JP2023129390A - REM sleep determination device, REM sleep determination method, and program - Google Patents

Abstract

To enhance determination accuracy on REM sleep in a non-restricted type rem sleep determination method.SOLUTION: A REM sleep determination device includes: a body motion data acquisition unit for acquiring body motion data indicating a size of a body motion in a sleep period at each time point; a body motion removal unit for removing first body motion data, which is the data corresponding to a body motion at a predetermined sleep stage, out of the body motion data; a body motion density moving average calculation unit for calculating body motion density indicating the number of times of body motions that have appeared in a first time of a predetermined length, and a moving average in a second time of a predetermined length from the body motion data in which the first body motion data is removed by the body motion removal unit; and a REM sleep determination unit for determining a period in which the body motion density calculated by the body motion density moving average calculation unit surpasses the moving average of the sleep period as a REM sleep period.SELECTED DRAWING: Figure 15

Description

The present invention relates to a REM sleep determination device, a REM sleep determination method, and a program.

According to the Ministry of Health, Labor and Welfare, more than 20% of adults in Japan currently suffer from some kind of sleep disorder. There is a close relationship between sleep and health, and poor sleep quality can lead to a variety of health problems, such as weakening the immune system, increasing the incidence of cognitive dysfunction, and increasing the risk of heart disease. Therefore, understanding the daily sleep state is important for detecting diseases and treating them through early detection.

In understanding sleep states, the cycle of sleep stages (sleep cycle) is one of the important indicators. Sleep stages are based on the Rechtschaffen & Kales (hereinafter referred to as the "R&K method") method, which is an international standard for determining sleep stages, and are divided into light sleep, deep sleep, wake (arousal), rapid eye movement (REM) sleep, and rapid eye movement (REM) sleep. Non-REM sleep is expressed in six stages, stages 1 to 4. The sleep cycle is a periodicity in which deep sleep (NREM sleep stages 3 to 4) and light sleep (NREM sleep stages 1 to 2, REM sleep) are repeated in a cycle of about 90 minutes, from falling asleep to REM sleep, Or during REM sleep. Therefore, sleep cycles can be measured by estimating REM sleep.

The current international standard sleep stage estimation method in medical practice is based on the R&K method and is estimated by evaluating biological data such as electroencephalograms, electromyograms, and eye movement diagrams obtained by overnight polysomnography (PSG) tests. However, PSG testing requires the attachment of multiple electrodes and sensors to the subject's head and body, and the restrained state places a heavy physical burden on the subject. Therefore, there is a high demand for an unrestrained sleep stage estimation method using a simple sensor such as a mat sensor. If a non-restrictive sleep stage estimation method is established, it will become possible to measure sleep states on a daily basis, leading to the prevention and early detection of diseases.

Currently, many unrestrained sleep stage estimation methods are being studied. In particular, methods that focus on REM sleep include RSSE (Realtime Sleep Stage Estimation), which is a method that focuses on physiological characteristics (Non-Patent Document 1), and Multi-Timescale REM Estimation (MTRE), which is based on machine learning. ) (Non-Patent Document 2) is known.

RSSE described in Non-Patent Document 1 focuses on the physiological characteristic that heart rate irregularly increases and decreases during REM sleep. In RSSE, as shown in equation (1), from the time when the rate of change between the median heart rate H ^Rrecent _med in the previous 5 minutes and the median heart rate HR ^prev _med 5 to 10 minutes ago increases by 0.04 or more, The time when the rate of change in heart rate once becomes negative and then becomes 0 or more is determined to be REM sleep.

On the other hand, MTRE described in Non-Patent Document 2 determines REM sleep based on machine learning. In MTRE, REM sleep is determined based on Random Forest (RF) using as input biological vibration data (complex vibration values of heartbeat, breathing, and body movement) obtained from a mat sensor. Specifically, we focused on the fact that there are multiple patterns of REM sleep characteristics, such as instantaneous and continuous, and analyzed the logarithmic spectra of biological vibration data for each window length (32 seconds, 64 seconds, 128 seconds, and 256 seconds). Use as input data. Learning is performed using different RFs using the input data, and REM sleep is determined. Thereafter, the number of REM sleep determinations for each RF is summed up by taking a window, and the epoch of REM sleep is determined so that the ratio of REM sleep is about 20% of the night's sleep time. Note that other sleep stages including non-REM sleep are determined in the same manner.

Tomohiro Harada, Fumito Uwano, Takahiro Komine, Yusuke Tajima, Takahiro Kawashima, Morito Morishima, Keiki Takadama, “Re” "Al-Time Sleep Stage Estimation from Biological Data with Trigonometric Function Regression Model", "AAAI Spring Symposium - Technical Report”, 2016, SS-16-01 Volume 07, Pages 348-353 Iko Nakari, Naoya Matsuda, Keiki Takadama, “REM Estimation Based on Combination of Multi-Timescale Estimations and Automat ic Adjustment of Personal Bio-vibration Data of Mattress Sensor”, “Proceedings of the AAAI 2022 Spring Symposium “How Fair” Is it fair? Achieving Wellbeing AI”, 2022, Volume 3276, Pages 74-80 Yuta Kambayashi, Kei Hagiwara, “Attempt to estimate sleep cycle using the frequency of body movements,” “Biomedical Engineering”, 2012, Vol. 50, No. 1, pp. 99-104

However, the methods described in RSSE described in Non-Patent Document 1 and MTRE described in Non-Patent Document 2 do not take into account the periodicity of sleep. Therefore, there is a problem in that even data without sleep periodicity is determined to be REM sleep as long as the determination conditions are met. On the other hand, as a method to estimate the sleep cycle with emphasis on periodicity, we focus on body movements that occur periodically during light sleep before and after REM sleep, and calculate the number of body movements by counting the number of body movements in 30-minute windows. A sleep cycle estimation method based on body movement density has been studied (Non-Patent Document 3).

The sleep cycle estimation method described in Non-Patent Document 3 focuses on body movements that occur periodically during light sleep before and after REM sleep, and estimates sleep cycles from body movement information during sleep acquired using an infrared motion sensor. do. In this sleep cycle estimation method, body movement density was introduced as an index representing the frequency of body movements in order to capture the periodicity of body movements that occur. Body movement density is the number of body movements that occurred in a total of 30 minutes, 15 minutes before and after.

In the sleep cycle estimation method, body movement density is first derived using body movement data during sleep acquired from an infrared motion sensor. FIG. 23 shows an example of a body movement and a body movement density calculated from the body movement. The vertical axis shows the number of body movements, and the horizontal axis shows the elapsed time.
In Non-Patent Document 3, "the time when the body movement density is convex upward" is the time when the value of the body movement density continuously exceeds the average value, and the time when the value of the body movement density is continuously below the average value. He describes the similarity to the sleep cycle by defining a set of ``times when the body movement density is downward convex'' as a body movement density cycle.

It is possible to estimate REM sleep by applying the sleep cycle estimation method based on body movement density described in Non-Patent Document 3 above. However, the following two problems exist in applying this sleep cycle estimation method to REM sleep estimation. The first problem is that light sleep determined by body movement density includes body movements due to wakefulness in addition to body movements during REM sleep. The second problem is that the magnitude of body movements in REM sleep and NREM sleep stage 2 are very similar, and the boundaries are unclear. Because of these problems, it is thought that sufficient determination accuracy cannot be obtained by simply applying the sleep cycle estimation method to REM sleep estimation.

As described above, the conventional unrestricted method for determining REM sleep does not have sufficient determination accuracy. There is a need to improve the accuracy of REM sleep determination in unrestrained REM sleep determination methods.

The present invention has been made in view of the above points, and provides a REM sleep determination device, a REM sleep determination method, and a program that can improve the accuracy of REM sleep determination in an unrestricted REM sleep determination method. .

The present invention has been made to solve the above problems, and one aspect of the present invention includes a body movement data acquisition unit that acquires body movement data that indicates the magnitude of body movement during a sleeping period at each time. , a body movement removing unit that removes first body movement data that is data corresponding to body movements in a predetermined sleep stage from the body movement data; and the body movement removing unit removes the first body movement data. body movement density that indicates the number of body movements that occurred during a first time period of a predetermined length and a moving average per second time period of a predetermined length from the body movement data; a moving average calculation unit; a REM sleep determination unit that determines a period of the sleep period in which the body movement density calculated by the body movement density moving average calculation unit exceeds the moving average as a period of REM sleep; This is a REM sleep determination device.

Further, in one aspect of the present invention, in the REM sleep determination device described above, the first body movement data includes body movement data during awakening, which is data corresponding to body movements during awakening.

Further, in one aspect of the present invention, in the above REM sleep determining device, the body movement removing unit may determine whether the body movement during wakefulness is determined based on the magnitude of the body movement indicated by the body movement data from among the body movement data. Remove dynamic data.

Further, in one aspect of the present invention, in the above REM sleep determination device, the body movement removing unit may select a size of the body movement indicated by the body movement data from among the body movement data at a predetermined top rate. The waking body movement data is removed based on certain data, and the ratio is determined from the ratio of waking to the entire sleep.

Further, in one aspect of the present invention, in the above REM sleep determination device, the body movement removing unit removes the waking body movement data from the body movement data based on a result of machine learning.

Further, in one aspect of the present invention, in the REM sleep determination device, NREM sleep stage 2 body movement data, which is data corresponding to body movements in NREM sleep stage 2, is included together with the waking body movement data, The motion removing unit removes the non-REM sleep stage 2 body motion data from the body motion data from which the waking body motion data has been removed.

Further, in one aspect of the present invention, in the REM sleep determination device described above, the first body movement data includes NREM sleep stage 2 body movement data that is data corresponding to body movements in NREM sleep stage 2.

Further, in one aspect of the present invention, in the above REM sleep determination device, the body movement removing unit may determine the non-REM sleep stage based on the magnitude of the body movement indicated by the body movement data from among the body movement data. Remove two-body motion data.

Further, in one aspect of the present invention, in the above REM sleep determination device, the body movement removing unit may select a size of the body movement indicated by the body movement data from among the body movement data at a predetermined top rate. The NREM sleep stage 2 body movement data is removed based on certain data, and the ratio is determined from the ratio of NREM sleep stage 2 to the entire sleep.

Further, in one aspect of the present invention, in the above REM sleep determination device, the body movement removing unit removes the non-REM sleep stage 2 body movement data from the body movement data based on a result of machine learning.

Further, in one aspect of the present invention, in the above REM sleep determination device, the first body movement data includes body movement during awakening which is data corresponding to body movement during awakening together with the non-REM sleep stage 2 body movement data. The body movement removing unit removes the waking body movement data from the body movement data from which the non-REM sleep stage 2 body movement data has been removed.

Further, in one aspect of the present invention, in the above REM sleep determination device, the body movement density moving average calculation unit calculates the moving average using a time window of a predetermined time length determined based on a sleep cycle. calculate.

Further, one aspect of the present invention includes a body movement data acquisition step of acquiring body movement data indicating the magnitude of body movement during a sleep period at each time, and a body movement data acquisition step of acquiring body movement data indicating the magnitude of body movement during a sleep period, and determining body movement data at a predetermined sleep stage from among the body movement data. a body movement removing step of removing first body movement data that is data corresponding to , and a first time period of a predetermined length from the body movement data from which the first body movement data has been removed by the body movement removing step. a body movement density moving average calculation step of calculating a body movement density indicating the number of body movements that occurred during the period, and a moving average of a predetermined length per second time; A REM sleep determining method includes a REM sleep determining step of determining a period during which the body movement density calculated in the moving average calculating step exceeds the moving average as a period of REM sleep.

Further, one aspect of the present invention includes a body movement data acquisition step of acquiring body movement data indicating the magnitude of body movement during a sleep period at each time, and a body movement data acquisition step of acquiring body movement data indicating the magnitude of body movement during a sleep period, and determining body movement data at a predetermined sleep stage from among the body movement data. a body movement removing step of removing first body movement data that is data corresponding to , and a first time period of a predetermined length from the body movement data from which the first body movement data has been removed by the body movement removing step. a body movement density moving average calculation step of calculating a body movement density indicating the number of body movements that occurred during the period, and a moving average of a predetermined length per second time; The present invention is a program for executing a REM sleep determining step of determining a period during which the body movement density calculated in the dynamic density moving average calculating step exceeds the moving average as a period of REM sleep.

According to the present invention, it is possible to improve the accuracy of REM sleep determination in an unrestricted REM sleep determination method.

FIG. 1 is a diagram illustrating an example of an outline of a REM sleep determination method according to an embodiment of the present invention. FIG. 3 is a diagram showing an example of the relationship between sleep stages and the magnitude of body movement according to an embodiment of the present invention. FIG. 6 is a diagram showing an example of body movements extracted by the process of removing body movements during non-REM sleep according to the embodiment of the present invention. FIG. 4 is a diagram showing the difference in magnitude between body movements during wakefulness and body movements during REM sleep according to an embodiment of the present invention. FIG. 6 is a diagram illustrating an example of a body movement extracted by a body movement removal process during wakefulness according to an embodiment of the present invention. FIG. 3 is a diagram showing an example of a 30-minute window according to an embodiment of the present invention. It is a figure showing an example of a moving average concerning an embodiment of the present invention. FIG. 3 is a diagram showing an example of REM sleep determination using the REM sleep determination method according to the embodiment of the present invention. FIG. 3 is a diagram illustrating an example of a conceptual diagram showing the number of body movements with respect to elapsed time according to an embodiment of the present invention. FIG. 3 is a conceptual diagram showing the number of body movements relative to the elapsed time after body movements during non-REM sleep stage 2 and body movements during wakefulness are removed according to an embodiment of the present invention. FIG. 2 is a conceptual diagram showing body movement density according to an embodiment of the present invention. FIG. 3 is a conceptual diagram showing a difference in body movement density for each REM sleep according to an embodiment of the present invention. FIG. 2 is a conceptual diagram showing a case where REM sleep is determined using a fixed threshold value according to an embodiment of the present invention. FIG. 2 is a conceptual diagram showing a case where REM sleep is determined using a moving average as a threshold according to an embodiment of the present invention. 1 is a diagram showing an example of the configuration of a REM sleep determination device 1 according to an embodiment of the present invention. FIG. 3 is a diagram illustrating an example of the flow of REM sleep determination processing according to an embodiment of the present invention. FIG. 3 is a diagram showing the results of REM sleep determination using the REM sleep determination method according to the embodiment of the present invention. It is a figure showing the result of each test subject of the 1st evaluation index concerning an example of the present invention. It is a figure which shows the result of each test subject of the 2nd evaluation index based on the Example of this invention. It is a bar graph showing the average value and standard deviation of the results of the first evaluation index according to the example of the present invention. It is a bar graph which shows the average value and standard deviation of the result of the 2nd evaluation index based on the Example of this invention. It is a figure which shows an example of the REM sleep determination result by the REM sleep determination method based on the Example of this invention, and the conventional method (RSSE, MTRE). It is a figure which shows an example of the body movement density based on a prior art.

(Embodiment)
Embodiments of the present invention will be described in detail below with reference to the drawings.
In this embodiment, sleep stages are classified into a total of six stages from light sleep to deep sleep based on the Rechtschaffen & Kales (hereinafter abbreviated as "R&K method") method. The six stages are arousal, rapid eye movement (REM) sleep, and non-REM sleep stages 1 to 4.
In the following explanation, among the NREM sleep stages 1 to 4, NREM sleep stage 2, NREM sleep stage 3, and NREM sleep stage 4 are treated without distinction, and may be referred to as NREM sleep stage 2.

FIG. 1 is a diagram illustrating an example of an overview of the REM sleep determination method according to the present embodiment. The REM sleep determination method includes removing non-REM sleep (step S30), removing wakefulness (step S40), counting the number of body movements (w minute window, w is a predetermined natural number), and calculating a moving average (z minute window, z is a predetermined natural number) (step S50).

In the REM sleep determination method, for example, two types of data are used as input data. The first data is body movement data for one night. The second data is the non-REM sleep stage 2 determination result.
Body movement data is obtained by extracting high frequency components from vibration values of body movement obtained from a biosensor during one night's sleep.
The NREM sleep stage 2 determination result is a result obtained by performing a NREM sleep stage 2 determination using machine learning (for example, Random Forest (RF)) using biological vibration data during sleep as input. Here, the biological vibration data is acquired from the biological sensor during one night's sleep. The biological vibration data includes a composite vibration value of a heartbeat vibration value, a breathing vibration value, and a body movement vibration value.
In the REM sleep determination method, the above body movement data and the non-REM sleep stage 2 determination result are input, and the REM sleep determination result is output.

As for the flow of the determination, one night's body movement data and the non-REM sleep stage 2 determination result are received as input. Thereafter, body movements during times determined to be non-REM sleep stage 2 by machine learning (RF) are removed (step S10). Thereafter, body movements for y seconds before and after the body movement of the top x% in size are removed (step S20). Body movements related to REM sleep are extracted in the order of step S10 and step S20.

Thereafter, the body movement density is calculated for the processed body movement data in 30 minute windows. Here, the body movement density is the number of body movements that occur during a predetermined period of time, and is calculated at each time. When calculated in a 30-minute window, the body movement density at a certain time is the number of body movements that occur in a total of 30 minutes, 15 minutes before and after the time.

A moving average is derived for the calculated body movement density in a z-minute window, and a time period in which the body movement density exceeds the moving average is determined to be REM sleep (step S30). Here, by determining the body movement density in a 30-minute window, it is possible to suppress the influence of erroneous determination of NREM sleep determined by machine learning, so that erroneous determination of NREM sleep determined by machine learning is tolerable to some extent.

The flow of the REM sleep determination method according to the present embodiment is based on the natural idea that if only body movements during REM sleep are left, REM sleep can be estimated based on the frequency of appearance of the body movements. In the REM sleep determination method, steps S10 and S20 are performed in order to separate and preserve body movements during REM sleep. The process of step S30 is performed to identify REM sleep from the derivation of the appearance frequency of body movements and the height of the appearance frequency.
The details of each process will be explained below.

[Removal of body movements during non-REM sleep]
The REM sleep determination method according to the present embodiment focuses on the physiological characteristic that body movements occur frequently during light sleep such as REM sleep. Here, FIG. 2 shows the relationship between sleep stages and the magnitude of body movements. As shown in FIG. 2, body movements during NREM sleep stage 2 are very similar in size to body movements that occur near REM sleep. Therefore, it is difficult to eliminate body movements in non-REM sleep stage 2 based on the characteristics of the size of body movements. Therefore, the time period of non-REM sleep stage 2 is determined using machine learning. By removing all body movements during the time period determined to be non-REM sleep stage 2, only body movements related to REM sleep are extracted.

FIG. 3 shows an example of body movements extracted by the process of removing body movements during non-REM sleep described above. FIG. 3 shows sleep stages based on body movement data, overnight polysomnography (PSG), and a range determined to be non-REM sleep stage 2 by machine learning. The vertical axis on the left axis indicates the magnitude of body movement. The vertical axis on the right axis indicates sleep stages. The horizontal axis indicates elapsed time. Looking at the period T1 to T5, it can be seen that body movements in the range determined to be NREM sleep stage 2 have been removed, although there are some parts that differ from the results with PSG.

[Removal of body movements during awakening]
As mentioned above, the REM sleep determination method focuses on the physiological characteristic that body movements occur frequently during light sleep, including REM sleep. It also includes body movements caused by arousal. As shown in Figure 4, there is a clear difference in the size of body movements during wakefulness and body movements during REM sleep. Focusing on this feature, we remove the influence of body movements due to arousal from the characteristics of the size of body movements.

First, body movements in the top x% of the body movement data for one night are removed because they are likely to be body movements due to awakening. Since the influence of a larger body movement continues for a certain period of time, body movements y seconds before and after the removed body movement are also removed. FIG. 5 shows an example of body movements extracted by the above process of removing body movements during wakefulness. FIG. 5 shows sleep stages based on body movement data and PSG. The vertical axis on the left axis indicates the magnitude of body movement. The vertical axis on the right axis indicates sleep stages. The horizontal axis indicates elapsed time.

Body movements BM1 and body movements BM2 are body movements whose magnitudes are in the top x% of the overnight body movement data. Range T61 and range T62 are ranges of y (=60) seconds before and after body movement BM1, which is the top x (=1)% of body movements. Range T71 and range T72 are ranges of y (=60) seconds before and after body movement BM2, which is the top x (=1)% of body movements. All body movements included in these ranges T61 and T62, and ranges T71 and T72 are removed.

[Derivation of body movement density and REM sleep determination]
For the processed body movement data, body movement density is calculated in a w minute window (15 minutes forward and backward, slide width 1 second). A moving average of a z-minute window (z/2 minutes forward and backward, slide width 1 second) is calculated for the calculated body movement density, and the time when the body movement density exceeds the moving average is determined to be REM sleep. The w-minute window is, for example, a 30-minute window. The z-minute window is, for example, a 90-minute window.

Here, FIG. 6 shows an example of a 30-minute window. FIG. 7 shows an example of a moving average. FIG. 6 shows sleep stages based on the number of body movements (body movement density) and PSG. FIG. 7 shows sleep stages based on the number of body movements (body movement density), moving average, and PSG. The vertical axis on the left axis indicates the number of body movements (body movement density). The vertical axis on the right axis indicates sleep stages. The horizontal axis indicates elapsed time. Note that in FIGS. 6 and 7, the scale indicating the elapsed time is different from that in FIGS. 3 and 5 described above.

The 30-minute window W1 shown in FIG. 6 is a 30-minute window at time P1. That is, as the body movement density at time P1, the number of body movements that occurred during a total of 30 minutes, 15 minutes before and after time P1, is counted. A 90-minute window W2 shown in FIG. 7 is a 90-minute window at time P2. The moving average at time P2 is a moving average for 45 minutes before and after time P2, for a total of 90 minutes.

FIG. 8 shows an example of REM sleep determination using the REM sleep determination method. FIG. 8 shows sleep stages based on body movement density, moving average, and PSG. The vertical axis on the left axis indicates the number of body movements (body movement density). The vertical axis on the right axis indicates sleep stages. The horizontal axis indicates elapsed time.

The range T81 to the range T86 each indicate a range in which the body movement density exceeds the moving average (broken line) of the z-minute window. The range T81 to T86 is REM sleep determined by the REM sleep determination method. The range T91 to T95 indicates actual REM sleep. The similarity between REM sleep determined using the REM sleep determination method and actual REM sleep can be seen.

Here, with reference to FIGS. 9 to 14, the reason for using a moving average as a threshold value for REM sleep determination will be explained.
Body movement density may vary greatly from subject to subject or from one REM sleep state to another. Therefore, the threshold value for REM sleep determination is not a fixed value, but a moving average of z minutes. It is considered that times when the density of body movement is high enough to exceed the moving average are likely to be REM sleep. Based on this basis, the time period in which the body movement density exceeds the moving average is determined to be REM sleep.

9 to 14 are conceptual diagrams showing the number of body movements relative to elapsed time. FIG. 9 shows body movements during wakefulness, body movements during REM sleep, and body movements during stage 2 of non-REM sleep. Furthermore, range T101, range T102, and range T103 each indicate actual REM sleep.

In FIG. 10, body movements during NREM sleep stage 2 and body movements during awakening are removed by removing body movements during NREM sleep stage 2 and during awakening, respectively. On the other hand, there are body movements that are erroneously removed from among the body movements of REM sleep as well as body movements of non-REM sleep stage 2 and wakefulness. REM sleep body movement B101 and REM sleep body movement B102 indicate body movements of REM sleep that were mistakenly removed.

FIG. 11 shows the body movement density C1 calculated using a time window. As mentioned above, body movement density may vary greatly depending on REM sleep. In FIG. 12, the body movement density in range T102 is lower than the body movement density in range T101 and range T103. Therefore, as shown in FIG. 13, when the fixed value threshold TH1 is used as the threshold for REM sleep determination, the range T102 will not be determined as REM sleep.

FIG. 14 shows a case where REM sleep is determined using the moving average MA1 as a threshold. When moving average MA1 is used as a threshold, range T102 is determined to be REM sleep. In other words, if the range in which the body movement density C1 exceeds the moving average MA1 is determined as REM sleep, it is possible to reduce the amount of REM sleep that is missed even if there is a large difference in body movement density from one REM sleep to another. I can do it.

Further, as shown in FIG. 10, body movements during REM sleep may be mistakenly removed. Even in that case, if the moving average is used for determination, there is a possibility that the range of body movements in REM sleep that was once mistakenly removed can be correctly determined as REM sleep.

[Configuration of REM sleep determination device 1]
The REM sleep determination method described above is executed by the REM sleep determination device 1, for example.
FIG. 15 is a diagram showing an example of the configuration of the REM sleep determination device 1 according to the present embodiment. The REM sleep determination device 1 is a personal computer (PC). The REM sleep determination device 1 includes a control section 2 and a storage section 3.

The control unit 2 includes a body movement data acquisition unit 20, a non-REM sleep body movement determination unit 21, a body movement removal unit 22, a body movement density moving average calculation unit 23, a REM sleep determination unit 24, and an output unit 25. Equipped with Each of these functional units is realized, for example, by a CPU reading a program from a ROM (Read Only Memory), loading it into a RAM (Random Access Memory), and executing processing according to the program. The ROM and the RAM are included in the storage unit 3.

The body movement data acquisition unit 20 acquires body movement data A1. Body movement data A1 is data that indicates the magnitude of body movement during the sleep period for each time. In the present embodiment, as an example, the body movement data acquisition unit 20 first acquires biological vibration data A2 from the biological vibration data measurement sensor 4. The biological vibration data A2 is data that indicates, at each time, a vibration value that is a combination of the magnitude of body movement, the magnitude of heartbeat, and the magnitude of respiration during the period of sleep. The body movement data acquisition unit 20 acquires body movement data A1 by extracting a high frequency component from the vibration value of the magnitude of body movement included in the acquired biological vibration data A2.

The non-REM sleep body movement determination unit 21 determines the period of non-REM sleep stage 2 among the sleep periods based on machine learning from the biological vibration data A2 acquired by the body movement data acquisition unit 20.
The machine learning used by the NREM sleep body movement determination unit 21 to determine NREM sleep stage 2 is supervised machine learning. The trained model M1 is a supervised machine learning model that has been trained in advance. When the learned model M1 receives the biological vibration data A2, it outputs the period of non-REM sleep stage 2 among the sleep periods.

Note that, when the body movement data A1 is input as the learned model M1, a learned model that outputs the period of non-REM sleep stage 2 among the sleep periods may be used. In that case, the body movement data acquisition unit 20 only needs to acquire body movement data from the biological vibration data A2.

In this embodiment, the machine learning used by the NREM sleep body movement determination unit 21 to determine NREM sleep stage 2 is, for example, the REM sleep determination method (Multi-Timescale REM Estimation: MTRE) described in Non-Patent Document 2. . Note that the machine learning may be RF other than MTRE, or may be supervised machine learning other than RF.

The body movement removal unit 22 removes the first body movement data from the body movement data A1 acquired by the body movement data acquisition unit 20. The first body movement data is data corresponding to body movements in a predetermined sleep stage.

In this embodiment, the first body movement data removed by the body movement removal unit 22 includes NREM sleep stage 2 body movement data. The NREM sleep stage 2 body movement data is, for example, data corresponding to body movements in NREM sleep stage 2. Note that the NREM sleep stage 2 body movement data may be data corresponding to body movements from NREM sleep stage 2 to NREM sleep stage 4, since the ratio of NREM sleep stages 3 and 4 is small and similar. . Further, the NREM sleep stage 2 body movement data may be data corresponding to body movements in NREM sleep stage 2 and NREM sleep stage 3.

Furthermore, in the present embodiment, the first body movement data removed by the body movement removal unit 22 includes body movement data during wakefulness. Awakening body movement data is data corresponding to body movement during awakening.

The body movement removing unit 22 removes at least one of the non-REM sleep stage 2 body movement data and the waking body movement data from the body movement data A1. When removing both the non-REM sleep stage 2 body movement data and the waking body movement data from the body movement data A1, the body movement removal unit 22 removes one and then the other. The body movement removal unit 22 may remove the waking body movement data from the body movement data A1 from which the non-REM sleep stage 2 body movement data has been removed, or may remove the body movement data from which the waking body movement data has been removed. The NREM sleep stage 2 body motion data may be removed from A1.

The method by which the body movement removal unit 22 removes the first body movement data from the body movement data A1 is based on the magnitude of the body movement indicated by the body movement data A1, or based on machine learning. Any method may be used. There are two methods for removing data based on the size of body movement: a method for removing data based on a threshold value for the size of body movement, and a method for removing data based on a predetermined percentage of body movement from body movement data A1. There is a method of removing based on Specific examples of these methods will be described later.

For example, based on the determination result of NREM sleep stage 2 by the NREM sleep body movement determining unit 21, the body movement removing unit 22 selects two NREM sleep stages from the body movement data A1 acquired by the body movement data acquisition unit 20. Remove dynamic data. Therefore, the body movement removal unit 22 removes the NREM sleep stage 2 body movement data from the body movement data A1 acquired by the body movement data acquisition unit 20 based on the result of machine learning.

For example, the body movement removing unit 22 removes the waking body movement data from the body movement data A1 based on the magnitude of the body movement indicated by the body movement data A1. In the present embodiment, as an example, the body movement removal unit 22 removes the waking body movement data from the body movement data A1 from which the non-REM sleep stage 2 body movement data has been removed. In addition, the body movement removing unit 22 removes the waking body movement data from the body movement data A1 based on the magnitude of the body movement indicated by the body movement data A1, for example.

The body movement density moving average calculation unit 23 calculates the body movement density and the moving average from the body movement data A1 from which the first body movement data has been removed by the body movement removal unit 22. The body movement density indicates the number of body movements that occur during a first period of predetermined length (for example, 30 minutes). The moving average is a moving average per second period of predetermined length (90 minutes, for example).

The REM sleep determination unit 24 determines a period of sleep in which the body movement density calculated by the body movement density moving average calculation unit 23 exceeds the moving average as a period of REM sleep.

The output unit 25 outputs the determination result by the REM sleep determination unit 24.

The biological vibration data measurement sensor 4 measures biological vibration data A2. The biological vibration data measurement sensor 4 is the biological sensor described above.
The biological vibration data measurement sensor 4 is a mat-type sensor placed under a bed mat on which the subject is placed. The biological vibration data measurement sensor 4 is a non-constraint type sensor.

[REM sleep determination processing]
Next, with reference to FIG. 16, the flow of the process (REM sleep determination process) based on the REM sleep determination method by the REM sleep determination apparatus 1 will be described.
FIG. 16 is a diagram illustrating an example of the flow of REM sleep determination processing according to the present embodiment. The REM sleep determination process is executed by the control unit 2.

Step S110: The body movement data acquisition unit 20 acquires the body movement data A1. As an example, the body movement data acquisition unit 20 first acquires biological vibration data A2 from the biological vibration data measurement sensor 4. The body movement data acquisition unit 20 acquires body movement data A1 by extracting a high frequency component from the vibration value of the magnitude of body movement included in the acquired biological vibration data A2.

Step S120: The body movement removal unit 22 removes the non-REM sleep stage 2 body movement data from the body movement data A1 acquired by the body movement data acquisition unit 20 based on the result of machine learning. The result of machine learning is the determination result of NREM sleep stage 2 by the NREM sleep body movement determination unit 21.

Note that the determination result of the NREM sleep stage 2 may be determined by the NREM sleep body movement determination section 21 and stored in the storage section 3 in advance before the start of the REM sleep determination process. Note that in step S120, the body movement data acquisition section 20 may acquire the biological vibration data A2, and the NREM sleep body movement determination section 21 may determine the NREM sleep stage 2 based on machine learning.
Note that the process in step S120 corresponds to the process in step S10 described above.

Note that the NREM sleep body movement determining unit 21 may determine the period of the NREM sleep stage 2 based on the magnitude of the body movement indicated by the body movement data A1 instead of using machine learning. For example, the NREM sleep body movement determination unit 21 may determine a period in which the magnitude of body movement is smaller than a predetermined threshold value as the period of the NREM sleep stage 2.

Step S130: The body movement removing unit 22 removes the waking body movement data from the body movement data A1 based on the magnitude of the body movement indicated by the body movement data A1. In the example shown in FIG. 16, the body movement removing unit 22 removes the waking body movement data from the body movement data A1 from which the NREM sleep stage 2 body movement data has been removed by the body movement removing unit 22 in step S120.

Furthermore, the body movement removing unit 22 removes the waking body movement data from the body movement data A1 based on the magnitude of the body movement indicated by the body movement data A1. Here, the body movement removal unit 22 removes data as waking body movement data based on data in which the magnitude of the body movement indicated by the body movement data A1 is at a predetermined higher rate from among the body movement data A1. The body movement removal unit 22 removes a predetermined range before and after the time when the data at the predetermined rate was measured as waking body movement data.

Preferably, the predetermined ratio is determined based on the ratio of wakefulness to total sleep. It is known that awakening occurs in 5 to 15% of total sleep. Therefore, the predetermined ratio is, for example, a range of 60 seconds before and after a body movement that is in the top 1% of body movements. Note that the ratio of the predetermined value may be set for each subject. Moreover, the ratio of the predetermined value may be set without being based on the ratio of wakefulness to total sleep.

The body movement removing unit 22 removes, from the body movement data A1, data in which the size of the body movement indicated by the body movement data A1 is equal to or larger than a predetermined size (threshold value) as waking body movement data. Good too.

Here, the magnitude of the body movement differs depending on the sleep stage, as shown in FIG. 4. The magnitude of body movement increases in the order of wakefulness, NREM sleep stage 1, REM sleep, NREM sleep stage 2, NREM sleep stage 3, and NREM sleep stage 4. The threshold value may be set, for example, as the magnitude of body movement corresponding to the boundary between NREM sleep stage 1 and REM sleep. Further, the threshold value may be set, for example, as the magnitude of body movement corresponding to the boundary between wakefulness and non-REM sleep stage 1.
Note that the process in step S130 corresponds to the process in step S20 described above.

Step S140: The body movement density moving average calculation unit 23 calculates the body movement density and the moving average from the body movement data A1 from which the waking body movement data has been removed by the body movement removal unit 22. The length of the time window used to calculate body movement density is 30 minutes. The length of the time window used to calculate the moving average is preferably determined based on the sleep cycle. The length of the time window used to calculate the moving average is 90 minutes. Therefore, the body movement density moving average calculation unit 23 calculates a moving average using a time window of a predetermined length determined based on the sleep cycle.

Step S150: The REM sleep determination unit 24 determines a period of sleep in which the body movement density calculated by the body movement density moving average calculation unit 23 exceeds the moving average as a period of REM sleep.
Note that the processing in step S140 and step S150 corresponds to the processing in step S30 described above.

Step S160: The output unit 25 outputs the determination result by the REM sleep determination unit 24. The output unit 25 outputs the determination result to, for example, a display device (not shown), and causes the display device to display the determination result. Note that the output unit 25 may output the determination result to an external information processing device (such as a server). Further, the output unit 25 may cause the storage unit 3 to store the determination result.
With this, the REM sleep determination device 1 ends the REM sleep determination process.

Note that the body movement removing unit 22 may obtain the determination result of the non-REM sleep stage 2 from an external device other than the REM sleep determination device 1. The external device has the same function as the non-REM sleep body movement determining section 21. In that case, the non-REM sleep body movement determination section 21 may be omitted from the configuration of the REM sleep determination device 1.

In the present embodiment, an example has been described in which the waking body movement data is removed from the body movement data A1 from which the NREM sleep stage 2 body movement data has been removed, but the present invention is not limited to this. The order in which the non-REM sleep stage 2 body movement data and the waking body movement data are removed may be reversed. That is, the non-REM sleep stage 2 body movement data may be removed from the body movement data A1 from which the waking body movement data has been removed. In that case, the process of step S120 is executed after the process of step S130.

Furthermore, the process of removing the NREM sleep stage 2 body movement data may be omitted. In that case, the body movement removal unit 22 removes the waking body movement data from the body movement data A1 acquired by the body movement data acquisition unit 20. In this case, the non-REM sleep body movement determining unit 21 and the body movement removing unit 22 may be omitted from the configuration of the REM sleep determining device 1.

Note that, in the present embodiment, an example has been described in which the body movement removing unit 22 removes the waking body movement data based on the magnitude of the body movement, but the present invention is not limited to this. The body movement removal unit 22 may remove the waking body movement data from the body movement data A1 based on machine learning. In this machine learning, a learned model is used that outputs the range of arousal when body movement data A1 is input.

In addition, in this embodiment, an example in which NREM sleep stage 2 body motion data is removed based on machine learning has been described, but the present invention is not limited to this. The body movement removing unit 22 may remove the non-REM sleep stage 2 body movement data from the body movement data A1 based on the magnitude of the body movement indicated by the body movement data A1. Alternatively, the non-REM sleep stage 2 body movement data may be removed from the body movement data A1 based on data in which the magnitude of the body movement indicated by the body movement data A1 is at a predetermined upper level ratio. Here, the ratio is determined from the ratio of non-REM sleep stage 2 to the total sleep.

In this embodiment, before the process of calculating the body movement density and the moving average (step S140), the NREM sleep stage 2 body movement data and the waking body movement data are removed from the body movement data A1 in advance. Although an example of the case where the process is performed (step S120 and step S130) has been described, the present invention is not limited to this. Either one of the non-REM sleep stage 2 body movement data and the waking body movement data is not removed from the body movement data A1, and the process of calculating the body movement density and the moving average (step S140) is executed. Good too. That is, either one of the processes of step S120 and step S130 may be omitted from the REM sleep determination process shown in FIG. 16.

Note that the REM sleep determination device 1 may be realized as a virtual server. Each functional unit included in the REM sleep determination device 1 may be distributed and provided in a plurality of servers. The REM sleep determination device 1 may be implemented as a cloud server.

(Example)
As an example using the REM sleep determination method according to the present embodiment, the results of an experiment will be described.
In this example, eight subjects of different ages and genders were used as subjects. Based on the heart rate data and body movement data measured by the first biosensor and the biovibration data measured by the second biosensor, the conventional methods (RSSE, MTRE) and the present embodiment are applied to the subject. The REM sleep estimation results will be compared with the REM sleep determination method according to the present invention.

A subject experiment was conducted with each subject wearing a PSG and two types of biosensors (a first biosensor and a second biosensor) placed under the bed mat. Each of the two types of biosensors is a mat type sensor. These two types of biosensors are used in a state where the two types of sensors are stacked on top of each other at positions directly below the chest of the subject. The two types of biosensors mainly acquire chest vibrations. The first biosensor outputs one sensor value per second, and the second biosensor outputs 16 sensor values per second. This vibration includes breathing, heartbeat, body movement, and noise.

The comparison method (RSSE) uses data acquired by the first biosensor. The comparison method (MTRE) uses data acquired by the second biosensor. In this embodiment, data from both biosensors is used. The first biological sensor and the second biological sensor are examples of the biological vibration data measurement sensor 4.

In addition, the acquired PSG data was asked to have a specialized engineer analyze the sleep stages, and the obtained sleep stages were used as correct data. The age groups of the subjects were 2 in their 20s, 1 in their 30s, 4 in their 40s, and 1 in their 50s. Regarding gender, 5 were male and 3 were female. Although the proportions are uneven, we selected subjects to cover all age groups (20s to 50s) and genders.

In addition, regarding polysomnography (PSG), changes in the sleeping environment place a large burden on the subject, and when the purpose is to analyze sleep structure and sleep indicators, it is necessary to perform polysomnography after the second or third night rather than the first night. There is a view that it is better to adopt the records of Therefore, in this example, all data from the second night onwards are used.

As evaluation indicators for sleep stage estimation, Accuracy, Precision, Recall, F-measure, etc. for each epoch (30 seconds) are generally used to confirm the accuracy of estimation. However, in order to estimate the sleep cycle, it is important to be able to estimate REM sleep even roughly, rather than the second-by-second accuracy of REM sleep estimation. Therefore, in this embodiment, two evaluation indicators, a first evaluation index and a second evaluation index, are used. The first evaluation index is the rate at which the REM sleep estimated by the REM sleep determination method according to the present embodiment and the conventional methods (RSSE, MTRE) captures actual REM sleep. Second evaluation index: False positive (FP) (determination as REM sleep when it is not actually REM sleep) is the number of times.

If the estimated REM sleep temporally overlaps with the PSG-based REM sleep, it is defined that the REM sleep has been captured.
Regarding the first evaluation index, in the sleep stages based on PSG, other sleep stages may appear during REM sleep, but if the duration of sleep stages other than REM sleep is within 5 minutes, one REM Count as sleep.
Regarding the second evaluation index, since REM sleep is a sleep stage that continues for a certain period of time, REM sleep of less than 5 minutes is not counted among the estimated REM sleep.

FIG. 17 is a diagram showing the results of REM sleep determination using the REM sleep determination method according to the present example. A method for deriving the first evaluation index and the second evaluation index will be described with reference to FIG. 17. FIG. 17 shows sleep stages based on body movement density, moving average, and PSG. The vertical axis on the left axis indicates the number of body movements (body movement density). The vertical axis on the right axis indicates sleep stages. The horizontal axis indicates elapsed time. Ranges T201 to T204 each indicate actual REM sleep.

According to the sleep stages based on PSG, the data shown in FIG. 17 shows that four times of REM sleep appear in one night. Therefore, the actual number of REM sleeps is four. The REM sleep estimated by the REM sleep determination method according to the present embodiment temporally overlaps with these four actual REM sleeps three times. Therefore, the number of times of REM sleep that was successfully estimated is three. Accordingly, the first evaluation index is derived as 3/4=75%. In addition, the time that is not actually REM sleep is estimated as REM sleep, and there is one range of REM sleep that continues for 5 minutes or more (from falling asleep to around 2:30). Therefore, the second evaluation index is one time.

Further, the hyperparameter settings used in this experiment were x = 1 [%], y = 60 [seconds], and z = 90 [minutes]. x is the percentage of the magnitude of the body movement when the body movement during wakefulness is removed. y is the interval (in seconds) before and after removing the top x% of body movements at the same time. z is the length of the time window (unit: minutes) when calculating the moving average. It is known that awakening occurs in 5 to 15% of total sleep. In order to remove body movements corresponding to arousal from the ratio of arousal to the total sleep, x=1 [%] and y=60 [seconds] are set. Furthermore, since the sleep cycle is approximately 90 minutes, z is set to 90 [minutes] in order to compare the level of body movement density within that cycle.

Results will be shown regarding two evaluation indicators, a first evaluation indicator and a second evaluation indicator. FIG. 18 is a diagram showing the results of each subject for the first evaluation index. FIG. 19 is a diagram showing the results of the second evaluation index for each subject. Moreover, FIG. 20 is a bar graph showing the average value and standard deviation of the results of the first evaluation index. FIG. 21 is a bar graph showing the average value and standard deviation of the results of the second evaluation index.

The average percentage of the number of times actual REM sleep could be estimated was 97±8% for the REM sleep determination method according to the present embodiment, 80±23% for RSSE, 84±23% for MTRE, and the average number of times for FP was , the REM sleep determination method according to the present embodiment was 1.9±1.5 times, RSSE was 2.6±1.4 times, and RF was 4.3±2.0 times. From this, compared to conventional methods (RSSE, MTRE), the REM sleep determination method according to the present embodiment has the following advantages: , it can be seen that the number of times that the time that is actually not REM sleep is incorrectly determined as REM sleep is small.

FIG. 22 is a diagram illustrating an example of REM sleep determination results using the REM sleep determination method according to the present embodiment and conventional methods (RSSE, MTRE). FIG. 22(A), FIG. 22(B), and FIG. 22(c) respectively show the estimation results of the REM sleep determination method, RSSE, and MTRE according to this embodiment. The vertical axis shows sleep stages, and the horizontal axis shows elapsed time. Further, in FIG. 22, sleep stages based on PSG are shown along with REM sleep estimated by each method.

The REM sleep estimated by the REM sleep determination method according to the present embodiment periodically captures actual REM sleep. On the other hand, REM sleep estimated by conventional methods (RSSE, MTRE) may not capture REM sleep, as shown in range T301, range T302, range T303, T304, and range T305. In range T303, REM cannot be determined continuously.

Furthermore, as shown in range T401, range T402, range T403, and range T404, erroneous determinations occur in all the methods. However, conventional methods (RSSE, MTRE) incorrectly determine short-term REM sleep at short intervals, and do not capture the characteristic of REM sleep that it occurs periodically at approximately 90-minute intervals. This is thought to be because conventional methods (RSSE, MTRE) do not take into account the periodicity of sleep, and as long as the judgment conditions are met, even data without sleep periodicity is judged to be REM sleep. .

On the other hand, the REM sleep estimated by the REM sleep determination method according to the present embodiment is obtained in a relatively long range that follows the range of actual REM sleep, and the range determined as REM sleep is approximately 90%. There is a cyclical pattern of sleep that seems to rise and fall at minute intervals.

As explained above, the REM sleep determination device 1 according to the present embodiment includes the body movement data acquisition unit 20, the body movement removal unit 22, the body movement density moving average calculation unit 23, and the REM sleep determination unit 24. , is provided.
The body movement data acquisition unit 20 acquires body movement data A1 that indicates the magnitude of body movement during the sleeping period for each time.
The body movement removing unit 22 extracts first body movement data (in this embodiment, non-REM sleep stage 2 body movement data and waking body movement data) which is data corresponding to body movements in a predetermined sleep stage from the body movement data A1. dynamic data).
The body movement density moving average calculation unit 23 calculates the data for a predetermined length of a first period of time (30 minutes in this embodiment) from the body movement data A1 from which the first body movement data has been removed by the body movement removing unit 22. A body movement density indicating the number of body movements that occurred in the second period of time and a moving average per a second period of predetermined length (in this embodiment, 90 minutes) are calculated.
The REM sleep determination unit 24 determines a period of sleep in which the body movement density calculated by the body movement density moving average calculation unit 23 exceeds the moving average as a period of REM sleep.

With this configuration, the REM sleep determination device 1 according to the present embodiment can remove body movements that are less relevant to REM sleep from the body movement data A1 and determine REM sleep from the characteristics of the remaining body movements. , it is possible to improve the accuracy of REM sleep determination in an unrestricted REM sleep determination method.

In the REM sleep determination device 1, when determining REM sleep from the characteristics of body movements that remain after removing body movements that are less relevant to REM sleep, it is possible to determine REM sleep based on the characteristics of body movements that remain after body movements that are less relevant to REM sleep are removed. REM sleep is determined based on the characteristic that large body movements occur multiple times within a time window of several minutes.
Furthermore, in the REM sleep determination device 1, the period in which the body movement density exceeds the moving average is determined as the period of REM sleep. , it is possible to improve the accuracy of determining REM sleep.

Further, in the REM sleep determination device 1 according to the present embodiment, the first body movement data includes body movement data during awakening, which is data corresponding to body movements during awakening.

With this configuration, the REM sleep determination device 1 according to the present embodiment can remove the waking body movement data from the body movement data A1 and determine REM sleep from the characteristics of the remaining body movements. In the method for determining REM sleep, the accuracy of determining REM sleep can be improved compared to the case where waking body movement data is not removed in advance.

Further, in the REM sleep determination device 1 according to the present embodiment, the body movement removing unit 22 removes the waking body movement data from the body movement data A1 based on the magnitude of the body movement indicated by the body movement data A1. .

With this configuration, the REM sleep determination device 1 according to the present embodiment can remove body movement data during awakening from body movement data A1 based on the feature that body movement during awakening is large compared to REM sleep. The accuracy of determining REM sleep can be improved compared to the conventional technology (for example, MTRE described in Non-Patent Document 2), which determines the range of wakefulness based on machine learning. When determining sleep stages other than REM sleep based on machine learning as in the conventional technology, erroneous determinations occur frequently and the determination results vary.

Further, in the REM sleep determination device 1 according to the present embodiment, the body movement removing unit 22 determines that the magnitude of body movement indicated by the body movement data A1 is determined to be a predetermined upper predetermined ratio (in the present embodiment). , 1%), the waking body movement data is removed. The ratio is determined from the ratio of wakefulness to total sleep.

With this configuration, the REM sleep determination device 1 according to the present embodiment can remove the waking body movement data from the body movement data A1 based on the ratio determined from the ratio of wakefulness to the whole sleep. The accuracy of determining REM sleep can be improved compared to the case where the ratio is not based on a predetermined ratio of wakefulness.

In addition, in the REM sleep determination device 1 according to the present embodiment, the first body movement data includes NREM sleep stage 2 body movement data (in this embodiment, NREM sleep stage 2 body movement data that corresponds to body movements in NREM sleep stage 2). Data corresponding to body movements from stage 2 to NREM sleep stage 4) are included. The body movement removal unit 22 removes the non-REM sleep stage 2 body movement data from the body movement data A1 acquired by the body movement data acquisition unit 20 based on the result of machine learning.

With this configuration, the REM sleep determination device 1 according to the present embodiment can remove the NREM sleep stage 2 body movement data from the body movement data A1 based on the result of machine learning. Therefore, the accuracy of determining REM sleep can be improved compared to the case where the results of machine learning are not used. As described above, since REM sleep and NREM sleep stage 2 are similar in the magnitude of body movement, it is difficult to distinguish between REM sleep and NREM sleep stage 2 based on the magnitude of body movement.

Further, in the REM sleep determination device 1 according to the present embodiment, the body movement density moving average calculation unit 23 calculates a time window of a predetermined length determined based on the sleep cycle (90 minutes in the present embodiment). Calculate the moving average using

With this configuration, the REM sleep determination device 1 according to the present embodiment can compare the height of the body movement density within the sleep cycle, so it is easier to determine REM sleep than when the time window is not determined based on the sleep cycle. Accuracy can be increased.

Note that some parts of the REM sleep determination device 1 in the embodiment described above, such as the body movement data acquisition unit 20, the non-REM sleep body movement determination unit 21, the body movement removal unit 22, the body movement density moving average calculation unit 23, and the REM sleep The determination unit 24 and the output unit 25 may be realized by a computer. In that case, a program for realizing this control function may be recorded on a computer-readable recording medium, and the program recorded on the recording medium may be read into a computer system and executed. Note that the "computer system" herein refers to a computer system built into the REM sleep determination device 1, and includes hardware such as an operating system (OS) and peripheral devices. Furthermore, "computer-readable recording media" refers to portable media such as flexible disks, magneto-optical disks, ROMs (Read Only Memory), and CD-ROMs (Compact Disc-Read Only Memory), as well as portable media built into computer systems. Refers to storage devices such as hard disks. Furthermore, a "computer-readable recording medium" refers to a medium that dynamically stores a program for a short period of time, such as a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. In this case, it may also include a device that retains a program for a certain period of time, such as a volatile memory inside a computer system that is a server or a client. Further, the above-mentioned program may be one for realizing a part of the above-mentioned functions, or may be one that can realize the above-mentioned functions in combination with a program already recorded in the computer system.
Moreover, a part or all of the REM sleep determination device 1 in the embodiment described above may be realized as an integrated circuit such as an LSI (Large Scale Integration). Each functional block of the REM sleep determination device 1 may be made into a processor individually, or some or all of them may be integrated into a processor. Further, the method of circuit integration is not limited to LSI, but may be implemented using a dedicated circuit or a general-purpose processor. Further, if an integrated circuit technology that replaces LSI emerges due to advances in semiconductor technology, an integrated circuit based on this technology may be used.

Although one embodiment of the present invention has been described above in detail with reference to the drawings, the specific configuration is not limited to that described above, and various design changes etc. may be made without departing from the gist of the present invention. It is possible to

DESCRIPTION OF SYMBOLS 1... REM sleep determination device, 20... Body movement data acquisition unit, 22... Body movement removal unit, 23... Body movement density moving average calculation unit, 24... REM sleep determination unit, A1... Body movement data

Claims (14)

a body movement data acquisition unit that acquires body movement data indicating the magnitude of body movement during a sleeping period at each time;
a body movement removing unit that removes first body movement data that is data corresponding to body movements in a predetermined sleep stage from the body movement data;
a body movement density indicating the number of body movements that have occurred during a first time period of a predetermined length from the body movement data from which the first body movement data has been removed by the body movement removal unit; a body movement density moving average calculation unit that calculates a moving average per second hour;
A REM sleep determination unit that determines a period of the sleep period in which the body movement density calculated by the body movement density moving average calculation unit exceeds the moving average as a period of REM sleep;
A REM sleep determination device. The REM sleep determination device according to claim 1, wherein the first body movement data includes body movement data during wakefulness, which is data corresponding to body movement during wakefulness. The REM sleep determining device according to claim 2, wherein the body movement removing unit removes the waking body movement data from the body movement data based on the magnitude of the body movement indicated by the body movement data. The body movement removal unit removes the waking body movement data from the body movement data based on data in which the magnitude of the body movement indicated by the body movement data is a predetermined upper ratio;
The REM sleep determining device according to claim 2, wherein the ratio is determined from the ratio of wakefulness to the total sleep. The REM sleep determining device according to claim 2, wherein the body movement removing unit removes the waking body movement data from the body movement data based on a result of machine learning. The first body movement data includes NREM sleep stage 2 body movement data, which is data corresponding to body movements in NREM sleep stage 2, along with the waking body movement data,
The body movement removing unit removes the NREM sleep stage 2 body movement data from the body movement data from which the waking body movement data has been removed. REM sleep determination device. The REM sleep determination device according to claim 1, wherein the first body movement data includes NREM sleep stage 2 body movement data that is data corresponding to body movements in NREM sleep stage 2. The REM sleep determination device according to claim 7, wherein the body movement removing unit removes the NREM sleep stage 2 body movement data from the body movement data based on the magnitude of the body movement indicated by the body movement data. . The body movement removal unit removes the non-REM sleep stage 2 body movement data based on data in which the magnitude of the body movement indicated by the body movement data is a predetermined upper predetermined ratio from among the body movement data,
The REM sleep determination device according to claim 7, wherein the ratio is determined from the ratio of non-REM sleep stage 2 to the entire sleep. The REM sleep determination device according to claim 7, wherein the body movement removing unit removes the NREM sleep stage 2 body movement data from the body movement data based on a result of machine learning. The first body movement data includes body movement data during awakening, which is data corresponding to body movements during awakening, along with the non-REM sleep stage 2 body movement data,
The body movement removing unit removes the waking body movement data from the body movement data from which the non-REM sleep stage 2 body movement data has been removed. REM sleep determination device. The REM sleep determination device according to claim 1, wherein the body movement density moving average calculation unit calculates the moving average using a time window of a predetermined length determined based on a sleep cycle. a body movement data acquisition step of acquiring body movement data indicating the magnitude of body movement during the sleeping period at each time;
a body movement removing step of removing first body movement data that is data corresponding to body movements in a predetermined sleep stage from the body movement data;
a body movement density indicating the number of body movements that have appeared during a first time period of a predetermined length from the body movement data from which the first body movement data has been removed in the body movement removal step; a body movement density moving average calculation step of calculating a moving average per second hour of
REM sleep determining step of determining a period of the sleep period in which the body movement density calculated by the body movement density moving average calculating step exceeds the moving average as a period of REM sleep;
A method for determining REM sleep. a body movement data acquisition step of acquiring body movement data indicating the magnitude of body movement during the sleeping period at each time;
a body movement removing step of removing first body movement data that is data corresponding to body movements in a predetermined sleep stage from the body movement data;
a body movement density indicating the number of body movements that have appeared during a first time period of a predetermined length from the body movement data from which the first body movement data has been removed in the body movement removal step; a body movement density moving average calculation step of calculating a moving average per second hour of
REM sleep determining step of determining a period of the sleep period in which the body movement density calculated by the body movement density moving average calculating step exceeds the moving average as a period of REM sleep;
A program to run.

Images