidden Markov model HMM is constructed in a state where (1 t T) has a continuous value with respect to the temperature.

Specifically, the processing device 21 does not perform data discretization processing. Therefore, body temperature detection data O_t(1 t T) is a continuous value with respect to temperature.

In addition, the model HMM is a continuous body temperature detection data O_tIt has a body temperature detection data output probability B consisting of a continuous probability distribution according to. Therefore, the body temperature detection data output probability B is expressed by the following equation 11 on the assumption that it is based on the synthesis of M Gaussian distributions. At this time, the restriction conditions shown in the equation of Equation 12 are added.

<maths num="11"><img file="JP2008264352A_D0011.tif" /></maths>

<maths num="12"><img file="JP2008264352A_D0012.tif" /></maths>

When updating (re-estimating) the model parameter λ according to the definition of the body temperature detection data output probability B, each parameter of the Gaussian distribution density function G (mixing coefficient c)_jm, Average vector μ_jm, Covariance matrix U_jm) Is obtained by using the following equations of equations 13 to 16.

<maths num="13"><img file="JP2008264352A_D0013.tif" /></maths>

<maths num="14"><img file="JP2008264352A_D0014.tif" /></maths>

<maths num="15"><img file="JP2008264352A_D0015.tif" /></maths>

<maths num="16"><img file="JP2008264352A_D0016.tif" /></maths>

Thus, even in the second embodiment configured in this way, almost the same effect and effect as in the first embodiment can be obtained. In particular, in the second embodiment, the processing device 21 uses the body temperature detection data O of the body temperature data series O._tSince a hidden Markov model HMM is constructed with a continuous value for temperature, the model HMM is a continuous body temperature detection data O._tIt has a body temperature detection data output probability B consisting of a continuous probability distribution according to. Therefore, continuous body temperature detection data O with respect to temperature_tState sequence Q using a body temperature data sequence O consisting of a continuous hidden Markov model HMM^*Can be sought. Further, since the continuous hidden Markov model HMM is used, the state series Q is compared with the case where the discrete hidden Markov model is used as in the first embodiment.^*The accuracy of

Next, FIGS. 13 and 14 show a third embodiment of the present invention, and the feature of the present embodiment is that the processing device displays the state of the person to be measured by using a pie chart and a bar graph. I have done it. In this embodiment, the same components as those in the first embodiment described above are designated by the same reference numerals, and the description thereof will be omitted.

FIG. 13 shows a pie chart 31 displayed on the home page by the processing device 11. Here, the pie chart 31 is divided into eight equal fan-shaped first to eighth stages 31A to 31H, and half of the first to fourth stages 31A to 31D correspond to the low temperature period, and the remaining fifth to The 8th stage 31E-31H corresponds to the high temperature period. As a result, the pie chart 31 displays whether the person to be measured corresponds to the low temperature period or the high temperature period (first to eighth stages 31A to 31H).

Then, it is assumed that one lap of the pie chart 31 corresponds to one cycle of basal body temperature, and one cycle (one lap) is basically 28 days. At this time, 4 days are assigned to each of the 1st, 2nd, 5th, and 6th stages 31A, 31B, 31E, and 31F, and the 3rd, 4th, 7th, and 8th stages 31C, 31D, 31G, and 31H are assigned. 3 days will be allocated to each. Further, for the first day of the first stage 31A, for example, the menstruation start date recorded by the subject H0 using the sensor device 1 is set.

Then, in principle, the processing device 11 sets the date on which the latest body temperature detection data D1 is measured (latest measurement date) to any of the first to fourth stages 31A to 31D according to the number of days elapsed from the latest change point. Determine if applicable. That is, in principle, the processing device 11 determines which of the first to fourth stages 31A to 31D corresponds to the latest measurement date according to the number of days elapsed from the first day of the first stage 31A. Then, the processing device 11 displays, for example, a character symbol (for example, a star) during the stage to which the latest measurement date corresponds. The corresponding stage is not limited to the character symbol, and may be configured to be displayed by, for example, blinking.

However, after passing the latest falling point, the low temperature phase q₁From high temperature phase q₂If the change point to (rising change point) has also passed, the day on which this rising change point occurs is determined to be the day when the fifth stage 31E is entered. Then, it is determined which of the 5th to 8th stages 31E to 31H corresponds to the latest measurement date according to the number of days elapsed from the rising change point, and for example, a character symbol is displayed in the corresponding stage.

On the contrary, when the latest measurement date is 12 days or more after the first day of the 1st stage 31A and no rising change point occurs, the character symbol is displayed in the 4th stage 31D as if it remains in the 4th stage 31D. To do. In particular, when the latest measurement date is 15 days or more after the first day of the first stage 31A and no rising change point occurs, it is judged that the low temperature period is prolonged and a message to that effect is displayed.

In addition, if the next falling change point does not occur even after 17 days or more have passed from the first day (the day of the latest rising change point) of the fifth stage 31E, the processing device 11 is the eighth stage. In addition to displaying letters and symbols during stage 31H, it will indicate that there is a possibility of pregnancy. Here, in the case of pregnancy, the basal body temperature generally gradually decreases from around 16 weeks, and the low temperature phase q around 20 weeks.₁Since it returns to, the low temperature phase q between 16 and 20 weeks₁When you return to, you will be notified that you are likely to become pregnant.

FIG. 14 shows a bar graph 32 displayed on the home page separately from the pie chart 31 by the processing device 11. Here, the bar graph 32 determines the date on which the latest falling change point occurs as the menstruation start date, and displays the date together with the symbol indicating the start of menstruation (for example, a circle symbol).

Further, the processing device 11 displays one or a plurality (for example, three) of character symbols (for example, black star marks) indicating that the high temperature period is reached on the day corresponding to the high temperature period. On the other hand, the processing device 11 displays one or a plurality (for example, three) of character symbols (for example, white star marks) indicating that the low temperature period occurs on the day corresponding to the low temperature period.

As a result, compared to the case where the display by the line graph is used, the display can be made easier to see even on the screen of the mobile phone PT, for example.

The character symbol indicating the high temperature period may be displayed in red, the character symbol indicating the low temperature period may be displayed in blue, or the like.

In addition, the person to be measured is asked to declare the physical condition of the day (good or bad), and the bar graph 32 shows the physical condition of the day reported at the position corresponding to the date with simple character symbols (for example, φ is good, It may be configured to display using # (normally with b, bad mood with b).

Further, the next menstrual period (falling change point) may be predicted based on the past menstrual cycle, and the predicted date may be displayed. In this case, the next menstrual period may be predicted by adding the number of days for one cycle of the immediately preceding menstrual cycle to the previous menstrual period, and the average number of days for these one cycle for the past several menstrual cycles may be calculated. It may be calculated and predicted by adding this average number of days to the previous menstrual cycle. Furthermore, assuming that the extension and contraction of the menstrual cycle occur periodically, the number of days for one cycle of the menstrual cycle is calculated in consideration of the periodicity of this menstrual cycle, and the number of days for this calculated one cycle is calculated as the previous time. You may predict the next menstrual cycle by adding it to your menstrual cycle.

In addition, a period of 2 days before and after the day when the rising change point occurs may be set as a possible ovulation period, and a configuration to that effect may be displayed. Further, it may be configured to predict the next ovulation period based on the past menstrual cycle. In this case, the prediction of the next ovulation period can be performed in substantially the same manner as the prediction of the next menstrual period.

In addition, when luteal dysfunction occurs in the subject, the secretion of progesterone is insufficient and the high temperature period cannot be maintained, so that the body temperature drops in the latter half of the high temperature period, or menstruation occurs while the high temperature period is extremely short. It becomes difficult to get pregnant. Therefore, when the period of the high temperature period is less than 10 days, it is presumed that there is a possibility of luteal dysfunction, and a configuration may be configured to indicate that fact.

In the anovulatory cycle, although there is periodic bleeding, the basal body temperature does not change and menstruation occurs in the low temperature period. In the case of such a cycle pattern, there is no ovulation and no pregnancy occurs. Therefore, when the biphasic change of the menstrual cycle does not appear, it may be presumed that there is a possibility of an anovulatory cycle, and a configuration to that effect may be displayed.

Thus, even in the third embodiment configured in this way, almost the same effect and effect as in the first embodiment can be obtained. In particular, since the processing device 11 is configured to display the pie chart 31 and the bar graph 32 showing the state of the person to be measured, the person to be measured can visually check the pie chart 31 and the bar graph 32 to see the state of the latest measurement date and the like. Can be easily grasped.

Further, in the third embodiment, the pie chart 31 and the bar graph 32 are created by using the processing device 11 according to the first embodiment. However, the present invention is not limited to this, and for example, the pie chart 31 and the bar graph 32 may be created by using the processing device 21 according to the second embodiment.

Further, in each of the above-described embodiments, the hidden Markov model HMM has a low temperature phase q.₁And high temperature phase q₂The configuration has only two states. However, the present invention is not limited to this, for example, as in the hidden Markov model HMM according to the modification shown in FIG. 15, the low temperature phase q₁And high temperature phase q₃Medium temperature phase q between₂May be provided so that the hidden Markov model HMM has these three states. As a result, the low temperature phase q₁From high temperature phase q₃Rising change point to and high temperature phase q₃From low temperature phase q₁You can investigate the falling point of change in detail. Also, the medium temperature phase q₂If is provided, the low temperature phase q₁And high temperature phase q₃Since it is considered that the probability of direct transition between and is low, it is effective to use the Viterbi algorithm when finding the maximum likelihood state sequence in order to reduce the calculation time and the calculation cost.

Further, in each of the above-described embodiments, the sensor device 1 has a configuration in which the start time and the end time of the measurement are set in advance, and the body temperature detection data D1 is detected at regular time intervals between them. However, the present invention is not limited to this, and the sensor device 1 starts reading and storing the body temperature detection data D1 when, for example, the body temperature T1 detected by the body temperature detection unit 3 rises above a certain temperature (for example, 32 ° C). However, when the temperature drops below a certain level, the reading of the body temperature detection data D1 may be terminated.

Further, when the sensor device 1 is provided with a micro directional switch and the subject changes the posture from the sitting or standing position to the lying position, the azimuth switch is turned on and the measured person changes the posture from the lying position to the sitting or standing position. When changed, the orientation switch may be turned off. As a result, the sensor device 1 can start and end reading the body temperature detection data D1 according to the on state and the off state of the direction switch.

Further, in each of the above-described embodiments, the body temperature detection data D1 during sleep is detected a plurality of times per day, but for example, the body temperature may be detected only once per day after the body temperature is sufficiently stabilized.

<figref num="1">It is an external view which shows the state which the wearable sensor device by 1st Embodiment of this invention wears | measured person.</figref><figref num="2">It is a front view which shows the wearable sensor device in FIG. 1 by itself.</figref><figref num="3">It is a rear view which shows the wearable sensor device in FIG.</figref><figref num="4">It is sectional drawing of the wearable sensor device as seen from the direction of arrow IV-IV in FIG.</figref><figref num="5">It is a block diagram which shows the structure of the control unit provided in the wearable sensor device.</figref><figref num="6">It is explanatory drawing which shows the menstrual cycle estimation apparatus which consisted of a wearable sensor apparatus and an external processing apparatus.</figref><figref num="7">It is explanatory drawing which shows the hidden Markov model by 1st Embodiment.</figref><figref num="8">It is a flow chart which shows the menstrual cycle estimation process.</figref><figref num="9">It is a flow chart which shows the construction process of the hidden Markov model in FIG.</figref><figref num="10">It is a characteristic diagram which shows the body temperature data series after pretreatment, the maximum likelihood state series, and the state series after state identification processing, respectively.</figref><figref num="11">It is explanatory drawing which shows the menstrual cycle estimation apparatus by 2nd Embodiment.</figref><figref num="12">It is explanatory drawing which shows the hidden Markov model by 2nd Embodiment.</figref><figref num="13">It is explanatory drawing which shows the pie chart by the 3rd Embodiment.</figref><figref num="14">It is explanatory drawing which shows the bar graph by 3rd Embodiment.</figref><figref num="15">It is explanatory drawing which shows the hidden Markov model by the modification.</figref>

Code description

1 Wearable sensor device 2 Casing 3 Body temperature detection unit (body temperature detection means) 5 Control unit (data reading means) 6 Storage unit (storage means) 9 Display unit (transfer means) 11,21 Processing device