CN114678119A

CN114678119A - Device, method and biomarker for predicting ovulatory and menstrual periods using heart rate

Info

Publication number: CN114678119A
Application number: CN202210225146.0A
Authority: CN
Inventors: 杨凌; 徐璎; 王瑶; 董莺莺; 周飞
Original assignee: Suzhou University
Current assignee: Suzhou University
Priority date: 2022-03-07
Filing date: 2022-03-07
Publication date: 2022-06-28

Abstract

The present invention relates to a device for predicting ovulatory and menstrual periods using heart rate, a data acquisition module collects continuous heart rate data of a subject; the data preprocessing module determines night resting state data from the continuous heart rate data, and averages the night resting state data to obtain a daily average heart rate; the marker extraction module takes the peak point phase of the daily average heart rate derivative curve as a first biomarker for evaluating that the subject enters an ovulation period and/or takes the valley point phase of the daily average heart rate derivative curve as a second biomarker for evaluating that the subject enters a menstrual period; an assessment module assesses an indicator of likelihood of a subject entering an ovulatory period based on a first biomarker; and/or assessing an indicator of likelihood of the subject entering a menstruation period based on the second biomarker. The invention only uses heart rate data as the input of feature extraction, does not need to additionally collect other physiological parameters as features, and has the advantages of no wound, convenience and long-time continuous measurement.

Description

Device, method and biomarker for predicting ovulatory and menstrual periods using heart rate

Technical Field

The invention relates to the technical field of female physiological hygiene, in particular to a device and a method for predicting an ovulation period and a menstrual period by using a heart rate and a biomarker.

Background

The prediction of the female's menstrual period and ovulatory period is of great importance for contraception or pregnancy preparation in women of child bearing age, and considerable attention has been paid to the development of methods for determining the number of ovulatory days and the date of onset of menstruation. There are many methods for detecting or predicting female menstrual period and ovulation, such as ultrasonography, uroxanthogenin, serum progesterone, uropregnanediol 3-glucuronide, basal body temperature, cervical mucus, and the like.

The maximum growth and subsequent size reduction of the dominant follicle can be observed by ultrasonography, and the time to ovulation between the two can be determined. Since this time can be so well defined, it is recognized as a standard reference examination for ovulation detection, mainly for artificial reproduction techniques. The detection of uroluteinizing hormone (LH), both in serum and urine, is very sensitive and specific to ovulation and provides a high degree of accuracy in determining fertility. Before luteinizing hormone proliferated, serum estrogen levels increased and some changes in body fluid composition occurred, including cervical mucus and saliva, and observation of these differences allowed a better understanding of the fertility window. Progesterone is secreted only from the corpus luteum after ovulation and the detection of progesterone or its metabolites can retrospectively confirm the occurrence of ovulation. Since progesterone causes an increase in Basal Body Temperature (BBT), measuring basal body temperature is also useful for determining ovulation. Since oocytes die shortly after ovulation, methods related to progestogen and its effects determine that the fertility window is closed. However, these methods are either not performed continuously, or are not non-invasive, or are not performed by the subject. It is therefore of utmost importance to develop a new biomarker system for detecting or predicting the menstrual and ovulatory periods.

Wearable sensor technology is rapidly developing. These devices provide mainly information that provides insight into the physical activity of the user, and are increasingly used in healthcare institutions, and in recent years, many scholars have begun to study the circadian rhythms of people using data measured by wearable devices. For example, Lan Luo et al propose a non-invasive wearable fertility monitoring device that measures ear canal temperature at intervals, and a Hidden Markov Model (HMM) with two hidden states, high and low, to identify statistical learning algorithms with greater ovulation probability, and use the data captured by the device to detect and predict ovulation. Mohaned Sh i l a i h et al studied the correlation between Wrist Skin Temperature (WST) measured by a wearable sensor and ovulatory urine tests, and used it as a fertility tracking method. However, in the above body temperature monitoring, no matter the wrist temperature or the ear canal temperature is greatly influenced by the external temperature change, which may affect the accuracy of body temperature monitoring, and thus directly affect the accuracy of the prediction result.

Disclosure of Invention

Therefore, the technical problem to be solved by the invention is to overcome the problems in the prior art and provide a device, a method and a biomarker for predicting the ovulation period and the menstrual period by using the heart rate.

To solve the above technical problems, the present invention provides an apparatus for predicting an ovulatory period and a menstrual period using a heart rate, comprising:

a data acquisition module for collecting continuous heart rate data of a subject using a device capable of collecting continuous heart rate;

the data preprocessing module is used for dividing the continuous heart rate data of the subject according to days to determine a night resting state time period to obtain night resting state data, removing a fluctuation peak caused by interference factors in the night resting state data, and averaging the night resting state data to obtain a daily average heart rate;

a marker extraction module for obtaining a peak phase of a daily average heart rate derivative curve and a valley phase of the daily average heart rate derivative curve, using the peak phase of the daily average heart rate derivative curve as a first biomarker for evaluating the entry of the subject into an ovulation period, and/or using the valley phase of the daily average heart rate derivative curve as a second biomarker for evaluating the entry of the subject into a menstrual period;

an assessment module for assessing an indicator of likelihood of a subject entering ovulation phase based on a first biomarker; and/or assessing an indicator of likelihood of the subject entering a menstruation period based on the second biomarker.

In one embodiment of the invention, the apparatus is at least one of an input device and a wearable device operatively attached to the computing device.

In one embodiment of the invention, the wearable device is a wearable device with a heart rate module capable of collecting continuous heart rates.

In one embodiment of the present invention, the marker extraction module comprises:

the filtering submodule is used for filtering the average daily heart rate to obtain a filtered average daily heart rate curve;

and the derivation submodule is used for deriving the daily average heart rate curve, calculating all peak point phases on the derivative curve and/or calculating all valley point phases on the derivative curve.

In one embodiment of the invention, the evaluation module comprises:

the prediction submodule is used for training a model by using the previous characteristic data and a machine learning method based on the peak point phase of a daily average heart rate derivative curve and combining the daily average heart rate curve derivative and a second derivative as characteristic data, and the possibility index of the next day of a subject entering the ovulation period can be output by using the trained model and the current characteristic as input; and/or based on the valley point phase of the daily average heart rate derivative curve, combining the current and previous daily average heart rate curve derivatives and second-order derivatives as characteristic data, training by using the previous characteristic data and a machine learning method to obtain a model, and outputting a possibility index that the test subject enters the menstrual period on the next day by using the trained model and the current characteristic as input.

In addition, the present invention provides a method for predicting an ovulation period and a menstruation period using a heart rate, comprising:

collecting continuous heart rate data of the subject using a device capable of collecting continuous heart rate;

dividing the continuous heart rate data of the subject according to days to determine a night resting state time period to obtain night resting state data, removing a fluctuation peak caused by interference factors in the night resting state data, and averaging the night resting state data to obtain a daily average heart rate;

acquiring a peak point phase of a daily average heart rate derivative curve and a valley point phase of the daily average heart rate derivative curve, and taking the peak point phase of the daily average heart rate derivative curve as a first biomarker for evaluating that the subject enters an ovulation period, and/or taking the valley point phase of the daily average heart rate derivative curve as a second biomarker for evaluating that the subject enters a menstrual period;

assessing an indicator of likelihood of the subject entering ovulation phase based on the first biomarker; and/or assessing an indicator of likelihood of the subject entering a menstruation period based on the second biomarker.

In one embodiment of the invention, the method for acquiring the peak point phase of the daily average heart rate derivative curve and the valley point phase of the daily average heart rate derivative curve comprises the following steps:

filtering the average daily heart rate to obtain a filtered average daily heart rate curve;

And (3) carrying out derivation on the daily average heart rate curve, and calculating all peak point phases on the derivative curve and/or calculating all valley point phases on the derivative curve.

In one embodiment of the invention, an indicator of the likelihood of a subject entering ovulation phase is assessed based on the first biomarker; and/or assessing an indicator of likelihood of the subject entering a menstruation period based on the second biomarker, comprising:

based on the peak point phase of the daily average heart rate derivative curve, combining the daily average heart rate curve derivative and the second derivative as characteristic data, training by using the previous characteristic data and a machine learning method to obtain a model, and outputting a possibility index of a next day subject entering an ovulation period by using the trained model and the current characteristic as input; and/or based on the valley point phase of the daily average heart rate derivative curve, combining the current and previous daily average heart rate curve derivatives and second-order derivatives as characteristic data, training by using the previous characteristic data and a machine learning method to obtain a model, and outputting a possibility index that the test subject enters the menstrual period on the next day by using the trained model and the current characteristic as input.

Also, the present invention provides a biomarker for predicting an ovulation period and a menstruation period, which is applied to the apparatus for predicting an ovulation period and a menstruation period using a heart rate as described above, the biomarker including a first biomarker for evaluating an onset of an ovulation period of a subject using a phase of a peak of a daily average heart rate derivative curve, and/or a second biomarker for evaluating an onset of a menstruation period of a subject using a phase of a valley of the daily average heart rate derivative curve.

Also, the present invention provides the use of a biomarker as described above in the manufacture of a device for assessing an indicator of the likelihood of a subject entering an ovulatory period or a menstrual period, including the use of a first biomarker in the manufacture of a device for assessing an indicator of the likelihood of a subject entering an ovulatory period, and/or the use of a second biomarker in the manufacture of a device for assessing an indicator of the likelihood of a subject entering a menstrual period.

Compared with the prior art, the technical scheme of the invention has the following advantages:

1. the invention only uses the heart rate data as the input of the feature extraction, and does not need to additionally collect other physiological parameters as the features, so that the invention has the advantages of no wound, convenience and long-time continuous measurement in the data collection process.

2. The invention obviously distinguishes the biomarker for predicting the menstrual period from the biomarker for predicting the ovulatory period, and can effectively improve the accuracy of prediction.

3. The invention takes the biomarker as an important index for detecting or predicting the menstrual period and the ovulatory period, and the disclosure of the biomarker has important significance for predicting the fertility window.

Drawings

In order that the present disclosure may be more readily and clearly understood, reference will now be made in detail to the present disclosure, examples of which are illustrated in the accompanying drawings.

Fig. 1 is a schematic flow chart of a method for predicting an ovulatory period and a menstrual period using a heart rate according to the present invention.

FIG. 2 is heart rate data, where a-b represent the original minute-scale heart rate; c denotes the average daily heart rate extracted from the resting night heart rate.

FIG. 3 is a flow chart of a marker extraction method.

FIG. 4 is a graphical representation of heart rate analysis results with thin lines representing the raw heart rate curves; the continuous thick line represents the filtered curve; the dashed line represents the curve fitted by the global cos; black hexagons represent the lowest points of the global fitting curve; x represents a local valley point; denotes local peak points; the bold line segments on both sides represent the heart rate curves at sleep and wake up, and the average daily heart rate is taken as the mean of the heart rates between sleep and wake up.

Fig. 5 is a flow chart of biomarker data segmentation.

Fig. 6 shows the average daily heart rate filtering result, where the broken line shows the average daily heart rate and the curve shows the filtering result.

Fig. 7 shows the result of the derivation of the daily average heart rate curve, in which the asterisk is the mark of real ovulation and the pentagram is the mark of the first day of real menstruation.

FIG. 8 is a graph showing the peak positions of the derivatives of the true average phase ovulation plotted against the average daily heart rate curve for a uniform period length, wherein the thin line indicates the uniform length heart rate derivative, the shading indicates the derivatives. + -. standard deviation, the dots indicate the ovulation day for the average phase, and the thick line indicates the ovulation day. + -. standard deviation.

FIG. 9 shows the first day of menstruation after the true average phase marked at the valley position of the derivative of the daily average heart rate curve of uniform length, where the thin line represents the heart rate derivative of uniform length, the shading represents the derivative. + -. standard deviation, the dots represent the first day of menstruation of the average phase, and the thick line represents the first day of menstruation. + -. standard deviation.

Fig. 10 is a prediction flow chart.

Fig. 11 is a result of prediction of ovulation in a subject, a, hatching indicates the actual ovulation day and two days before and after the actual ovulation day, dots indicate the result of prediction, a value of 1 indicates no ovulation, and a value of 2 indicates the period of prediction of ovulation; b. the horizontal line is the threshold k, the curve represents the value of the probability index obtained by the model predicting ovulation, and the shading is the same as that in the graph a.

FIG. 12 shows the menstruation prediction of a subject. a. The shading shows the first day and every two days before and after the real menstruation, the point shows the prediction result, the value of the point is 1 to show that the menstruation is not performed, and the value of the point is 2 to show that the menstruation is predicted; b. the horizontal line is a threshold value k, the curve represents the value of the probability index obtained by predicting menstruation with the model, and the hatching represents the same as that in fig. a.

Detailed Description

The present invention is further described below in conjunction with the following figures and specific examples so that those skilled in the art may better understand the present invention and practice it, but the examples are not intended to limit the present invention.

In this application, the singular forms "a", "an" and "the" include plural referents unless the context clearly dictates otherwise.

As used herein, the term "subject" is preferably a human non-menopausal female, and the subject presents a menstrual cycle rhythm.

As used herein, the term "resting data," also referred to as "nighttime resting data," refers to heart rate data for a period of nighttime sleep obtained after calibration time, data partitioning, and filtering, and after removal of interfering data such as sleep cycles, overnight activity, and the like. The term "night rest period" refers to a period of night sleep state obtained by calibration time, data division, and filtering. The night resting state data is data subjected to data preprocessing.

As used herein, the term "phase of the peak of the nocturnal resting heart rate derivative" refers to the time at the maximum point after the nocturnal resting heart rate pre-processing, filtering and derivation, and recalculation; the term "phase of the valley point of the nighttime resting heart rate derivative" refers to the time at the minimum point after nighttime resting heart rate preprocessing, filtering and derivation, and recalculation.

As used herein, the term "wearable device" refers to a portable device worn directly on a user with a heart rate module capable of collecting continuous heart rates. The wearable device may be a hardware device that may be supported by software for data interaction, cloud interaction, or data analysis purposes.

The experimental procedures in the following examples are conventional unless otherwise specified. The invention will be further understood with reference to the following non-limiting experimental examples.

The present invention provides an apparatus for predicting an ovulatory period and a menstrual period using a heart rate, comprising:

The invention only uses the heart rate data as the input of the feature extraction, and does not need to additionally collect other physiological parameters as the features, so that the invention has the advantages of no wound, convenience and long-time continuous measurement in the data collection process.

In one embodiment of the invention, the evaluation module comprises:

In accordance with an embodiment of the above apparatus, referring to fig. 1, the present invention also provides a method for predicting an ovulation period and a menstruation period using a heart rate, comprising:

s11: collecting continuous heart rate data of the subject using a device capable of collecting a continuous heart rate;

s12: dividing the continuous heart rate data of the subject according to days to determine a night resting state time period to obtain night resting state data, removing a fluctuation peak caused by interference factors in the night resting state data, and averaging the night resting state data to obtain a daily average heart rate;

S13: acquiring a peak point phase of a daily average heart rate derivative curve and a valley point phase of the daily average heart rate derivative curve, and taking the peak point phase of the daily average heart rate derivative curve as a first biomarker for evaluating that the subject enters an ovulation period, and/or taking the valley point phase of the daily average heart rate derivative curve as a second biomarker for evaluating that the subject enters a menstrual period;

s14: assessing an indicator of likelihood of the subject entering ovulation phase based on the first biomarker; and/or assessing an indicator of likelihood of the subject entering a menstruation period based on the second biomarker.

Specifically, in step S11, continuous heart rate data of the subject is collected using a device capable of collecting continuous heart rates, such as a wearable device with a heart rate module, wherein the heart rate data density is conventionally 1 data point per minute, and the nighttime rest portion of each day is selected to be averaged to generate a daily average heart rate of 1 point per day.

Specifically, in step S13, selecting a suitable filtering parameter, and performing filtering processing on the daily average heart rate by using a Butterworth Filter (Butterworth Filter), wherein a filtering curve shows an obvious monthly rhythm; the average daily heart rate curve is derived, the phases of all peak points (and/or valley points) on the derivative curve, namely the maximum value (and/or minimum value) points of the curve, are calculated, and then the phases of the peak points (and/or valley points) can be used for detecting the ovulation period (and/or the menstrual period).

Specifically, in step S14, based on the phase of the peak point of the daily average heart rate derivative curve, the daily average heart rate curve derivative and the second derivative are used as feature data, the previous feature data and the machine learning method are used to train and obtain a model, and the trained model is used to output the possibility index that the next day of the test subject enters the ovulation period by using the current feature as input; and/or based on the valley point phase of the daily average heart rate derivative curve, combining the current and previous daily average heart rate curve derivatives and second-order derivatives as characteristic data, training by using the previous characteristic data and a machine learning method to obtain a model, and outputting a possibility index that a subject enters the menstrual period on the next day by using the trained model and the current characteristic as input.

For example, if the output possibility index is probability1 ═ model _ kde1(x), x ∈ R^n×3，probability1∈R^n×1The probability1(i) that the day i +1 is the ovulation day can be obtained by training the features before the day i to obtain a model _ ked1 and inputting the feature x of the day i. Set the threshold k 0.006, if probability1(i) ≧ k and [ probability1(i)]' > 0 and [ viability 1(i-1)]' > 0, then prediction1(i + 1: i + t) ═ 2, namely from the (i + 1) th day to the (i + t) th day, the mark is ovulation, t is defined as the number of days of ovulation period duration, and the next prediction starts from the (i + 20) th day; otherwise prediction1(i) ═ 1, predicts that day i +1 is not ovulation, and predicts next from day i + 2.

In accordance with an embodiment of the above apparatus, the present invention also provides a biomarker for predicting an ovulation period and a menstruation period, which is applied to the apparatus for predicting an ovulation period and a menstruation period using a heart rate as described above, the biomarker including a first biomarker for evaluating an onset of an ovulation period of a subject as a phase of a peak point of a daily average heart rate derivative curve, and/or a second biomarker for evaluating an onset of a menstruation period of a subject as a phase of a valley point of a daily average heart rate derivative curve.

In accordance with embodiments of the biomarkers described above, the present invention also provides the use of a biomarker as described above in the manufacture of a device for assessing an indicator of the likelihood of a subject entering an ovulatory period or a menstrual period, including the use of a first biomarker in the manufacture of a device for assessing an indicator of the likelihood of a subject entering an ovulatory period, and/or the use of a second biomarker in the manufacture of a device for assessing an indicator of the likelihood of a subject entering a menstrual period.

The present invention will be further described below by way of specific embodiments and experimental data.

Example 1 Heart Rate data analysis processing

The analysis method of the present invention uses continuously recorded heart rate data, which can be derived from wearable devices with heart rate modules, and the heart rate data density is conventionally 1 data point per minute, as shown in fig. 2a and 2 b.

91 non-menopausal female volunteers are recruited, the volunteers need to wear the intelligent bracelet for at least one month for data acquisition, meanwhile, the volunteers record the menstrual time every month, and the menstrual cycles are recorded in 603 total, wherein 27 volunteers record the ovulation date through test paper measurement. Wearable devices were purchased from two independent vendors. The heart rate HR data of the frequency of 1 minute collected by the intelligent wrist strap is retrieved from a manufacturer cloud server through an application program interface after the user authorization, and is stored in a local database.

Because the heart rate data in the daytime is influenced by various subjective factors such as motion, working strength, social contact and the like and external factors, the biological rhythm of an individual in the day cannot be objectively reflected, the heart rate data in a resting state at night is selected as a data source for acquiring the rhythm parameters. After the interference of sleep cycle, night activity and the like in the resting state data at night is removed through preprocessing, a trigonometric function is used for fitting to obtain the core global parameters. In addition, small peak-to-valley parameters in the nighttime resting data were also obtained using Butterworth Filter (Butterworth Filter) filtering. Referring to fig. 3, the general flow chart specifically includes the following steps:

1) obtaining a time period of resting state data:

1-1) calibrating time

The time data is a reference for subsequent processing, and the initial time needs to be calibrated before work starts in consideration of the particularity and difference of different software on time data storage.

1-2) data partitioning

The time series data was divided in units of days with 14 points per day as the division point. And meanwhile, screening the incomplete data.

1-3) Filtering

The daily heart rate data is plotted as a curve formed by the superposition of a macroscopic biological clock curve and a number of small noises (jitters), i.e. consisting of a low frequency component and a high frequency component, respectively. Wherein the curve corresponding to the low frequency component changes slowly, and the curve corresponding to the high frequency component changes violently.

In order to study the biological clock curve of the sleep stage, it is necessary to filter out high frequency noise, so a low frequency filter is used to filter or attenuate the high frequency component of the curve by a large margin, and let the low frequency component pass. Here, a Butterworth low pass filter is used to obtain filtered data.

1-4) obtaining sleep data (resting state)

a. And calculating the average value B _ mean of the heart rate of the whole day and the median value B _ prct i l e of the heart rate of the whole day by using the filtered data smooth _ f i filter, and taking the minimum value of the average value B _ mean and the median value B _ prct i l e of the heart rate of the whole day as a critical value B _ i nf.

b. And (3) calculating a sleep data starting point, namely, taking the length i nterva l (1) as 240 (unit: min), and calculating the heart rate which corresponds to how many points in the time interval [1: i nterva l (1) ] is less than B _ i nf, namely, the number of points positioned below B _ i nf is recorded as count _ poi.

c. If count _ poi nt is equal to 0, the time interval slides to the right by i nterva l (1), and the operation similar to b is performed on the new time interval.

If 0< count _ poi < i nterva l (1), slide 1 unit to the right, do the similar operation of b for the new time interval.

If the count _ poi nt is i nterva l (1), the left end point of the corresponding time interval is marked as the starting point of the sleep data, and the loop is ended.

d. Finding a sleep data termination point:

the principle is the same as a, and judgment is carried out from right to left instead.

e. If count _ poi nt is 0, the time interval is slid to the left by i nterva l (1), and the operation d is repeated for the new time interval.

If 0< count _ poi nt < i nterva l (1), slide 1 unit to the left and repeat d for the new time interval.

If the count _ poi nt is i nterva l (1), the right end point of the corresponding time interval is marked as the sleep data end point, and the cycle is ended.

In the case of excessively short night sleep time, the above method may not be able to effectively acquire the required data, and at this time, the interval length is changed from i nterva l (1) to 240 to i nterva l (2) to 120 (unit: min), and the relevant steps are repeated.

f. To obtain more complete sleep data, consider to extend both ends slightly.

If there is an extension margin at the left end (start end), it is considered that the start point is moved to the left by cont i nue _ l en equal to 60 (unit: min). If an overflow occurs, that is, the starting point is out of the actual recording range, 0<, which is the movement amount < cont i nue _ l en, is taken.

For this extended segment of data, a first order difference is made. The rising number is defined as the number greater than zero in the differential data, and the falling number is defined as the number less than zero in the differential data. And calculating the ascending number and the descending number, and if the descending number > is 2 times of the ascending number, determining that the extension is necessary and changing the coordinates of the starting point, otherwise, the extension is not needed, namely, the coordinates are not needed to be modified.

g. For the right end (terminating end), if there is an extension margin, it is considered to move the terminating point to the right by cont i nue _ l en of 60 (unit: min). If an overflow occurs, that is, when the end point is out of the actual recording range, 0<, which is the movement amount < cont i nue _ l en, is taken.

For the extended data, a first difference is made, the ascending number and the descending number are calculated according to the definition, if the ascending number > is 2 times of the descending number, the extension is considered to be necessary, the coordinates of the starting point are changed, otherwise, the extension is not needed, namely, the coordinates are not needed to be modified.

h. The sleep data is segmented from the coordinates determined by f and g.

2) In the time period of rest state, removing the fluctuating peak (considered as interference factors such as sleep cycle and night activity) to obtain the preprocessed night rest state data:

for the acquired sleep data, because the whole sleep stage comprises a plurality of small sleep cycles which reflect a plurality of small fluctuations on the heart rate curve, in order to acquire macroscopic sleep resting state data, a sliding window is used for traversing and comparing the data, and a starting point/an end point of a local peak and a local trough is defined and deleted.

3) Averaging the preprocessed night resting state data to obtain a daily average heart rate, and obtaining daily average heart rate data of one point in a day:

the sleep data are divided into three equal parts, namely left, middle and right parts.

And obtaining the slope value when the sleep state is entered by using the left section data. A time interval having a length i nterva l (3) ═ 50 is taken, and a statistic "decline score" is defined to determine the quality of the time interval selected at that time (i.e., the effect of obtaining the slope for that interval). Defining: first-order difference is carried out on the interval data, the descending number (the number of the data less than zero in the difference data) is calculated, the descending amount (the sum of the values less than zero in the difference data and then the absolute value is taken), and then the two are summed by taking 0.3 and 0.7 as weights, so that the descending score of the interval is called. Traversing the left segment, searching the segment with the highest score, searching the maximum value max _ i nfo of the left half segment of the segment and the position max _ p l ace of the left half segment of the segment, and searching the minimum value mi n _ i nfo of the right half segment and the position of the right half segment of the segment. The slope of going to sleep is the number of heart rate drops per minute (negative) over the period. The falling end time is the time corresponding to the last moment of the sleep time period, and the falling end heart rate is the heart rate corresponding to the falling end time.

And obtaining the slope value of the wake-up stage by using the right-segment data. The "rise score" is defined as a time interval of length i nterva l (3) ═ 50, and is used to determine the quality of the time interval selected at that time (i.e., the effect of obtaining the slope from the interval). Defining: and (3) performing first-order difference on the interval data, calculating the rising number (the number of the difference data larger than zero), the rising amount (the sum of the values of the difference data larger than zero), and summing the rising number and the rising amount by taking 0.3 and 0.7 as weights, wherein the sum is called the rising score of the interval. And traversing the right segment and searching the segment with the highest score. And searching the minimum value mi n _ i nfo of the left half section of the interval and the position mi n _ p l ace of the minimum value, and searching the maximum value max _ i nfo of the right half section and the position max _ p l ace of the maximum value. The value of the slope of the wake-up period is the number of heart rate rises per minute over the period. The rising start time is a time corresponding to the first time of the wake-up period, and the rising start heart rate is a heart rate corresponding to the rising start time.

The average daily heart rate is the average of the resting night heart rate between the last time of the sleep session and the first time of the wake session, see fig. 4, and the average daily heart rate is shown in fig. 2 c.

Example 2 biomarker assay result determination

To demonstrate that the peak phase of the nocturnal resting heart rate derivative can be used as a biomarker for ovulation and the valley phase of the nocturnal resting heart rate derivative can be used as a biomarker for menstruation, the inventors compared the biomarkers collected from volunteers using ovulation records and processed heart rate data at the same time period as the ovulation records, and the analysis steps and results were as follows:

1) Data acquisition:

biomarker extraction was based on night resting heart rate, which, as in example 1, refers to the average daily heart rate generated by taking the night sleep part from the minute-scale heart rate data and averaging the minute-scale data of the night sleep part, a time series of 1 data point in 1 day.

2) Data preprocessing:

2-1) because of a plurality of reasons such as uploading is not carried out in time, the volunteers do not wear the bracelet continuously, the heart rate data has some missing values, and the incomplete data is screened.

2-2) performing linear interpolation supplement on missing values which are continuously missing for less than five days, wherein the missing value supplement mode is not limited to a linear interpolation method, and the missing values can not be supplemented if a filter is used to support data containing the missing values.

3) Data filtering:

the preprocessed heart rate data are plotted as a curve, which contains a macroscopic biological clock and a lot of high-frequency noise jittering up and down, i.e. consisting of low-frequency components and high-frequency components, respectively. Wherein the curve corresponding to the low frequency component changes slowly and the curve corresponding to the high frequency component changes violently. In order to study the monthly rhythm, high frequency noise needs to be filtered out, so a low frequency filter is used for filtering or greatly attenuating the high frequency component of the curve, and the low frequency component is passed through. Here a Butterworth low pass filter is used resulting in filtered data, see fig. 6, which exhibits a pronounced monthly rhythm.

4) Derivation:

the filtered heart rate is derived, see fig. 7.

5) Marking of menstruation and ovulation:

the menses and ovulation recorded by the volunteers were marked on the heart rate filter data and derivative data according to the corresponding date, wherein the marking of the menses on the heart rate curve refers to the first day of menses.

6) Ovulation day versus ovulation biomarker:

6-1) data slicing

Firstly, normalizing the derivative of the heart rate to be between-1 and 1, segmenting the normalized derivative curve of the heart rate from valley points to form a segment between every two valley points to represent a physiological cycle, and the segmentation flow chart is shown in figure 5;

6-2) uniform period length

Since the lengths of the menstrual cycles of different volunteers are different and the lengths of the menstrual cycles of the same volunteer in different time periods are not completely the same, all the divided periods need to be unified to 28 days in order to better compare all the menstrual cycles of different volunteers. If the period length is L, the distance between every two points in the period is multiplied by 28/L, and the period length is changed into 28 days. The relative position of the markers in each cycle is constant during the cycle, i.e. if the ovulation is centred in a split menstrual cycle, it is also centred in the 28 day cycle after stretching.

6-3) period averaging

And interpolating all the periods which are segmented and have the same length between 0 and 28 by using the same sequence as time, so that the lengths of all the periods are 28 days and have numerical values at the same time point, n periods are total, and the ith period is marked as ai.

After processing as above, all cycles are averaged

The relative position of ovulation in the cycle is also averaged over the averaged menstrual cycle and the results are shown in figure 8, and the comparison shows that the phase of the peak of the nocturnal resting heart rate derivative can mark the day of ovulation well and is a good marker.

7) First day of menstruation compared to biomarkers of menstruation:

7-1) data slicing

Firstly, normalizing the derivative of the heart rate to be between-1 and 1, segmenting a derivative curve of the heart rate after normalization from peak points to form a segment between every two peak points, representing a physiological cycle, segmenting into m cycles, and a segmentation flow chart is shown in figure 5;

7-2) uniform period length

Since the lengths of the menstrual cycles of different volunteers are different and the lengths of the menstrual cycles of the same volunteer in different time periods are not completely the same, all the divided periods need to be unified to 28 days in order to better compare all the menstrual cycles of different volunteers. If the period length is L, the distance between every two points in the period is multiplied by 28/L, and the period length is changed into 28 days. The relative position of the marked ovulation in each cycle is constant during the cycle, i.e. if the first day of menstruation is centred in a divided menstrual cycle, it is also centred in the 28 day cycle after stretching.

7-3) period averaging

And (3) interpolating all the periods which are divided and have the same length between 0 and 28 by using the same sequence as time, so that all the periods have the length of 28 days and have numerical values at the same time point, and m periods are total, wherein the ith period is marked as bi.

After processing as above, all cycles were averaged

The relative position of the first day of menstruation in the cycle was also averaged over the averaged menstrual cycle and the results are shown in fig. 9, the comparison above shows that the trough phase of the nocturnal resting heart rate derivative can mark menses well and is a good marker.

Example 3 prediction of ovulation using markers of ovulation

The prediction flow chart is shown in fig. 10. In ovulation prediction, 27 volunteers participating in ovulation monitoring are used, luteinizing hormone ovulation test paper is used for detecting on day 8 of a menstrual cycle until a positive result is obtained, 86 cycles are monitored in total, 78 positive ovulation results are obtained, in order to ensure the accuracy of prediction, in subsequent data processing, data which is more than one month in front of heart rate data participating in prediction or training is required to facilitate data preprocessing, so that the first positive detection result of part of the volunteers is not included in a training data set or a prediction data set, and 60 ovulation cycles which are finally involved in training and prediction are obtained. One subject used two wearable devices at different times, so the data was grouped into 28 groups, numbered 1-28.

Training the heart rate data of the testee in the previous n days according to the important characteristic of the phase of the peak point of the derivative of the daily average heart rate curve, and realizing the prediction of whether the (n + 1) th day is ovulated: the procedure and results were as follows:

1) data acquisition:

biomarker extraction is based on the nocturnal resting heart rate, a time series of 1 data point per day, as in example 1, the nocturnal resting heart rate refers to the average daily heart rate generated by taking the nocturnal sleep part from the minute-scale heart rate data and averaging the minute-scale data of the nocturnal sleep part.

2) Data preprocessing:

2-1) because of a plurality of reasons such as not uploading in time, the volunteer does not wear the bracelet in succession, the heart rate data has some missing values, screens incomplete data.

2-2) performing linear interpolation supplement on missing values which are continuously missing for less than five days, wherein the missing value supplement mode is not limited to a linear interpolation method, and the missing values can not be supplemented if the used filter supports the data containing the missing values.

2-3) directly filter above data and then get the numerical value of current day, the serious problem of data fluctuation can appear, therefore before filtering, data carry out preliminary smooth processing earlier: namely, the heart rate average value of the current day plus the current day for six days replaces the heart rate of the current day.

3) Data filtering:

the low frequency filter is used to filter or attenuate the high frequency components of the curve by a large amount to allow the low frequency components to pass. Here, a Butterworth low pass filter is used to obtain filtered data.

4) Derivation

The filtered heart rate is derived. The probability that the ovulation day is at the peak of the derivative is high, and since our goal is to predict the state of the next day, knowing the current day and previous data, the data of the next day is unknown, we use the forward derivation formula as follows:

where f (x2) represents the filtered heart rate for the current day, f (x)₀) And f (x1) represents the heart rate two days before the current day. The second derivative is also a very important feature, since the peak or valley of the derivative curve is determined, i.e. the range where the second derivative is zero is first sought.

5) Least squares fit

After derivation of the data, the previous 30 days including the current day are subjected to least square fitting, the function form is b0+ b1 · cos (2 pi · b2 · x + b3), the initial value b0 is the average value of the heart rate data of the 30 days, b1 is 1, b2 is 1/30, and b3 is 0, and if the current day is preceded by data of more than 45 days, the data of the previous 45 days are used for fitting. Because of the preliminary smoothing process, the current day data is replaced by the result of the multi-day averaging, so that the ovulation is not on the peak of the derivative curve, but is shifted to the left, and the predicted result also deviates from the range of true ovulation. Therefore, we further compensate for this deviation. And when the value of the periodic curve fitted from the current day to the previous 30 days is taken, the value of 7 days is taken to the right, and the curve is aligned with the last day of the derivative curve. The peak of the fitted curve r is exactly the ovulation area, so instead of the derivative of the heart rate, the curve consisting of the last point of the derivative of the r-curve (day-by-day fitted curve) is the second feature of the current day (called feature f2), instead of the second derivative of the heart rate, as a curve consisting of the last point of the r-curve (day-by-day fitted curve) as one feature of the current purpose (called feature f 1). In the function fitting process, the purpose of limiting undetermined parameters of the function is realized by setting a penalty function, so that the function fitted every time does not shake very severely, and the effect of smoother than the day-by-day derivative is achieved.

6) Model training and prediction:

and on the daily average heart rate derivative curve, taking the distance d from the current day to the previous valley point as the phase of the current day on the cycle, and if the current day is the peak point, indicating the phase of the peak point on the daily average heart rate derivative curve by the distance from the current day to the previous valley point.

The last 4 volunteers were selected as a test set (9, 10, 11, 12, 25, 26, 27, 28) from 1-12 volunteers and 13-28 volunteers with only one ovulation cycle for two or more consecutive cycles, all data had 60 cycles, the test set occupied 22 cycles, and the data of 38 cycles of the other 20 volunteers were used as a training set. After the series of data preprocessing, the process is repeated every time every day, the calculated d, f1 and f2 are used as the characteristics of the current day, and after all data are processed day by day, an n multiplied by 3 matrix is obtained to represent the characteristics of a total of n days. Taking the values of f1, f2 and d in five days before and after all real ovulation-marking days as characteristics, a KDE method is adopted to obtain a probability density model _ KDE of the three characteristics of d, f1 and f2, which is a common model commonly used for different volunteers.

7) An evaluation module:

if the output possibility index is probability1 ═ model _ kde1(x), x ∈ R^n×3，probability1∈R^n×1Inputting the feature x of the day i, the probability1(i) that the day i +1 is the ovulation day can be obtained. Setting a threshold k if the probability1(i) ≧ k and [ probability1(i)]' > 0 and [ viability 1(i-1)]' > 0, then prediction1(i + 1: i + t) ═ 2, namely from day i +1 to day i + t, marked as ovulation, t is defined as the ovulation period or the day of the duration of the menstrual period, and the next prediction starts from day i + 20; otherwise prediction1(i) ═ 1, i.e. prediction that day i +1 is not ovulation, next prediction starts from day i +2, and the prediction results for one subject are shown in fig. 11.

Example 4 prediction of menstrual period Using markers for menstruation

The data from subjects # 1-28 above were also used in menstruation prediction and the prediction flow chart is shown in FIG. 10.

According to the important characteristic that the valley point phase of the derivative of the daily average heart rate curve is used, the heart rate data of the testee in the previous n days is trained, and whether the (n + 1) th day is the first day of menstruation or not is predicted: the procedure and results were as follows:

1) data acquisition:

2) Data preprocessing:

3) Data filtering:

4) Derivation:

Where f (x2) represents the filtered heart rate for the current day, and f (x0) and f (x1) represent the heart rates two days prior to the current day. The second derivative is also a very important feature, since the peak or valley of the derivative curve is determined, i.e. the range where the second derivative is zero is sought first.

5) And (3) least square fitting:

after derivation of the data, the previous 30 days including the current day are subjected to least square fitting, the function form is b0+ b1 · cos (2 pi · b2 · x + b3), the initial value b0 is the average value of the heart rate data of the 30 days, b1 is 1, b2 is 1/30, and b3 is 0, and if the current day is preceded by data of more than 45 days, the data of the previous 45 days are used for fitting. Because of the preliminary smoothing process, the current day data is replaced by the result of the multi-day averaging, so that the ovulation is not on the peak of the derivative curve, but is shifted to the left, and the predicted result also deviates from the range of real menses. Therefore, we further compensate for this deviation. And when the value of the cycle curve fitted from the current day to the previous 30 days is taken, the value of 7 days is additionally taken rightwards, and then the curve is aligned with the last day of the derivative curve to fit the r curve. The trough of the fitted curve r is exactly the menstruation area, so that instead of the derivative of the heart rate, the curve consisting of the last point of the derivative of the r-curve (day-by-day fitted curve) is the second feature of the current day (called feature f2), instead of the second derivative of the heart rate, the curve consisting of the last point of the r-curve (day-by-day fitted curve) is taken as one feature of the current day (called feature f 1). In the function fitting process, the purpose of limiting undetermined parameters of the function is realized by setting a penalty function, so that the function fitted every time does not shake very severely, and the effect of smoother than the day-by-day derivative is achieved.

6) Model training and prediction

And on the daily average heart rate derivative curve, taking the distance d from the current day to the previous peak point as the phase of the current day on the cycle, and if the current day is the valley point, indicating the valley point phase on the daily average heart rate derivative curve by the distance from the current day to the previous peak point.

The last 4 volunteers were selected as test sets (9, 10, 11, 12, 25, 26, 27, 28) from 1-12 volunteers and 13-28 volunteers with only one ovulation cycle for two consecutive cycles or more, respectively. After the series of data preprocessing, the process is carried out again every time the time advances by one day, the calculated d, f1 and f2 are used as the characteristics of the current day, and after all data are processed by one day, an n x 3 matrix is obtained to represent the characteristics of a total of n days. Taking the values of f1, f2 and d in five days before and after the first day of all real menses as features, a probability density model _ KDE _ m of three features of d, f1 and f2 is obtained by adopting a KDE method, and the model is a common model which is generally used for different volunteers.

7) Evaluation module

If the output possibility index is probability2 ═ model _ kde1(x), x ∈ R^n×3，probability2∈R^n×1When the feature x of the day i is input, the probability2(i) that the day i +1 is the first day of menstruation can be obtained. Setting a threshold k if the mobility 2(i) ≧ k and [ mobility 2(i) ]' > 0 and [ basic 2(i-1)]' > 0, then prediction2(i + 1: i + t) ═ 2, i.e. from day i +1 to day i + t, marked as menstruation, t is defined as the number of days of duration of menstruation, and the next prediction starts from day i + 20; otherwise, prediction2(i) ═ 1, i.e., prediction is not performed on day i +1 and is not performed during the menstrual period, and prediction is performed from day i +2 next time, wherein the prediction results of one subject are shown in fig. 12.

It should be understood that the above examples are only for clarity of illustration and are not intended to limit the embodiments. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. And obvious variations or modifications of the invention may be made without departing from the spirit or scope of the invention.

Claims

1. An apparatus for predicting an ovulatory period and a menstrual period using a heart rate, comprising:

2. The apparatus for predicting the ovulatory period and the menstrual period using the heart rate as set forth in claim 1, wherein: the apparatus is at least one of an input device and a wearable device operatively attached to the computing device.

3. The apparatus for predicting the ovulatory period and the menstrual period using the heart rate as set forth in claim 2, wherein: the wearable device is a wearable device with a heart rate module capable of collecting continuous heart rates.

4. The apparatus for predicting the ovulatory and menstrual periods using the heart rate as set forth in claim 1, wherein the marker extracting module includes:

5. The apparatus for predicting the ovulatory and menstrual periods using the heart rate as claimed in claim 1, wherein said evaluation module comprises:

6. A method for predicting an ovulatory period and a menstrual period using a heart rate, comprising:

collecting continuous heart rate data of the subject using a device capable of collecting a continuous heart rate;

7. The method for predicting the ovulatory period and the menstrual period using the heart rate as set forth in claim 6, wherein: the method for acquiring the peak point phase of the daily average heart rate derivative curve and the valley point phase of the daily average heart rate derivative curve comprises the following steps:

and (3) performing derivation on the daily average heart rate curve, and calculating all peak point phases on the derivative curve and/or calculating all valley point phases on the derivative curve.

8. The method for predicting ovulatory and menstrual periods using heart rate according to claim 6, wherein: assessing an indicator of likelihood of the subject entering ovulation phase based on the first biomarker; and/or assessing an indicator of likelihood of the subject entering a menstruation period based on the second biomarker, comprising:

9. A biomarker for predicting the ovulatory and menstrual periods, characterized by: the biomarker for use in the device for predicting ovulation and a menstrual period using heart rate as claimed in any one of claims 1 to 5, the biomarker comprising a phase of a peak of a derivative curve of a daily average heart rate as a first biomarker for assessing entry of a subject into an ovulation period, and/or a phase of a valley of a derivative curve of a daily average heart rate as a second biomarker for assessing entry of a subject into a menstrual period.

10. Use of a biomarker according to claim 9 in the manufacture of a device for assessing an indicator of the likelihood of a subject entering an ovulatory and menstrual period, comprising the use of a first biomarker in the manufacture of a device for assessing an indicator of the likelihood of a subject entering an ovulatory period and/or the use of a second biomarker in the manufacture of a device for assessing an indicator of the likelihood of a subject entering a menstrual period.